P

R
O
C
E
E
D
I
N
G
S
MpCCI
2
4
W
E
L
C
O
M
E
8
TH
MpCCI USER FORUM 2007 AT SCHLOSS BIRLINGHOVEN
COUPLED SIMULATIONS
A new market segment for Computer Aided Engineering
MpCCI (Mesh-based parallel Code Coupling Interface) has opened a new market segment for CAE:
multi disciplinary simulation through coupling of established and best in-class simulation tools offers a variety
of new disciplines and solution strategies.
MpCCI has been developed at the Fraunhofer Institute SCAI in order to provide a code-independent interface
for the coupling of different simula tion codes. Most of the leading CAE software vendors support the MpCCI
Interface.
About 100 licensed users and 40 partner installations worldwide (in 2006) have demonstrated usability and
benet of the MpCCI open concept.
Supported Codes (MpCCI 3.0.6):

ABAQUS

ANSYS

Flowmaster (in preparation)

FLUENT

FLUX

Icepak

MSC.Marc

PERMAS

STAR-CD

RadTherm

Programmable Interface for inhouse Codes

On behalf of the MpCCI Team
Klaus Wolf
Coupled Application Areas

Fluid Structure Interaction

Thermal Coupling and Radiation

Thermo-Electrical Interaction

Magneto-Hydro-Dynamics

Coupling of 3D-CFD and 1D-Pipeline

Fluid-Fluid Combinations

and many more

Klaus Wolf
Fraunhofer Institute for Algorithms
and Scientic Computing SCAI
Schloss Birlinghoven
53754 Sankt Augustin
Germany
Phone: +49 (0) 2241 / 14 - 2557
klaus.wolf@scai.fraunhofer.de
www.scai.fraunhofer.de
www.mpcci.de
C
O
N
T
A
C
T
5
6
MpCCI STATUS AND BUSINESS
News around MpCCI Development and Projects
Klaus Wolf Fraunhofer Institute SCAI
10-17
FLUID-STRUCTURE INTERACTIONS
A new CFD and thermal analysis method based on MpCCI
Ernst-Peter Weidmann FKFS, Walter Bauer, Martin Maihfer
DaimlerChrysler and Uwe Scholl Fraunhofer SCAI
18-27
Coupling FLUENT with FEM codes
Mark Pelzer FLUENT Germany
28-35
Solving uid-structure interaction problems using MpCCI and
the Component Template Library (CTL)
J. Rang, M. Krosche, R. Niekamp, H. G. Matthies Technical University
of Braunschweig
36-41
OPEN INTERFACE FOR COUPLING OF INHOUSE CODES
How to create your own code adapter
Holger Spiess and Pascal Bayrasy Fraunhofer SCAI
42-45
Coupled simulations using the elasti city solver DDFEM and
FLUENT via MpCCI
Dimitar Stoyanov Fraunhofer ITWM
46-56
Fluid-structure simulations using MpCCI 3.0.5 with ABAQUS and an
in-house CFD-Code on distributed resources
Felix Lippold, University of Stuttgart IHS
56-61
C
O
N
T
E
N
T
7
ELECTRICAL ENGINEERING
Simulation of the switching arc for a motor protective
circuit breaker
Christian Rmpler Fraunhofer SCAI and Albert Zacharias Moeller GmbH
62-69
MpCCI-coupled simulations for electrical-distribution equipment
Ian Lyttle, Benjamin Pulido Schneider Electric
70-75
SOLUTIONS FROM OUR SOFTWARE PARTNERS
Pressure moulding simulation using FLUENT and ABAQUS
Jasper Kidger FLUENT UK
76-83
Fluid-structure solutions based on ABAQUS
Albert Kurkchubatsche ABAQUS US
84-95
Coupling STAR-CD with ABAQUS to model FSI interaction of
a body entering a uid
Philip Morris Jones CD-adapco
96-101
CIVIL ENGINEERING
R&D of nuclear power plant simulator on atomic energy grid
infrastructure
Noriyuki Kushida, Yoshio Suzuki, Osamu Hazama, Hitoshi Matsubara,
Akemi Nishida, Fumimasa Araya, Tetsuo Aoyagi and Norihiro Kakajima
Japan Atomic Energy Agency JAEA
102-107
8
3-code coupling in ood simulation
Jrg-Volker Peetz Fraunhofer SCAI and Thomas Sommer Dresden
Ground water Research Center
108-117
BIOMEDICAL
Preconditions for a uid-structure interaction of the human heart
using FLUENT, ABAQUS and MpCCI
S.B.S. Krittian, H. Schmid, T. Schenkel, H. Oertel University of Karlsruhe
118-125
Code adapter for arbitrary prescribed motion and deformation
using FLUENT and MpCCI
M.G. Perschall, H. Spiess, T. Schenkel, H. Oertel University of Karlsruhe
126-131
C
O
N
T
E
N
T
9
PARTICIPANTS OF THE 8
th
MPCCI USERFORUM 2007
10
NEWS AROUND MPCCI DEVELOPMENT AND PROJECTS
Klaus Wolf Fraunhofer Institute SCAI
Solving multi-disciplinary problems by using coupled simulation environments is becoming a
state-of-the-art technology for a fast growing engineering community. The MpCCI interface
has been accepted as a de-facto standard for such code coupling environments.
MpCCI 3.0.6
In spring 2007 the new version of MpCCI 3.0.6 will be released. Besides various
improvements for the system internal processing there are some major enhancements for the
user interface and code support:
In MpCCI 3.0.6 for most codes the self-code-coupling is available. Instead of coupling
two different simulations tools a user may decide to set up an interaction between two
instances of the same code.
In its public production version MpCCI already supports nine codes (ABAQUS, ANSYS,
FLUENT, Flux3D, ICEPAK, MSC.Marc, PERMAS, STAR-CD 3 & 4, RadTherm). The new
Code-Adapter-API in 3.0.6 now opens the door for further code-writers to create MpCCI
interfaces for their own codes which are fully compatible with the codes listed above.
In most multi-disciplinary cases the user will take care that the coupled surfaces match
to each other as far as possible. For large applications with hundreds of (independent)
parts and a total size of coupling surface with 10
5
to 10
6
nodes/cells such a 100 percent
geometric compatibility cannot be guaranteed due. MpCCI 3.0.6 now provides some
extrapolation schemes which provide coupled values even for such regions which otherwise
would be orphaned.
To support deforming FSI-structures when coupling STAR-CD with FEM codes MpCCI
provides a separate morphing tool. This spring-based morpher has been extended to run
also polyeder based models as used by STAR-CD 4.
For MpCCI 3.0.6 we put a lot of effort in a much more detailed documentation and
extended tutorials. Various examples for different code combinations and comparisons of
coupled simulation results should allow MpCCI beginners to run rst simple application
cases in a do-it-yourself mode.
M
p
C
C
I

S
T
A
T
U
S

A
N
D

B
U
S
I
N
E
S
S
Coupling two instances of the same Code
Klaus Wolf
Fraunhofer Institute for Algorithms
and Scientic Computing SCAI
Schloss Birlinghoven
53754 Sankt Augustin
Germany
Phone: +49 (0) 2241 / 14 - 2557
klaus.wolf@scai.fraunhofer.de
www.scai.fraunhofer.de
www.mpcci.de
Coupling two instances of the same Code
Although most applications use MpCCI to couple two different codes there are also some
problems where interaction between two instances of the same code is useful. Compared
to a single complex model a coupled environment provides more exibility for the model
set-up. The coupling of two independent sub-models allows to specify appropriate runtime
parameters (e.g. relaxation factors or discretization schemes) for each sub-model with
different values. In MpCCI 3.0.6 such self-code couplings can easily be specied by
selecting two models of same code type in the Model Panel of the MpCCI GUI. All
following steps (coupling specs, parameters and startup) are handled in same way as for
coupling two different codes.
C
O
N
T
A
C
T
An example of a coupled CFD-CFD application would be a Glass Tank simulation using a
FLUENT-FLUENT environment. The rst sub-domain is computing the combustion chamber
the second CFD domain is for the glass space itself. This is a thermal coupling only.
Each model has its own radiation model no radiation properties will be transferred. The
temperature prole is send from the glass tank domain to the combustion area; heat ux is
returned from combustion to glass area. First test cases with a simple 2D case show similar
results in the single FLUENT solution and the coupled FLUENT-FLUENT model (example
provided from Fluent Germany/ Fluent Asia Pacic).
MpCCI
MpCCI
GUI
Server
data
run
code
adapter
GUI
integration
control
Simulation
code
12
M
p
C
C
I

S
T
A
T
U
S

A
N
D

B
U
S
I
N
E
S
S
Programming Interface to create Adapters for in-house Codes
The basic idea of MpCCI is to enable the coupling between any two simulation codes
independent if they are commercial, in-house or research tools.
MpCCI provides two levels of programming interfaces (APIs) to link the coupling software
MpCCI with a code.
MpCCI SDK is a lower level subroutine interface and
MpCCI Adapter level used since MpCCI 3.0 for all commercial codes.
The MpCCI SDK interface level provides various basic functions to dene coupling regions,
control communication between the coupled codes and to handle other MpCCI parameters.
The usage of MpCCI SDK is very similar to the usage of the MPI Message Passing Interface
- there is no xed protocol when to communicate which data with which other coupled
components. This MPI-like exibility allows the implementation of different coupling
algorithms. However, this exibility has one big disadvantage MpCCI itself cannot guarantee
that the sequence how MpCCI procedure calls (dene meshed, send and receive operations)
used in code A matches to a sequence of MpCCI calls realised in code B.
To avoid such protocol incompatibilities between different code integrations MpCCI 3.0 has
introduced the concept of MpCCI code adapters. These adapters have internal mechanisms
to control the current coupling state and actions of each integrated code. These Coupling
Control Managers guarantee a consistent behaviour and communication protocol for all
adapted codes.
MpCCI 3.0.6 now provides an open programming interface (API) to implement code adapters
for own in-house codes (details on how to use this new API can be found in the article from
Holger Spiess). First solutions based on this open API already have been realised by Fraunhofer
ITWM (in-house FEM code), University of Stuttgart (own CFD code), University of Karlsruhe
(image processing code) and an industrial customer using a large research code for CFD
which will be coupled to the ABAQUS code.
The current MpCCI 3.0 still allows
both levels of integration (SDK
and Adapter level). However, it
should be noted that for reasons
of compatibility the MpCCI
Adapter Level should be preferred
for all new activities. In midterm
the MpCCI SDK interface will be
regarded as a pure internal MpCCI
and thus might be subject to larger
changes.
Orphaned Regions in FEM without coupled
values from CFD
Orphaned Regions in FEM with
extrapolated CFD-values
13
Extrapolation
The MpCCI interpolation algorithms are based on the assumption that both coupled surfaces
are well dened and geometrically compatible to each other. The CFD and FEM domain can
be modelled individually based on accuracy and runtime demands. MpCCI then is responsible
to map the discretisations used in CFD and FEM to each other. However, in some cases it
might happen that parts which are modelled in detail in the FEM domain are not represented
in the coupled CFD domain. In the gure above the pipe is only part of the FEM model and
not of the CFD model.
To handle such model mismatches a method is needed for assigning data and thus emulating
a coupling for this pipe region. One way is to assign default values. This leads to a number
of problems from listing all regions up to guessing values for each region. Therefore MpCCI
was extended to deal automatically with non matched regions by extrapolating values from
matched regions into the non matched ones. The extrapolation is based on a weighting
function over the distances to the borders of the non matched region. With d being the
distance between a non matched cell x and a border cell b, f(b
n
) is a value assigned to b, w is
a weighting function such that,
1=w(d(x,b
n
)) (1)

n
the value of x is then extrapolated by:
f(x)=w(d(x,b
n
))f(b
n
) (2)

n
Test case for morpher performance
14
M
p
C
C
I

S
T
A
T
U
S

A
N
D

B
U
S
I
N
E
S
S
MpCCI Grid Morpher
The MpCCI grid morpher will smooth a grid, which contains displaced nodes / vertices,
typically encountered in uid-structure interactions. The MpCCI grid morpher is based on a
spring-based morphing technique applying a modied radial-based function theory.
1. The MpCCI mesh morpher is started as a separate process and connected to the parallel
STAR-CD program via sockets. At start-up the morpher receives the initial mesh denition
of the STAR-CD.
2. During coupled simulation the FEM codes sends deformation information to the code
adapter of STAR-CD.
3. These structural changes of the uid boundary are then transferred to the grid morpher.
The morpher calculates the new positions of boundary and inner vertices and does an
additional smoothing step on the CFD mesh.
4. The morphed mesh model is returned back to the STAR-CD code adapters; the newxyz-
function is used to pass the adapted model to the STAR-CD solver.
Some basic benchmarks for the morpher may demonstrate the performance of this MpCCI
add-on tool. The tests were done on a 3 GHertz Xeon machine. The model was a rectangular
channel ow (STAR-CD 3.26) with a simple ap placed in the middle of the channel (ABAQUS
6.7). The MpCCI morpher shows a very good speed-up related to the number of nodes in
volume:
~ 1,000 nodes on surface / 25,000 Elements in volume 0,15 seconds
~ 4,000 nodes on surface / 200,000 Elements in volume 0,8 seconds
~15,000 nodes on surface / 1,500,000 Elements in volume 6,0 seconds
15
THE OPEN MpCCI COMMUNITY
The vision and goal for MpCCI is to be the leading and most used open interface for
simulation code coupling. MpCCI was and is open for any interested code writer (from
research, industry, and commercial software vendors) who wants his code to interact with
other simulation software.
The open API in MpCCI 3.0.6 now provides a straight forward strategy to combine own
codes with other MpCCI enabled codes in one multi-disciplinary environment. All technical
information and software interfaces to adapt further simulation tools to the MpCCI concept
are part of the public MpCCI product version.
Fraunhofer SCAI will use its position as an independent organisation for applied research
to push this vision of an Open MpCCI Community by motivating and supporting any
organisation to adapt its simulation solution to the MpCCI interface concept.
Trilateral FSI Co-operations
In 2004 ABAQUS Inc. and FLUENT Inc. (now part of Ansys Inc.) announced their strategic
partnership to solve uid-structure problems with MpCCI based environments. In joint
development and marketing efforts all three partners together could demonstrate the
usability and quality of FSI environments. Since the rst product version (MpCCI 3.0.4,
ABAQUS 6.5 and Fluent 6.2.18) was launched in summer 2004 around 60 users in industry
and research switched from a previous mono-disciplinary working environment to a now
multi-disciplinary coupled solution.
At the 8
th
MpCCI User Forum 2007 Dassault Systmes and CD-adapco announced MpCCI
as solution for analysing uid-structure interaction with STAR-CD and ABAQUS. Dassault
Systmes and CD-adapco, world leading providers of simulation solutions, follow the strategy
of delivering an open platform for multi-physics simulation. The bi-directional coupling
leverages a strong partnership of both companies with each other and the independent,
open solution for the coupling MpCCI software to enable ABAQUS and STAR-CD to work
together to solve a wide range of important FSI problems.
Both code combinations ABAQUS / MpCCI / FLUENT as well as ABAQUS / MpCCI / STAR-CD
are fully supported by all the related partners. End-users who decide to start and use one of
these combinations will receive best FSI support either from the CFD, or from the FEM or from
the SCA as interface provider. Both tri-lateral co-operations are settled as strategic solutions
for a long-term usage by the engineering community.
16
M
p
C
C
I

S
T
A
T
U
S

A
N
D

B
U
S
I
N
E
S
S
Bilateral Co-operations
CEDRAT SA in France for the integration of the Flux3D EMAG code. Our customer Schneider
Electric US is using a code combination ICEPAK Flux3D to compute thermal behaviour
virtual switching devices and busbar systems (see article from Ian Lyttle, Schneider US in this
proceedings);
Flowmaster Ltd. UK provides a system simulation code (Flowmaster) for 1D pipeline
computations. Users in automotive and aerospace industry want to set-up 1D-3D (FLUENT or
STAR-CD) environments to compute air condition or cooling networks in combination with
detailed 3D parts. This 1D-3D environment has reached a prototype stage rst steady state
cases tests show promising results. Public product versions are planned for summer 2007.
Intes GmbH provides their Permas code (FEM) which is combined with STAR-CD and POSRAD
to form a 3-code solution for the thermal management of automotive underhood (see
contribution from Ernst-Peter Weidmann, DaimlerChrysler);
MSC Software Corporation has enabled the MSC.Marc code for FSI applications with STAR-
CD or FLUENT. First users from electronics industry in Japan recently have started to utilize this
FSI solution for thermal problems.
The radiation code RadTherm from ThermoAnalytics Inc. is used in combination with CFD
codes to simulate the thermal/radiation effects in automotive applications.
RESEARCH PROJECTS
In its role as an organisation for applied research Fraunhofer SCAI participates in various
national or European funded projects. In some of them MpCCI is used as a technical
integration platform in others new technical approaches are investigated for a potential
further usage inside MpCCI.
3ZM-GRIMEX - Ground water simulation: The ood 2002 at the Elbe River showed that
besides the apparent damages by surface waters, considerable damage was caused by
groundwater and waters from the sewer system. Fast rising groundwater levels due to
inltration of overland ow or water from the sewer system can damage basements and
can be a serious thread to the static of houses. The objective of the project 3ZM-GRIMEX
is the development of a software system for the coupled simulation of three components:
surface water, groundwater, and sewer water. The coupled system will be used in ood risk
management. (More details are given in the article from Jrg-Volker Peetz, SCAI)
17
INGRID Grid Based environments: There is an obvious trend to use distributed grid-
environments for large and complex applications. Compute resources, data bases and services
from different sites may be combined to form a (temporary) solution for a specic problem
case. Coupled FSI simulations could benet from this resource sharing strategy in grids.
The coupled components CFD, FEM and MpCCI could be offered as grid services
(following the upcoming SOA standards); a user might then select available services and
combine them in one application. A rst prototype of such a grid based coupled simulation is
presented in the article from Felix Lippold, University of Stuttgart.
KOPFEM Code API and Extrapolation Schemes: Nowadays, despite the variety of commercial
solvers available on the market, the question of the design and development of effective
numerical solvers is still open. It is a fact that many problems can not nd their proper
treatment using the existing (commercial) codes. Typical examples are large sized problems
with complicated geometry, e.g. problems in micromechanics, see [1], which necessarily
require a parallel execution. In this sense, a specic-purpose in-house code, specially designed
and programmed for a particular parallel architecture, can often provide much better
performance than the existing general-purpose tools (see article from Dimitar Stoyanov,
Fraunhofer ITWM).
IMAUF Advanced Stamping Simulation: Simulation of manufacturing processes like metal
stamping is a well established procedure in all of the automotive companies today. Dedicated
stamping codes like Autoform, Indeed, Pam-Stamp and other more general codes like
LS-Dyna or Abaqus Explicit can be used for these purposes. However, in all applications
of stamping simulation today there is one major assumption about the behaviour of the
stamping tool itself: the tool is expected to be stiff enough and no deformations due to
friction of the stamped plate are considered during simulation. In a recently started project
the coupling of a stamping simulation with the structural analysis of the deforming stamping
tool will be realised. The deformation of the stamping tool will lead to locally different friction
parameters compared to a standalone stamping simulation. It is one of the challenges in
this research project to evaluate inuence of changing friction parameters onto the quality of
the nal stamped plate (see also http://www.imauf.de).
CONCLUSION
MpCCI 3.0 provides a lot of new features for the coupling of simulation codes. Together with
MpCCI code adapters a complete toolbox for multidisciplinary simulation is ready for use with
standard commercial simulation codes. Various solutions demonstrate the applicability of this
concept and the valuable outcome for the end users. A growing number of commercial codes
as well as inhouse and research codes are supported by MpCCI code adapters.
18
A NEW CFD AND THERMAL ANALYSIS METHOD
BASED ON MpCCI
Ernst-Peter Weidmann FKFS; Walter Bauer, Martin Maihfer DaimlerChrysler;
Uwe Scholl Fraunhofer Institute SCAI
ABSTRACT
Vehicle thermal protection is of vital importance for the development process of passenger
cars. Tightly-packaged engine compartments and strongly increased engine power
demand extensive testing and more and more numerical analysis. Underhood component
temperatures are sensitive to all three modes of heat transfer, conduction, convection and
radiation. Due to very high temperatures next to the exhaust system, in particular, a thermal
radiation analysis is required. The different numerical properties of the three physical modes
can be handled separately by sophisticated but specialized simulation codes. This paper
proposes a partitioned approach for the multi physics 3-D simulation methodology based
on coupling the commercial CFD code STAR-CD, the radiation code POSRAD and the nite
element code PERMAS by the Coupling Software MpCCI.
The applicability of this methodology is demonstrated by a simplied exhaust system. The
surface temperature is computed considering both internal and external ow. Air and
exhaust gas temperatures are coupled with the thermal analysis.
In the near future a similar methodology will be used to perform thermal analyses of the
engine compartment and the entire car. The new trilateral coupling approach is supposed to
be more accurate, especially in regions with a very strong coupling between temperature and
velocity eld.
INTRODUCTION
Thermal protection is increasingly important in the development process of passenger cars.
Tightly-packaged engine compartments and strongly increased engine power demand
extensive testing and analysis. Traditionally thermal protection is tested in climate wind
tunnels and road tests. Reduced time to market and high costs for prototypes result in the
need for building up the rst prototype very close to its optimum design. The introduction
of numerical methods allows for the optimization of cooling requirements as well as thermal
analysis of temperature sensitive components in an early stage of the vehicle development
process. Although road tests will also be necessary in the future to verify the numerical
predictions, the use of numerical methods helps to focus such tests on the most critical
points.
Fluid-structure interaction covers a broad scope of problems. The interaction can be
mechanical, thermal or even both. Underhood component temperatures are sensitive to
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
19
all three modes of heat transfer: conduction, convection and radiation. Due to very high
temperatures next to the exhaust manifold, a thermal radiation analysis is required.
There are basically two ways of approaching this problem. One way is to model the problem
as one closed monolithic system. The other solution is based on a so called partitioned
approach where each physical domain is modeled separately. Solving the system is to solve
each domain and to exchange boundary values between them. In contrast to the monolithic
approach, existing specialized simulation tools can be assigned to the different domains and
modelling can be done individually with respect to accuracy and runtime demands. One
drawback is the need of an additional tool for handling the exchange of boundary values
between the simulation codes. For industrial application it is important to ensure that the
tool is easy to use such that the overhead for setting up the coupling is small with respect
to the effort for modelling and solving the different domains. It has been known for some
time that there is no numerical analysis code available able to deal with the thermal problems
in the complex geometry of a whole vehicle by a monolithic approach. Several examples
of partitioned multi-mode heat transfer computations can be found in recently published
literature (1, 2, 3).
Reister and Maihfer (4) and Bauer and Maihfer (5) described a procedure to import heat
transfer data from a CFD computation into a thermal analysis code. In the case of forced
convective heat transfer the coupling between component temperature and ow eld
is relatively weak, although the opposite is not true. It is therefore sufcient to provide
convective heat uxes in a preliminary CFD analysis. On the other hand the strong coupling of
the velocity and temperature eld in buoyancy driven ows requires a two-way coupled multi-
mode heat transfer analysis (6, 7).
The methodology presented in this paper is an extension to the one presented by Maihfer
and Bauer (5) by considering the feedback between the structure and the CFD in both
directions and in an iterated way. The method by Maihfer and Bauer is based on the
computational uid dynamics code STAR-CD (8), the nite element code PERMAS (9) and
the thermal radiation solver POSRAD (8). It is integrated in the development process of all
new Mercedes-Benz passenger cars. To enable a coupling between the codes a software
tool was needed. There are some vendor specic approaches to couple codes from different
application elds via proprietary mechanisms (10). There are also some area specic coupling
Dipl.-Ing. Ernst-Peter Weidmann
Forschungsinstitut fr Kraftfahrwesen
und Fahrzeugmotoren Stuttgart FKFS
Pfaffenwaldring 12
70569 Stuttgart
Germany
Phone: +49 (0) 7031 / 90 45727
ernst-peter.weidmann@fkfs.de
www.fkfs.de
C
O
N
T
A
C
T
solutions available; but their use is restricted to special application elds like aeroelastic
coupling in airplane design or the coupling of ocean and atmosphere models in climate
research (11). This situation motivated the development of a software environment which
provides a general solution for multidisciplinary simulations.
The radiative heat ux is calculated in a third program named POSRAD. POSRAD uses a fast
beam tracking method based on an automatically generated voxel mesh for computing
viewfactors for the surface to surface radiation.
POSRAD and PERMAS are sequentially coupled using a script-based methodology. Radiative
heat transfer is considered as a thermal load and modies in each coupling step the right side
of equation 2.
The CFD mesh and the PERMAS grid are based upon the same CAD data but are generated
with different meshing tools with different accuracy for the representation of the surface.
STAR-CD V3.2x works most accurately with a trimmed mesh consisting of hexahedral volume
cells in almost all parts of the computational domain. For PERMAS a tetrahedral grid, which
is in most cases automatically created, is adequate. The common interface of both codes is
the surface of both meshes. Therefore a mesh mapping tool is needed to map the physical
properties from one mesh to the other (Figure 2).
MpCCI (Mesh based parallel Code Coupling Interface) (12) has been developed at the
Fraunhofer Institute SCAI in order to provide an application independent interface for the
coupling of different simulation codes. MpCCI is a software environment which enables
the exchange of data between the meshes of two simulation codes in the coupling region.
The coupling is setup in 4 steps with the help of a graphical user interface. There are no
restrictions on the type of transferred quantities; different discretisations at the coupling
regions (interface) are handled by interpolation and the communication is based on common
network protocols such that the simulation codes can run on different machines without
requiring a shared le system. Despite this features MpCCI needed to be extended to meet
the requirements for coupling three domains.
STEADY-STATE UNDERHOOD LOAD CASES
Maximum component temperatures are amongst others reviewed considering two steady-
state load cases: 40 km/h trailer towing uphill and driving at maximum speed. For these
load cases it can be assumed that the energy equation does not signicantly feed back
into the momentum equations. Therefore the ow through the engine compartment
and the convective heat transfer can be computed separately in a rst preliminary CFD
computation. Only relevant heat sources such as the engine, the gearbox and the exhaust
pipes are considered and experimental results are provided as thermal boundary conditions
(5). Afterwards the convective heat transfer data is transferred to the thermal solver and a
thermal analysis is carried out.
20
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
Figure 1: Heat transfer at the interface
boundary of uid and solid
n
For the thermal analysis the nite-element code PERMAS is used. Equation (1) describes the
energy equation solved in PERMAS.
c
p

T

__

t

(

2
T

___

x
2
+

2
T

___

y
2
+

2
T

___

z
2

)
= q
i
(1)
with
T

__

t
= 0 in steady-state computations.
The thermal equilibrium at the interface boundary (Figure 1) of uid and solid parts yields
n
t

grad (T) q
s
+ (T -T
Film
) = 0 (2)
The radiative heat ux q
s
is calculated in a third program named POSRAD. POSRAD uses a fast
beam tracking method based on an automatically generated voxel mesh for computing
viewfactors for the surface to surface radiation.
POSRAD and PERMAS are sequentially coupled using a script-based methodology. Radiative
heat transfer is considered as a thermal load and modies in each coupling step the right side
of equation 2.
The CFD mesh and the PERMAS grid are based upon the same CAD data but are generated
with different meshing tools with different accuracy for the representation of the surface.
STAR-CD V3.2x works most accurately with a trimmed mesh consisting of hexahedral volume
cells in almost all parts of the computational domain. For PERMAS a tetrahedral grid, which
is in most cases automatically created, is adequate. The common interface of both codes is
the surface of both meshes. Therefore a mesh mapping tool is needed to map the physical
properties from one mesh to the other (Figure 2).
This methodology performs quite well in problems involving forced convection which is the
dominating convective heat transfer mode when the vehicle is driving, either at high speed
21
or with an operating fan. However buoyancy driven ows, as the heat soak problem after
stopping and shutting off the engine, cannot be computed by this approach. In this case the
ow in the engine compartment is only driven by local density gradients caused by different
air and component temperature. Flow eld and component temperatures are tightly linked to
each other.
Besides buoyancy driven ows a methodology is required that is able to deal with two uid
streams determining the surface temperature of a component. This is in particular important
for exhaust pipes with an inner exhaust ow and an outer air ow, e.g. the underbody ow.
As mentioned before the temperatures of most heat sources in the engine compartment
were set to known temperatures. As accurate experimental values for the outer surface of an
exhaust system are often not available in an early stage of the development process, a closed
simulation solution is necessary.
COUPLED SYSTEM
Both applications demand feedback between the systems. In the case of underhood
simulation the problem is reducible to two bilateral couplings. One is the reaction of the
structure on radiation and vice versa, and the other is the heat exchange between the uid
and the structure. This can be modelled by a so called staggered loop as illustrated in Figure
3, based on an iterative coupling of the three software tools. In the steady-state simulation
the heat transfer coefcient and lm temperature are sent from the uid and the heat ux
from the radiation code to the structures code, starting a new simulation of the structural
problem. The resulting wall temperature is sent back to the radiation solver and to the uid
code, which start their simulation (Figure 4). The system is converged if the results of all codes
keep equal related to their proceeding results. To improve convergence a relaxation can be
applied to the values at the boundary.
This methodology can be extended such that the staggered system can solve transient
problems. Further investigations are necessary to prove if staggered loops between the
Figure 2: Mesh mapping on different meshes for STAR-CD and PERMAS
22
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
A
t
x(t
n
) x(t
n+1
)
t
B B B B
A A A
domains for each time step are necessary to get a reliable solution (Figure 3 (right)) or if
simple predictor based systems may give accurate solutions as well (Figure 3 (left)).
Figure 3: (left) solution of B is
used to predict the bound-
ary values P(y) of next time
step for A. (right) Inner loops
are used to nd the accurate
solution for each timestep.
TEST CASE EXHAUST SYSTEM
As an example a part of a simplied exhaust system and surrounding components are
considered. The mass ux and the temperature of the exhaust gas are set as boundary
conditions.
Figure 4: Coupling Methodology
Additionally the contact surface of the manifold and cylinder head is set to a xed
temperature. For condential reasons all boundary conditions presented here are arbitrarily
chosen. In a rst step the inner surface of the exhaust system is considered as adiabatic and
the preliminary STAR-CD run for the script-based approach is done. The resulting near wall
temperatures and heat transfer coefcients are mapped to the PERMAS grid as described
earlier. Afterwards the wall temperature is computed with xed boundary conditions for the
convective heat ux.
In a second step the same problem is run with the MpCCI based trilateral coupling approach.
The inner surface of the exhaust system is computed by coupling STAR-CD, PERMAS and
POSRAD iteratively. For the outer surface PERMAS and POSRAD are coupled iteratively while
the convective heat transfer is read from include les analogue to the script-based approach.
The rst computation neglects the cooling effect on the exhaust ow at the outer side of
the exhaust pipes. Therefore the computation tends to result in temperatures of the exhaust
system which are much too high (Figure 5 (right)). To overcome this problem, the temperature
23
steering
column
adiabatic wall
treatment in STAR-CD
MpCCI approach
element 1
Figure 5: Surface temperatures of the simplied exhaust system and surrounding components
(left) temperature of the exhaust gas system for adiabatic wall treatment in STAR-CD and the
MpCCI based approach (right)
distribution of the rst computation with the old approach is used as boundary condition
in a new STAR-CD computation and the resulting convective heat uxes are once again
transferred to PERMAS. After three loops of the old approach similar results compared to
the new one can be observed. It is important to emphasize that in the old approach data is
only manually transferred to the other code after fully converged runs while the MpCCI based
method performs an automatic iterative exchange.
Figure 6: Results for the exhaust system (left) and the steering column (right)
Figure 6 shows the results for element 1 which is located on the exhaust system for the
old approach with different iterations and the result with the new approach. The steering
column (Figure 5 (left)) is chosen for plotting temperature values (Figure 6). It is strongly
affected by the radiation from the exhaust pipes. The overestimated temperatures in case of
an adiabatic wall treatment in STAR-CD can also be found in the component temperatures.
The computational effort comparing both approaches is given in Figure 7. The old
numerical approach is set to the value of one as a reference value. The proposed MpCCI
24
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
25
approach has a little more computational effort then a single loop with the old methodology,
but is much more efcient, especially compared with two or more loops for the old one.
The runtime differences result from the iterative coupling, the leaner communication
architecture of MpCCI compared to the le based old approach and from the reuse of
calculated neighbourhoods for each coupling iteration step. These numbers are strongly
affected by the model size and may differ for other examples.
CONCLUSION
The trilateral partitioned approach presented in this paper demonstrates its capability to
deliver accurate numerical results, with minor effort with respect to runtime and set up
overhead. It is a straight forward extension to the already established method presented in
(4,5) and can be used for buoyancy driven ows or for computing surface temperatures of an
exhaust system considering both the inner exhaust ow as well as the air ow at the exterior.
It also allows as introduced above for extension to transient problems in the near future.
The new approach fulls accuracy and runtime demands within a development process
for passenger cars. In the near future this approach will be used to compute component
temperatures in an entire car underhood, e.g. in case of thermal soak.
ACKNOWLEDGEMENTS
This research was funded by DaimlerChrysler AG, department EP/SAE. The authors would like
to thank Dr.-Ing. Raimund Siegert for supporting this work and for permission to publish this
paper.
Figure 7: Scaled computational effort for
both methodologies
c
p
[J/KgK] specic heat
d [m] Distance
g [m/s
2
] acceleration due to gravity
h [W/m
2
K] convective heat transfer coefcient
q [W/m
2
] heat ux
t [s] Time
P Predicator
T [K] Temperature
x Boundary values of code A
y Boundary values of code B
Greek symbols
[W/mK] Conductivity
[Kg/m
3
] Density
Abbreviations
CFD computational uid dynamics
con Vonvection
S Radiation
MpCCI Mesh-based parallel Code Coupling Interface
NOMENCLATURE AND ABBREVIATIONS
26
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
27
REFERENCES
[1] Bendell, E. Investigation of a Coupled CFD and Thermal Modelling Methodology for
Prediction of Vehicle Underbody Temperatures VTMS 7, 2005-01-2044, Toronto 2005
[2] Srinivasan, K., Woronowycz G., Zabat M., Tripp J. An Efcient Procedure for Vehicle
Thermal Protection Development, SAE 2005-01-1904, Detroit 2005
[3] Fortunato, F., Damiano, F., Di Matteo, L., Oliva P. Underhood Cooling Simulation for
Development of New Vehicles, VTMS 7, SAE 2005-01-2046, Toronto 2005
[4] Reister, H., Maihfer, M. Underhood Component Temperature Analysis for Passenger
Cars VTMS 6 SAE Conference, 2003. C599/028/2003.
[5] Bauer W., Maihfer M. Numerische Simulation der Bauteiltemperaturen eines
Gesamtfahrzeuges, VDI Tagung Berechnung und Simulation im Fahrzeugbau,
Wrzburg, 2004
[6] Binner T., Reister H., Weidmann E.P., Wiedemann J. Underhood Temperature Analysis in
Case of Natural Convection, VTMS 7, SAE 2005-01-2045, Toronto 2005
[7] Binner T., Reister H., Weidmann E.P., Wiedemann J. Aspects of Underhood Thermal
Analyses, in Hucho/ Wiedemann (Editors), Progress in Vehicle Aerodynamics IV,
Numerical Methods, expert Verlag, Renningen 2006
[8] STAR-CD Users manual version 3.15/ Methodology,Computational Dynamics Limited,
London, 2001.
[9] INTES, PERMAS user guide, Version 10.00, INTES GmbH, Stuttgart, 2004
[10] CFDLink, www.owmaster.com
[11] Piperno, S., Farhat, C, Design and evaluation of staggered partitioned procedures for
uidstructure interaction simulations., INRIA, Rapport de recherche no. 3241, 1997
[12] Fraunhofer Gesellschaft SCAI, MpCCI user guide, St. Augustin, 2004
28
COUPLING FLUENT WITH FEM CODES
Mark Pelzer FLUENT Germany
ABSTRACT
The computational uid dynamics code FLUENT provides best-in-class simulation solutions
in its respective application area. With the MpCCI technology provided by Fraunhofer SCAI,
the code can be used to simulate multi-physics like uid-structure interaction (FSI) or thermal
electro magnetism phenomena. This enables the user to simulate both sequential and fully
coupled multi-physics applications with high accuracy. The communication of results between
FLUENT and the nite element codes is handled by Fraunhofer SCAI technology MpCCI,
making the FSI coupling between FEM codes and FLUENT accurate and user friendly. In the
presentation the technical background of the multi-physics approach will be presented and
several examples will be shown.
INTRODUCTION
So called multi-physics problems have been of high interest for recent years. Fields of
combination of multi-physics are structural analysis, uid dynamics, aeroelastics, heat transfer,
radiation, electrodynamics, magnetodynamics, stamping, crash. Simulations have been
able to address some of these combination within single tools. For example uid dynamics,
combustion, multiphase, conjugated heat transfer and radiation within FLUENT or structural
dynamic, heat transfer and electrodynamics in FE tools are already feasible. For some cases
one could also use CFD and FE tools programmable interfaces to implement additional
multi-physics. This is currently very often done in CFD codes since CFD tools often provides an
option to account for moving geometry.
If such on-board means are not sufcient the user can couple software tools manually.
A prerequisite then is identical computational meshes in both tools. A one-way coupling can
be established by exporting mesh and results from one tool in the appropriate format and
import the result into the other tool. In case of a loose physical coupling of the problem even
a two-way coupling can be achieved this way.
A closely coupled problem or a problem that requires different computational meshes in both
tools requires the use of a coupling software. This software has to handle data exchange,
mapping of data from one mesh to the other and the synchronization of physical time in
both tools. For some application also the deformation of the geometry is handled by the
coupling software. Following we will discuss the various level of complexities by examples
between nite element programs and the CFD software FLUENT using the coupling software
MpCCI [1].
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
S
Dr.-Ing. Dipl.-Phys. Mark Pelzer
Fluent Deutschland
Birkenweg 14a
64295 Darmstadt
Germany
Phone: +49 (0) 6151 / 3644 - 120
mp@uent.com
www.uent.de
29
FLUENT
FLUENT provides comprehensive modeling capabilities for a wide range of incompressible
and compressible, laminar and turbulent uid ow problems. Steady-state or transient
analyses can be performed. In FLUENT, a broad range of mathematical models for transport
phenomena (like heat transfer and chemical reactions) is combined with the ability to model
complex geometries. Examples of FLUENT applications include laminar non-Newtonian
ows in process equipment; conjugate heat transfer in turbomachinery and automotive
engine components; pulverized coal combustion in utility boilers; external aerodynamics;
ow through compressors, pumps, and fans; and multiphase ows in bubble columns and
uidized beds.

To permit modeling of uid ow and related transport phenomena in industrial equipment
and processes, various useful features are provided. These include porous media, lumped
parameter (fan and heat exchanger), streamwise-periodic ow and heat transfer, swirl, and
moving reference frame models. The moving reference frame family of models includes the
ability to model single or multiple reference frames. A time-accurate sliding mesh method,
useful for modeling multiple stages in turbomachinery applications, for example, is also
provided, along with the mixing plane model for computing time-averaged ow elds.
Another very useful group of models in FLUENT is the set of free surface and multiphase
ow models. These can be used for analysis of gas-liquid, gas-solid, liquid-solid, and gas-
liquid-solid ows. For these types of problems, FLUENT provides the volume-of-uid (VOF),
mixture, and Eulerian models, as well as the discrete phase model (DPM). The DPM performs
Lagrangian trajectory calculations for dispersed phases (particles, droplets, or bubbles),
including coupling with the continuous phase. Examples of multiphase ows include channel
ows, sprays, sedimentation, separation, and cavitation.
Robust and accurate turbulence models are a vital component of the FLUENT suite of models.
The turbulence models provided have a broad range of applicability, and they include the
effects of other physical phenomena, such as buoyancy and compressibility. Particular care
has been devoted to addressing issues of near-wall accuracy via the use of extended wall
functions and zonal models.
C
O
N
T
A
C
T
30
Various modes of heat transfer can be modeled, including natural, forced, and mixed
convection with or without conjugate heat transfer, porous media, etc. The set of radiation
models and related submodels for modeling participating media are general and can take
into account the complications of combustion. A particular strength of FLUENT is its ability
to model combustion phenomena using a variety of models, including eddy dissipation and
probability density function models. A host of other models that are very useful for reacting
ow applications are also available, including coal and droplet combustion, surface reaction,
and pollutant formation models [2].
SOLVING MULTI-PHYSICS PROBLEMS WITH A COUPLING SOFTWARE
A closely coupled problem or a problem that requires different computational meshes in both
tools requires the use of a coupling software. This software has to handle data exchange,
mapping of data from one mesh to the other and the synchronization of physical time in both
tools. For some application also the deformation of the geometry is handled by the coupling
software.
MpCCI (Mesh-based parallel Code Coupling Interface) has been developed at the Fraunhofer
Institute SCAI as a general software platform for coupling different simulation codes.
MpCCI is a software environment which enables the exchange of data between the meshes
of two or more simulation codes in their coupling region. Since the meshes belonging
to different simulation codes are not compatible in general, MpCCI performs an ux
conservation interpolation. In case of parallel codes MpCCI keeps track on the distribution
of the domains onto different processes.
MpCCI allows the exchange of nearly any kind of data between the coupled codes; e.g.
energy and momentum sources, material properties, mesh denitions or global quantities.
The intricate details of the data exchange are hidden behind the concise interface of MpCCI.
Most of the commercial CFD/FEM applications allow users to add additional features, physical
models, or boundary conditions via a programming interface. Within these user routines
access to internal data structures is possible, either through subroutine parameters and global
variables, or via internal modules for reading and storing data. MpCCI uses these capabilities
for code adaptation. A user-subroutine called after each iteration or time step works as a
hook to MpCCI [3].
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
S
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
S
Figure 1: MpCCI uses a
client-server architecture.
The MpCCI Code-Adapter
links the FEM code to
FLUENT through the MpCCI
Coupling Server.
DESCRIPTION SENDER RECEIVER
NPosition Nodal positions FEM FLUENT
OverPressure Relative pressure FLUENT FEM
RealWallForce Relative wall force FLUENT FEM
Temperature Ambient (uid) temperature FLUENT FEM
WallForce Absolute wall force FLUENT FEM
WallHTCoeff Wall heat transfer coefcient (lm coefcient) FLUENT FEM
WallHeatFlux Wall normal heat ux density FLUENT FEM
WallTemp Wall temperature (nodal temperature) FEM FLUENT
31
COUPLING EXAMPLES FEM-FLUENT (FSI)
The coupling capabilities between FLUENT and FEM codes includes both sequential and
coupled simulations. The following list of quantities gives an overview about the possible data
exchange.
Figure 2: Prediction of temperature on the surface of a cylinder head. Red shows the hottest
regions and blue the coldest regions.
Plotted in EnSight Gold from CEI. Courtesy of Deutz AG
32
Cylinder head thermal-stress prediction
The prediction of thermal stresses in an cylinder head depends strongly on the inuence of
the uid. The heated air enters the ow domain and heats up the material. The temperature
gradient in the material causes thermal stresses. Via one way code coupling it is possible to
send the uid temperature and heat transfer coefcient from FLUENT to ABAQUS/Standard.
With these values ABAQUS can calculate the heat ux and so the temperature distribution
in the material. Afterwards ABAQUS calculates the thermal stresses. Red shows the hottest
regions and blue the coldest regions (Figure 2). A comparison between an ABAQUS stand
alone simulation and an ABAQUS-FLUENT coupled simulation shows a signicant difference
in the temperature distribution on the surface of the cylinder head.
Vernay-VernaFlo

Fluid Flow Valve

In this application a liquid ow through a valve is examined. The pressure inlet boundary
condition changes from 0-40 psi and deforms the valve more and more (Figure 3). A steady
state analysis was done in both FLUENT and ABAQUS. FLUENT started with a low pressure
inlet boundary condition (Figure 4) and sent the relative forces after getting the converged
ow eld. ABAQUS calculates the new shape (Figure 4a) of the geometry and send back this
information via MpCCI. After this the inlet bc. will be increased and the complete procedure
is repeated until it reached the maximum value of 40psi at the inlet (Figure 4b). A comparison
between experimental data and simulation results shows a good agreement (Figure 5).
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
S
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
S
Figure 3: Illustration of the
undeformed and deformed
shape of the uid ow valve.
Plotted in EnSight Gold from CEI
Figure 4: Contours of path lines of the ow eld and von Misses stresses in the valve.
Picture a) shows the starting point, picture b) shows the end shape of the valve.
Figure 5: Comparison
between experimental and
simulation results. The values
are in good agreement.
Under low pressure Under high pressure
Flow control value cross-section
Figure 4b Figure 4a
33
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
S
34
SIMULATION OF A HARD DISK USING ANSYS AND FLUENT
A hard disk is very complex ow system. The high speed of rotation of the disc causes a very
turbulent ow system. This high-grade unsteady ow interacts with the reading head.
The reading head starts to vibrate if the turbulence vortices hit a resonance frequency of
System. To avoid this situation it is necessary to start a lot of difcult tests or the other option
is to simulate this process with a coupled transient simulation between ANSYS and FLUENT.
Figure 6 shows the complete FLUENT model. For this kind of simulations it is necessary to
resolve the reading head geometrically. In general a very ne mesh (ca. 10 mill. cells) is
required because of the turbulence effects in the system. For a realistic simulation one should
use the large eddy simulation (LES) model for turbulence description. Just for simplication
we used a much more coarse mesh (ca. 250.000 cells) and the RNG k-E model with enhanced
wall treatment. The ow medium is air with a constant density.
Figure 8 shows the ANSYS model. It was used the sparse solver with ca. 3000 linear elastic
(solid 45) elements.
The exchanged quantities at the surface of the reading head are the forces (from FLUENT to
ANSYS) and the nodal displacements (from ANSYS to FLUENT). The resulting history plot of
displacement for a special node on the tip of the reading head shows very small values
(Figure 9). The behavior of the reading head is more or less similar to a damped harmonic
oscillator. For a realistic simulation we would expect a different behavior. The reason for this
damped answer from the structural part is the coarse mesh and the simple turbulence model.
CONCLUSION
With the capability of automatic data exchange via MpCCI it is possible to describe different
complex multi physics systems. The interpolation method guaranteed a correct data exchange
between FLUENT and different FEM codes. This avoids time consuming le based exchange
between the codes and gives the exibility to use different meshes.
REFERENCES
[1] Sitzki L., Kaufmann F.H., Dehning C.: Fluid-Structure Interaction: Status and Strategy,
2004 Conference Virtual Product Development (VPD) in Automotive Engineering
[2] FLUENT Inc.: FLUENT 6.3 User Guide Vol 2A, 2005, 9-2
[3] Fraunhofer SCAI: MpCCI 3.0 Overview and Release Notes, 2004, 4
Figure 6: Complete FLUENT
model for the hard disc
Figure 7: ANSYS model for the reading head
Figure 8: History plot of displacements for a
special node on the tip of the reading head
35
36
SOLVING FLUID-STRUCTURE INTERACTION
PROBLEMS USING MPCCI AND THE COMPONENT
TEMPLATE LIBRARY (CTL)
Joachim Rang, M. Krosche, R. Niekamp, H. G. Matthies Technical University of Braunschweig
INTRODUCTION
Problems of uid-structure interaction contain of a uid and structure part. Both problems
are coupled by the boundary conditions on a common interface [MO95]. In the simulation of
uid-structure interaction often so-called partitioned methods are used, i.e. both problems
are solved by different codes. In this talk the programs are controlled by the help of the
component template library (CTL) [KNM03]. To use the CTL, component-based software
systems are needed since the codes are controlled by the help of the CTL. Component-based
software systems have the advantage that the implementations have a longer lifetime and
that linkage of may be incompatible libraries is avoided. Since the uid- and the structure-
code use different meshes and different space and time discretizations the nodes on the
common interface have to be modied with the help of MpCCI. In this talk algorithmic
aspects of the interaction of the uid-, the structure-code, MpCCI, and the CTL are explained.
The CTL has the main task to distribute the subproblems and to control the loops for solving
the linear systems of equations and the loop for the time discretization.
THE PROBLEM
In our case we have a coupling of uid and structure (see [MW01]). The domain is given in
Figure 1. The uid part is modelled with the help of incompressible Navier-Stokes equations,
i.e.

f
( + (( ) )v) + p = r
f
2 = (()

)
= 0.
Moreover some boundary and initial conditions are given. We use the following notations:

f
density of the uid
v velocity
position of the referece ALE-coordinate system
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
S
Figure 1: Domain of the
problem
37
velocity of the referece ALE-coordinate system
p pressure
r
f
body force
The structure is modelled by the following equations

s
u (FS) = r
s
, F = u,
S = (tr E)I + 2E, E = F

F I.
Furthermore we some boundary and initial conditions and the variables have the
following meanings.

s
density of the structure
u displacement
F gradient of the displacement
S second Piola-Kirchhoff stress
, Lam moduli, E Lagrange-Green strain
r
s
body load
Dr. Joachim Rang
Technische Universitt Braunschweig
Institut fr Wissenschaftliches Rechnen
Hans-Sommer-Str. 65
38106 Braunschweig
Germany
Phone: +49 (0) 531 / 391-3007
j.rang@tu-bs.de
www.tu-braunschweig.de
C
O
N
T
A
C
T
38
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
S
SOLUTION TECHNIQUES
There are two classes of solution techniques known in the literature. On the one hand-side
we have direct or the monolithical approach. In this case we solve the whole problem with
one software package. On the other hand a modular approach is possible, i.e. we solve
the subproblems with different codes. Therefore this methods are often called partitioned
methods.
In our case we use for the uid-part the Finite-Volume code OpenFOAM and for the structure
part the Finite-Element code ParaFep. Since OpenFOAM and ParaFep use different meshes
the nodes of the interface have to be approximated. Therefore we use the software-package
MpCCI.
Partitioned methods have several advantages if we compare them with direct methods.
existing software can be used
the best codes for the subproblems can be used (i.e. nite volumes for the uid, nite
elements for the structure)
the discretized problem is smaller than the whole problem
Of course partioned methods have some disadvantages. The numerical results depends
strongly on the iterative solution of the subsystems. Therefore in the literature is a difference
between weak and strong coupling. We use strong coupling methods [MS02, MNS06] as
block-Jacobi
block-Gau-Seidel
block-Newton
THE CODES
As we mentioned before the uid part is solved with software package OpenFOAM. This code
is open source and can be downloaded from the internet. The space discretization is done
with help of nite volumes. The main application eld of OpenFOAM is CFD. It supports mesh
motion and topology changes. The structural part is solved with code ParaFep, developed at
the University of Hannover. The code is open source, too, and uses nite elements.
The main applications are structural mechanics problems. It performs adaptivity and algo-
rithms can be solved in parallel.
SOFTWARE COMPONENTS
Clemens Szyperski and David Messerschmitt dene software components as follows. Deni-
tion A piece of software offering (via an interface) a predened service and which is able to
communicate with other components.
This is a rather general denition and so we mention some criteria for software components.
Multiple-use parallel execution
Non-context-specic exchangeable
Composable with other components
Encapsulated i.e., non-investigable through its interfaces
39
A unit of independent deployment and versioning
It is clear that we need a further code which controls our software-components. This is
done by the so-called Component Template Library (CTL) [KNM03] which has the following
features.
rst partially realised as part of the parallel FE-code ParaFep at the Institute of Structural and
Numerical Mechanics in Hannover, 1995
originally thought as an enhancement of C++ a template (header) library like the STL = no
further libraries or tools are needed
an implementation of the component concept with an RMI semantic similar to CORBA or
Java-RMI.
a tool for component building from existing libraries
suitable for High Performance computing on parallel hardware
also usable by C or FORTRAN programs
compared with CORBA very easy to use
It may a little bit surprising but the idea of software components is rather old as we see in the
following.
1960 so-called software crisis
1968 The idea to componentize prefabricated software rst published at the NATO
conference on software engineering in Garmisch, Germany, titled Mass Produced
Software Components by Douglas McIlroy
Subsequent inclusion of pipes and lters into the Unix operating system
The modern concept of a software component largely dened by Brad Cox of Stepstone,
Objective-C programming language
1990 IBM: System Object Model
Microsoft: OLE and COM
today several successful software component models exist
There are several reasons why we should use software components.
Growing number of (freely) available libraries and programs worth to be reused
Exchangeable software units
Support of distributed parallel run time systems
Avoids linkage of may be incompatible libraries
Longer lifetime of implementations
Next we have to specify how the codes can interact. In this talk we present the Explicit
Message Passing and Remote Method Invocation (RMI). First we consider the Explicit Message
Passing, which is used by MPI and PVM.
Coupling programs by explicit message passing needs inserting of communication points
into source code source and expertise for each program needed
No separation of communication and algorithm difcult maintenance of code
Figure 2: Control of the
whole problem
Solver
CTL
Interface
MpCCI
v
s
solve (v
f
, p
f
)
r
f
p
f
u
s
translator (
f
,
s
)

f

s
Fluid
OpenFOAM
Structure
ParaFep
solve (u
s
, r
s
)
40
F
L
U
I
D
-
S
T
R
U
C
T
U
R
E

I
N
T
E
R
A
C
T
I
O
N
S
Each new pair of coupling produces amount of programming
The Remote Method Invocation (RMI) is used by the CTL has following properties.
Needs component framework
Keeps functional programming style
Exchangeability of coupled components
Type safe communication
The CTL supports the following communication protocols/linkage types.
MPI (Message Passing Interface)
PVM (Parallel Virtual Machine)
TCP/IP (directly via sockets)
dynamic linkage
threads
pipes (to get through a rewall via ssh)
daemon (connect to a running process)
le (reasonable for dump of data to disc)
In our case the interaction of the software components, i.e. of OpenFOAM, ParaFep and
MpCCI, is done in the following way (Figure 2).
The CTL controls the whole process. First some forces are given to the solver ParaFep which
solves his problem and give some displacements values back. The CTL transforms these values
into velocities and send them to OpenFOAM. This code solves the uid part and gives some
pressure values back. The corresponding nodes of displacement and pressure are send to
MpCCI which give back some modied quantities.
Figure 3 shows the control of OpenFOAM. The code obtains some displacement values from
the CTL. Then a mesh is generated. Then OpenFOAM solves the uid problem using the
function icoFoamAutoMotionMod. Finally some pressure values are given to the CTL.
Figure 3: Usage of OpenFOAM
41
OpenFOAM
blockMesh
pBoundary
icoFoamAutoMotionMod
motionU
U
case
read
mesh
call
write displacements
CTL protocol
mesh motion
read pressures
CTL protocol
a
o
f
O
m
t
i
m
e

t
REFERENCES
[KNM03] M. Krosche, R. Niekamp, and Hermann G. Matthies. A component based
architecture for coupling optimisation and simulation software in a distributed
environment. In Walter Dosch and Roger Y. Lee, editors, Proceedings of the ACIS
Fourth International Conference on Software Engineering, Articial Intelligence,
Networking and Parallel/Distributed Computing (SNPD03), 1618 October, 2003,
Lbeck, Germany. ACIS, 2003. http://dblp.uni-trier.de.
[MNS06] H. G. Matthies, R. Niekamp, and J. Steindorf. Strong coupling methods. Computer
Methods in Applied Mechanics and Engineering, 195:2028 2049, 2006.
[MO95] H. Morand and R. Ohayon. Fluid-Structure Interaction. John Wiley & Sohn,
Chichester, 1995.
[MS02] H. G. Matthies and J. Steindorf. Partitioned but strongly coupled iteration schemes
for nonlinear uid-structure interaction. Comput. Struct., 80:19911999, 2002.
[MW01] D. P. Mok andW. A.Wall. Partitioned analysis schemes for the transient interaction
of incompressible ows and nonlinear exible structures. In K. Schweizerhof W. A.
Wall, K.-U. Bletzinger, editor, Trends in Computational Structural Mechanics, pages
689698. CIMNE, Barcelona, 2001.
Figure 1: The two basic parts
of the MpCCI Code API
MpCCI
MpCCI
GUI
Server
data
run
code
adapter
GUI
integration
control
inhouse
code
42
HOW TO CREATE YOUR OWN CODE ADAPTER
Holger Spiess and Pascal Bayrasy Fraunhofer Institute SCAI
THE MPCCI 3.0.6 CODE API
The MpCCI Code API is an interface which allows to create code adapters for inhouse or
research codes. These codes are then treated by MpCCI like any other code in the list of
supported codes. It is thus possible to couple the code with any other supported code or even
couple two inhouse codes using MpCCI.
It is recommended to have at least a basic knowledge of the structure and procedures of
MpCCI before starting to work on the code interface. The rst step towards a code adapter is
to dene which goals shall be achieved:
Which quantities shall be transferred, which kind of simulation transient or stationary
- shall be carried through, and whether the partner code can send and/or receive the selected
quantities.
Access to the source code of the inhouse code is a prerequisite, the mesh denitions must
be read and the coupling manager functions of the adapter library must be called during
execution. Of course, the quantities which shall be sent have to be written and the quantities
which are received must be stored at appropriate places in your code.
The MpCCI Code API consists of two basic parts, which are needed for a coupled simulation.
The rst is the integration into the MpCCI GUI, which allows the user to choose options and
start your code.
The second part is the actual code adapter which is responsible for data exchange with the
MpCCI server.
O
P
E
N

I
N
T
E
R
F
A
C
E

F
O
R

C
O
U
P
L
I
N
G

O
F

I
N
H
O
U
S
E

C
O
D
E
S
43
An MpCCI adapter license for your own code is required to run a coupled simulation.
SCAI also offeres the APIKit for easy code integration, which will be discussed below.
It is recommended to proceed as follows for code integration:
Get the MpCCI Code Integration Kit APIKit with templates for
GUI integration and adapter.
Create GUI integration.
Test GUI integration alone.
Write code adapter.
Test coupling with sample problem and analyze log les.
A more detailed description of procedure and further information is given
in the MpCCI manuals.
GUI INTEGRATION
All conguration les for integrating a code into the MpCCI GUI must be placed in a
conguration directory, which is located in the subdirectory <MPCCI_HOME>/codes.
This contains the les:
gui.xcfdenes the GUI entries for the code.
Scanner.pmPerl script for scanning the input le.
Starter.pmPerl script for starting the code.
Stopper.pmPerl script for stopping the code.
Info.pmget basic information on the code.
Subcmd.pmdenes an MpCCI subcommand.
A template with all of the above les is part of the APIKit. As soon as a conguration
directory is present, the code appears in the list of codes.
The le gui.xcf is the central conguration le for the MpCCI GUI, which is written in XML
format. It contains basic information like name and type of the code, the entries for Models
and Go Step, a list of quantities which can be handled by the code and which information is
passed to Scanner, Starter and Stopper.
Dr.-Ing. Holger Spiess
Fraunhofer Institute for Algorithms and
Scientic Computing SCAI
Schloss Birlinghoven
53754 Sankt Augustin
Germany
Phone: +49 (0) 2241 / 14 - 2647
holger.spiess@scai.fhg.de
www.scai.fraunhofer.de
www.mpcci.de
C
O
N
T
A
C
T
44
O
P
E
N

I
N
T
E
R
F
A
C
E

F
O
R

C
O
U
P
L
I
N
G

O
F

I
N
H
O
U
S
E

C
O
D
E
S
The Scanner is responsible to retrieve basic information which is needed for coupling from the
input le of your code. This includes space dimension, coordinate system, solution type, load
cases, unit system, oating point precision and further code-specic information.
Furthermore the possible coupling components must be identied. The mesh denitions
are not needed, they are directly sent during code execution. All information gained by the
Scanner is stored in a *.scan le.
The Starter script is called when the Start button is pressed in the MpCCI GUI, it needs to
know how to start the code and should pass command line options as dened in the MpCCI
GUI.
The Stopper script is called to stop a code when the Stop button is pressed and should
cause the code to end the simulation as soon as possible. The Info script is an optional
additional script to gather information for later use in Starter and Stopper.
CODE ADAPTER
The code adapter is a plug-in into a simulation code, which can also be based on a user-
dened function interface. The adapter uses the MpCCI adapter library (mpcci.h, libmpcci.
a) to communicate with the MpCCI server. This library is written in C and can be linked with
C/C++ or FORTRAN code.
Figure 2: Code adapter
structure
The code coupling manager functions must be called at specic times during the simulation.
Before any other function, MpCCI_Message_init should be called to specify a list of output
functions to be used by the adapter. Before the time loop is started, MpCCI_Init must be
called to dene the basic communication settings, the list of driver functions and initialize
the connection with the MpCCI server. During the execution of MpCCI_Init the rst driver
function MPCCI_Driver_deneGrid is called to obtain node coordinates and mesh topology.
45
During the simulation the code must call MpCCI_Transfer for data exchange. The adapter
then calls driver functions to either read data for sending or write data which was received
from the partner code via the MpCCI server. Finally, MpCCI_Exit should be called at the end
of the simulation to terminate the connection to the MpCCI server. MpCCI_Exit should also
be used to abort the simulation in case of errors.
MPCCI CODE INTEGRATION KIT APIKIT
The APIKit shall simplify code integration. For this purpose it contains commented templates
for all les in the conguration directory as well as for the code adapter. In addition a simple
example code is included with source code, Makeles and a sample coupling problem.
All steps of code integration are described in detail in the Programmers Guide which is part
VIII of the MpCCI 3.0.6 documentation. A description of the APIKit can be found there as
well. Please note that the APIKit is not included in the standard MpCCI distribution but must
be ordered separately together with an appropriate license.
46
COUPLED SIMULATIONS USING THE ELASTI CITY SOLVER
DDFEM AND FLUENT VIA MpCCI
Dimitar Stoyanov Fraunhofer Institute ITWM
ABSTRACT
DDFEM is a parallel nite element solver for 3D linear elasticity. The solver possesses
signicantly better performance than other universal FE-solvers, especially for certain specic
types of problems, e.g. in the eld of micro mechanics. This higher performance is mainly due
to the domain decomposition approach used to parallelize the nite elements method. The
solver is highly portable over parallel computer architectures supporting the MPI-standard.
DDFEM has been recently extended to handle heat problems, being able in this way to take
into account the temperature effects appearing in the elasticity models: the thermal stress
and strain.
Another new feature in the recent development of DDFEM is the MpCCI-adapter
implemented in a collaboration with the MpCCI group at Fraunhofer SCAI. This enables our
elasticity code to be used in the numerical investigation of FSI-models. Here we present two
test cases of coupling DDFEM with FLUENT using MpCCI:
3D elastic ap in a ow channel;
heat transfer within an exhausting pipe, including the thermo-elastic effects.
INTRODUCTION
Nowadays, despite the variety of commercial solvers available on the market, the question
of the design and development of effective numerical solvers is still open. It is a fact that
many problems can not nd their proper treatment using the existing (commercial) codes.
Typical examples are large sized problems with complicated geometry, e.g. problems in
micro mechanics, see [1], which necessarily require a parallel execution. In this sense, a
specic-purpose in-house code, specially designed and programmed for a particular parallel
architecture, can often provide much better performance than the existing general-purpose
tools.
The parallel 3D-elasticity solver DDFEM is an example for such an in-house development,
undertaken to meet the needs of certain macro- and micromechanical problems. DD in the
name stays for domain decomposition this is the method for parallelizing the numerics.
First and second order hierarchical local approximations are employed ([1]) within tetrahedral
nite elements because the latter are appropriate for automatic mesh generation, providing
at the same time a good approximation for complicated geometries. Recently DDFEM has
been extended to tackle heat problems as well.
O
P
E
N

I
N
T
E
R
F
A
C
E

F
O
R

C
O
U
P
L
I
N
G

O
F

I
N
H
O
U
S
E

C
O
D
E
S
47
The MpCCI driver for DDFEM is a new development also, which makes possible to use
DDFEM in FSI simulations employing the so called weak coupling. Note, that in some
observable future it is rather not very probable some general-purpose (commercial) package
with strong coupling between the ow- and solid solvers to appear. Therefore, tolls like
MpCCI, technically facilitating and intermediating the weak-coupling approaches, are
indispensable.
The present paper consists of two main parts: rst we will briey present the elasticity solver
DDFEM, its basic applications and performance. Then two test examples of coupled FLUENT-
DDFEM simulations via MpCCI will be discussed.
DDFEM: BRIEF DESCRIPTION, BASIC APPLICATIONS, PERFORMANCE
DDFEM has been designed as a reliable, portable and high-performance application for
parallel computers. Its C++ implementation follows the object oriented approach. The solver
references two external libraries:
the mesh partitioning in subdomains uses ParMETIS, see [2], and
the (parallel) iterative solution of the assembled linear system is performed via PETSc [3].
The iterative linear solver, employed in DDFEM, is the conjugate gradients method with
(block-)Jacobi preconditioning for the elasticity and GMRES for the heat equation. Both
methods are provided by PETSc as routines.
The higher performance of DDFEM compared to the one of many other FE-solvers, including
commercial ones, is achieved via effective parallelization: with exception of the input, all
stages of the numerics are performed in parallel. The serial input in many cases can not be
avoided because most of the mesh generators produce a global mesh description i.e. for
the whole domain. After reading it and partitioning the mesh into subdomains, the local
discretization and further the linear system solver work in parallel. This standard approach of
a posteriori mesh partitioning, mentioned so far, is not always appropriate. In some cases,
e.g. in micro mechanics, an a priori partitioning would be more suitable and DDFEM provides
it. Another advantage of the solver is the higher order approximation with hierarchical shape
functions we use. It allows a basis of a degree k+1 to be obtained as a correction to that of
degree k. The advantage in this case, contrary to the lagrangian approach, is that the higher
degree basis should not be entirely reconstructed (recomputed), but only the necessary
Dr. Dimitar Stoyanov
Fraunhofer Institute for
Industrial Mathematics ITWM
Fraunhofer-Platz 1
67663 Kaiserslautern
Germany
Phone: +49 (0) 631 / 31600 - 42 68
stoyanov@itwm.fraunhofer.de
www.itwm.fraunhofer.de
C
O
N
T
A
C
T
Figure 1: Micro mechanics: a bre structure
48
corrections should be found. Further details can be found in [1].
A special input language for DDFEM has been designed, where the mesh description is a list
of nodes and elements, and the boundary condition specication is always local, i.e. related
to a certain nite element (correspondingly to its edge, face) or to a node.
DDFEM has been used at ITWM as both
a standalone elasticity solver for complex large-sized problems in macro- and
micromechanics (Figure 1), and as a
FE-solver, iteratively referenced to perform structural analysis in a shape-optimization
loop (Figure 2).
Now, extending it with a heat solver, one can resolve models including temperature effects.
Both solvers are independent and can be separately referenced. If the coupled problem is
solved, then the heat solver is rst referenced and after obtaining the temperature eld one
could resolve the elasticity model, taking into account the thermal strain effects also.
Typical geometries resolved by DDFEM can be seen on the gures. Figure 1 represents a
complex structure, containing bres of three different materials. DDFEM is used to calculate
the effective elastic properties of the sample: i.e. to nd an energetically equivalent
reference material. In fact this means to solve six boundary value problems for the
domain. Such problems are computationally expensive, require parallel execution because
of their large size (usually ~5-6 millions unknowns), and also need effective ways to handle
the complicated geometry. To meet this requirements DDFEM rst reads the so called voxel
description of the domain (voxel is a small cube, located at a point). Note, that such a
description naturally arises from the computer tomography images, which are usually the
source for the digital reconstruction of such real materials. An a priori domain partitioning,
e.g. using cutting planes, would be more suitable for such problem and DDFEM performs it.
Then in parallel the numerical mesh is generated within the subdomains by subdividing each
voxel into ve tetrahedra and nally the solution starts to obtain the effective elastic tensor.
Figures 2 and 3a represent macroscopic geometries. Particularly the one on Figure 3a is an
outcome of a shape optimisation loop, where the optimisation tool after reshaping the
O
P
E
N

I
N
T
E
R
F
A
C
E

F
O
R

C
O
U
P
L
I
N
G

O
F

I
N
H
O
U
S
E

C
O
D
E
S
Figure 2: Macroscopic geometry: closing plate Figure 3a: Macroscopic geometry
subject of a shape optimization
Figure 3b: Shape optimization loop with TOSCA using DDFEM as elasticity solver
DDFEM
TOSCA
PERMAS DDFEM
INTERFACE
POSTPROCESSOR
DDFEM
(*.uci, *.dat)
PERMAS interface
(*.post)
PERMAS interface
DDFEM input
language
49
domain on each new iteration, e.g. by drilling a new hole - refers to DDFEM as a structure
mechanics solver. These examples use global tetrahedral meshes with further a posteriori
partitioning.
Coupling heat and elasticity gives the possibility to model a large set of physical phenomena.
Figure 4 presents the solution of a simplied model of a casting process, where the
solidication heat is introduced via the source term in the heat equation. Then, after
obtaining the temperature eld, one can calculate the (elastic) deformations in the metal cast.
The geometry of the problem can be seen on Figures 4a and 4c a metal alloy cast surrounded
by a mold of sand. A source of a (latent) heat is associated with the middle rib of the metal
cast and on the outer surfaces of the sand mold a convective cooling via the surrounding air
is supposed.
Figure 4a: Temperature
distribution at the mold
surface
Figure 4b: Displacements in
the metal alloy cast
Figure 4c: Temperature
distribution in the cast and in
the cross section of the mold
50
Figures 4a and 4c represent the temperature distribution on the mold surface and inside it
correspondingly, while on Figure 4b the displacements of the metal alloy cast are given.
O
P
E
N

I
N
T
E
R
F
A
C
E

F
O
R

C
O
U
P
L
I
N
G

O
F

I
N
H
O
U
S
E

C
O
D
E
S
Figure 5: DDFEM scalability
up to 1000 CPUs
(IBM RS6000 Power 3)
0 100 200 300 400 500 600 700 800 900 1000
0
0.5
1
1.5
2
2.5
3
3.5
4
x 10
10
IBM RS6000/SP POWER3, seaborg.nersc.gov
Num PEs
T
o
t
a
l

F
l
o
p
s
/
s
e
c

(
r
e
p
o
r
t
e
d

b
y

P
E
T
S
c
)
51
The performance of the numerical procedures is extremely important in the practical
simulations. For example, the repeating references to DDFEM in the shape optimization loop
(see Figure 3b) increase signicantly the performance requirements before the solver.
We may say that DDFEM successively meets them and it replaced practically entirely some
other commercial solvers used before in the shape optimisation iterations at ITWM on our
parallel distributed architectures. In fact the real speed up is up to 15-20 times in favour of
DDFEM, see [1].
Here by real we mean the acceleration obtained in a normal everyday working environment
for real problems of middle size.
The core of the DDFEM numerics has been shown to scale very well up to 1000 CPUs, see
Figure 5. This has been tested by a special bench-mark version of the solver, see [1], where
one considers pre-dened geometries of a simple form a combination of spheres, cylinders,
etc. Such geometries could be easily partitioned a priory and therefore the mesh generation
within each subdomain could be also performed in parallel, providing in this way very good
scalability for the whole combination of the grid generator and the FE-solver. The scalability in
the real applications is also good enough, but the serial input of the mesh puts some upper
limit and is a kind of performance bottle-neck.
Figure 6b:
Initial displacements
of the ap on
iteration
Figure 6c:
Flap displacements,
iteration 2
Figure 6d:
Flap displacements,
iteration 9
Figure 6a:
Conguration of
the ap within
the channel
52
PROGRAMMING A MPCCI ADAPTER FOR DDFEM
We will skip the technical details here. Instead it is better to say that to create an own adapter
is quite achievable task, providing that a detailed documentation of the API-functions of
MpCCI is available (which soon will be the case). The necessary extensions of the source code
of DDFEM, caused by the adapter, were rather auxiliary and certainly did not affect the overall
design of the solver.
FSI SIMULATIONS WITH DDFEM AND FLUENT
The following test models are a kind of classics for MpCCI. Although the geometry is not
that complicated, the complexity of a FSI-simulation is valid for the both examples.
3D elastic ap
This is the example from the MpCCI coupling cases tutorial. The geometry of the channel,
the position of the ap inside it, and the ow mesh can be seen on Figure 6a. The calculated
pressure by the ow solver (FLUENT) at the ap walls is further considered as distributed load
over the ap surface, and then the solid solver (DDFEM) calculates the displacements, i.e. the
new position of the ap in the channel. The transfer quantities ow pressure and ap nodal
coordinates are transmitted by MpCCI on each time iteration. The size of the solid problem
is about 90 000 unknowns. The ap positions on iterations 0 (initial position), 2, and 9 could
be seen on Figures 6b, 6c, and 6d correspondingly. The residuals of the FLUENT solution are
presented on Figure 6e.
O
P
E
N

I
N
T
E
R
F
A
C
E

F
O
R

C
O
U
P
L
I
N
G

O
F

I
N
H
O
U
S
E

C
O
D
E
S
Scaled Residuals
(Time= 9,2750e-02)
FLUENT 6.3 (3d, pbns, dynamesh, ske, unsteady) Feb 07,2007
Figure 6e: Fluent residuals, 3D elastic ap
53
Exhausting pipe
As a physical model this is the same example as the well known exhaust manifold of the
MpCCI meetings, but in our case the geometry is more simplied. We consider an inow of
air with a temperature of 300[C] in a metal tube a kind of exhausting pipe, see Figures 7.
The outer surface of the pipe is a subject of convective cooling by the surrounding air, the
inner surface of the pipe is the interface between the ow- and the solid domains.
The inlet is on the right-hand side of the pictures, the outlet on their left-top part. The goal
is to investigate both the uid and solid for a steady-state temperature distribution and the
resulting thermal strain and stress in the metal pipe. As a ow solver we use FLUENT, as a
heat and elasticity solver in the solid DDFEM. On each iteration the ow solver calculates
the heat ux normal to the walls inside the pipe and this ux is then transmitted by MpCCI
as a boundary condition for the heat equation solver in the solid. Solving for both the heat
and the elasticity equations in the solid, one nds the temperature distribution on the
internal wall of the pipe and the new nodal positions at the interface, which are transferred
back to the ow solver by MpCCI. The iterations continue until steady-state for the ow is
reached. Figures 7a, 7b, and 7c represent the nal ow solution the velocity, temperature,
and pressure elds correspondingly, on the Figure 7d one can see the residual history.
Figure 7e represents the nal temperature distribution in the solid, while Figure 7f give the
displacements.
Figure 7a: Velocity eld within the pipe Figure 7b: Temperature distribution of the
ow within the pipe
Figure 7c: Pressure distribution of the ow
within the pipe
Figure 7d: Fluent resudual history
Figure 7e: Temperature distribution within
the solid pipe
Figure 7f: Displacements of the solid pipe
54
O
P
E
N

I
N
T
E
R
F
A
C
E

F
O
R

C
O
U
P
L
I
N
G

O
F

I
N
H
O
U
S
E

C
O
D
E
S
CONCLUSION
This paper presents two examples for coupled uid-structure simulations between FLUENT
and the elasticity solver DDFEM via MpCCI. The heat- and elasticity solver DDFEM is shortly
introduced and the implementation of a MpCCI driver for the solver is briey discussed.
Finally two test cases for FSI are considered: 3D elastic ap in a ow channel and a simple
model of an exhausting pipe.
55
ACKNOWLEDGEMENTS
For Figure 5 this research used resources of the National Energy Research Scientic
Computing Center, which is supported by the Ofce of Science of the U.S. Department of
Energy under Contact No. DE-AC03-76SF00098.
Special thanks also to Inga Shklyar, Fraunhofer ITWM, who helped to automatically produce
the descriptions of the uid-solid interfaces in the coupled simulations. Many thanks to the
MpCCI team at Fraunhofer SCAI, particularly to Dr. Holger Spiess, for the help and assistance
to program the MpCCI code adapter for DDFEM.
REFERENCES
[1] Andr, Stoyanov, Error indicators in the parallel nite element solver DDFEM, (Berichte des
Fraunhofer ITWM, No 83, 2006).
[2] METIS: Multilevel Partitioning Algorithms, (http://glaros.dtc.umn.edu/gkhome/views/metis).
[3] Portable, Extensible Toolkit for Scientic Computation, (http://www-unix.mcs.anl.gov/
petsc/petsc-as/).
56
FLUID-STRUCTURE SIMULATIONS USING MpCCI 3.0.5 WITH
ABAQUS AND AN IN-HOUSE CFD-CODE ON DISTRIBUTED
RESOURCES
Felix Lippold University of Stuttgart IHS
ABSTRACT
Current trends in the development of new technologies for renewable energy systems
show the importance of tidal and ocean current exploitation. But this also means to enter
new elds of application and to develop new types of turbines. Latest measurements at
economically interesting sites show strong uctuations in ow and attack angles towards
the turbine. In order to examine the dynamical behaviour of the long and thin structure of
the turbine blades, coupled simulations considering uid ow and structural behaviour need
to be performed. For this purpose the CFD code FENFLOSS, developed at the IHS, is coupled
with the commercial FEM-Code ABAQUS. MpCCI is used for steering of the simulation and
managing the data transfer between the two computational elds. Since the CFD domain
has to be modelled in a certain way, the grid size tends to be in the range of about one
million grid points. Hence, to solve the coupled problem in an acceptable time frame the CFD
calculations have to run on more than one CPU. Whereas, the structural grid is quite compact
and, does not request that much computational power. The implementation of the code
driver for the inhouse CFD code FENFLOSS for the current version MpCCI 3.0.5 is presented.
Further issues concerning programmers interfaces and moving grid algorithms for CFD codes
are mentioned. Furthermore, the setup of the technical and computational problem as well as
the results obtained on a distributed computational grid will be discussed.
BASIC EQUATIONS
In order to simulate the ow of an incompressible uid the momentum equations and mass
conservation equation are derived in an innitesimal control volume. Including the turbulent
uctuations yields the incompressible Reynolds-averaged Navier-Stokes equations, see
Ferziger et al.[3]. Considering the velocity of the grid nodes U
G
due to the mesh deformation
results in the Arbitrary-Langrange-Euler (ALE) formulation, see Hughes [5].
O
P
E
N

I
N
T
E
R
F
A
C
E

F
O
R

C
O
U
P
L
I
N
G

O
F

I
N
H
O
U
S
E

C
O
D
E
S

U
i

___

x
i
= 0 (1)

U
i

___

t
+ (U
j
U
G
)
U
i

___

x
j
= -
1

_

p

___

x
i
+

__

x
i

(

U
i

___

x
j
+
U
j

___

x
i

)

i .


(2)
Reynolds Stresses
Figure 1: Physical and numerical coupling model
pressure
loads
pressure
loads
Structure (CSM) Structure (CSM)
Fluid (CFD)
Mesh (CMD)
Fluid (CFD)
deformation
deformation
update
57
The Reynolds Stresses are usually modelled following Boussinesqs vortex viscosity principle,
see Boussinesq [1]. To model the resulting turbulent viscosity, for most engineering problems
k- and k-w-models combined with logarithmic wall functions or Low-Reynolds formulations
are applied.
The discretization of the momentum equations using a Petrov-Galerkin Finite Element
approach, see Zienkiewicz [10] and Gresho et al. [4], yields a non-linear system of equations.
In FENFLOSS a xed point iteration is used to solve this problem numerically. For each
iteration the equations are linearized and then smoothed by an ILU-preconditioned iterative
BICGStab(2) solver, see van der Vorst [9]. The three components can be solved coupled or
decoupled, see Ruprecht [8]. Working on parallel architectures MPI is applied in the smoother
to exchange data in the matrix-vector and scalar products, see Maihoefer [7].
The discretised structural equations with mass, damping and stiffness matrices M, D, and K,
load vector f, and displacments u can be written as
M + Du + Ku = f , (3)
see Zienkiewicz [10].
Dipl.-Ing. Felix Lippold
University of Stuttgart IHS
Pfaffenwaldring 10
70550 Stuttgart
Germany
Phone: +49 (0) 711 / 685 63267
lippold@ihs.uni-stuttgart.de
www.ihs.uni-stuttgart.de
C
O
N
T
A
C
T
Figure 2: Exchange schemes for loosely coupled problems
Fluid
Structure
2
3
5
n n+1
1
4
Fluid
Structure
n-1/2 n+1/2
n n+1
1
2
3
4
t
Fluid
Structure
1
2
3
4
t
n n+1

1. Deformations for timestep n

2. Update grid and integrate
fluid field to timestep n+1
4. Put fluid loads to structure
5. Advance structural solution
to timestep n+1
Weak coupling
First order (or less!)
" Subcycling with
predictor for structure
" Time shifting
1. Deformed grid at time n+1/2
2.-4. see above
Second order (almost) for struc-
ture middle point intergration
r
n+1/2
= d
n
+ d
n

2

58
However, a data exchange between the codes has to be arranged, including the interpolation
between the two computational meshes. This is the reason why for engineering problems
usually the second approach is employed. In order to account for the coupling and to avoid
unstable simulations, some coupling schemes have been developed, e.g. see Farhat et. al. [2].
Figure 2 shows three of the most used schemes.
COUPLING FLUID AND STRUCTURE
Seen from the physical point of view a uid-structure interaction is a two eld problem.
But numerically, the problem has three elds, where the third eld is the uid grid that has to
be updated after every deformation step of the structure, to propagate the movement of the
uid boundaries, the wetted structure, into the ow domain, see Figure 1: threeeld.
The solution of the numerical problem can be arranged in a monolithic scheme, which solves
the structural and ow equations in one step. This method is applied for strongly coupled
problems, e.g. for modal analyses. Another scheme, suitable for loosely and intermediately
coupled simulations, is the separated solution of ow and structural equations with two
independent codes. In the separated scheme well proven codes and models can be used.
O
P
E
N

I
N
T
E
R
F
A
C
E

F
O
R

C
O
U
P
L
I
N
G

O
F

I
N
H
O
U
S
E

C
O
D
E
S
Figure 3: Coupling scheme
CFD-Start
P
i
c
a
r
d

i
t
e
r
a
t
i
o
n
d
e
f
o
r
m
a
t
i
o
n
s

y
s
u
b
i
t
e
r
a
t
i
o
n
p
r
e
s
s
u
r
e

l
o
a
d
s
(
i
n
t
e
r
p
o
l
a
t
i
o
n
)
END
Update
Mesh
CSM-Start
Solve for y
Matrix assembly u
i
Solve turbulence
Solve for u
i
Update p
59
Using the simple coupling scheme results in the ow chart shown in Figure 3. The left part
shows the solution of the ow problem as it is implemented in FENFLOSS. On the right side
the structural solver and the grid deformation is sketched. The moving grid algorithm may
be implemented either directly in FENFLOSS or as a stand alone software called during the
coupling step. Here, an extended interpolation method, which is implemented in FENFLOSS,
see Lippold [6], is used.
COUPLING FENFLOSS AND ABAQUS WITH MPCCI
In order to couple the in-house code FENFLOSS with the commercial software ABAQUS to
perform uid-structure simulations, an interface between FENFLOSS and MpCCI has to be
available. The old MpCCI 2.0 SDK interface in FENFLOSS is replaced with a new one based
on the adapter-driver connection required by the current MpCCI versions. In this case it
means mixing the MpCCI-API written in C with classical FORTRAN code. The implementation
of some special function calls in FENFLOSS is possible, creating an unexible interface
introducing new code into the program. Furthermore, the code then can only be used for
special analysis types, i.e. uid-structure simulations. Since, it is recommended to link the
adapter and driver via an external library, the best and most exible way is by using the
shared object interface implemented in FENFLOSS. Once written, all algorithms needed for
coupling the two codes, updating the grid, as well as coupling schemes can be implemented
by user-subroutines without changing the simulation code itself. This yields a better code
Figure 4: Architecture for coupling FENFLOSS with the shared object adapterdriver
library and ABAQUS
Initialisation
CFD
API
User functions
(MpCCI-Adapter)
Solver
Grid-update
CFD-solution
API-calls (Data
transfer)
Finalization
Store data
A
d
a
p
t
e
r
D
r
i
v
e
r

(
A
P
I
)
MpCCI
GUI/Client
Scan
Start
Stop
Server
Data transfer
Interpolation
Log files
P
e
r
l
ABAQUS v6.5
Solver
Initialisation
Data transfer
Fixed transfer
points
A
d
a
p
t
e
r
D
r
i
v
e
r
Visualisiation
COVISE
60
O
P
E
N

I
N
T
E
R
F
A
C
E

F
O
R

C
O
U
P
L
I
N
G

O
F

I
N
H
O
U
S
E

C
O
D
E
S
quality and a faster development of the interface. The set-up and conguration for the
graphical MpCCI interface is done with simple perl scripts. Figure 4 shows the architecture of
the coupled codes. Because not all MPI releases provide the export option of the environment
on the master node to the slave nodes, the MpCCI environment variables are exported via
les written and set in the MpCCI code adapter.
CONCLUSION
For most engineering applications it is found that loosely coupled schemes are appropriate
to model uid-structure interaction problems. Therefore, proven and established codes
and models can be used. In order to couple the in-house CFD code FENFLOSS with the
commercial structural analysis code ABAQUS MpCCI is used as an interface. Since the MpCCI
SDK implementation cannot be reused, a new programmers inter face is developed for
FENFLOSS and the adapter-driver unit can be linked as shared object library. It turns out that
this approach brings for effort in the beginning but saves time and problems in the end.
61
REFERENCES
[1] J.Boussinesq, Thorie de lcoulement tourbillonant et tumultueux des liquides dans les
its rectilignes grande section, Tome I - II, Gautier-Villard (1897).
[2] C. Farhat, M. Lesoinne, P. le Tallec, Load and motion transfer algorithms for uid/
structure interaction problems with non-matching interfaces, Computer Methods in
Applied Mechanics and Engineering, No. 157, 95 (1998).
[3] J. H. Ferziger, M. Peric, Computational Methods for Fluid Dynamics, Springer
(third Ed. 2002).
[4] P.M. Gresho, R.L.Sani, Incompressible Flow and the Finite Element Method,
Vol. I, John Wiley & Sons (1999).
[5] T. J. R. Hughes, W.K. Liu, T.K. Zimmermann, Lagrangian-Eulerian Finite Element
Formulation for Viscous Flows, Computer Methods in Applied Mechanics and
Engineering, 29, 329-349 (1981).
[6] F. Lippold, Fluid-Structure-Interaction in an Axial Fan, HPC-Europa report (2006).
[7] M. Maihoefer, Efziente Verfahren zur Berechnung dreidimensionaler Strmungen mit
nichtpassenden Gittern, PhD-Thesis, University of Stuttgart (2002).
[8] A. Ruprecht, Finite Elemente zur Berechnung dreidimensionaler turbulenter Strmungen
in komplexen Geometrien, PhD-Thesis, University of Stuttgart (1989).
[9] H. A. van der Vorst, BI-CGSTAB: A fast and smoothly converging variant of BICG for the
solution of nonsymmetric linear systems., SIAM J. Sci. Stat. Comp., 13, 631-644 (1992)
[10] Zienkiewicz, O. C., Taylor, R. L., The Finite Element Method, Vol. I, McGraw-Hill (1989)
Figure 1: Motor protective
circuit breaker PKZM0
62
SIMULATION OF THE SWITCHING ARC FOR A MOTOR
PROTECTIVE CIRCUIT BREAKER
Christian Rmpler Fraunhofer Institute SCAI and Albert Zacharias Moeller GmbH
ABSTRACT
The electric current is omnipresent and indispensable in our life. But there are always
threatening dangers connected with it. One of them is the electric short circuit and we have
to protect people and facilities from its effects. This protection is often provided by power
circuit breakers or motor protective circuit breakers. These devices are able to detect the short
circuit, limit the current, switch it off and provide save insulation before any harm can occur.
During the switching process of short circuits an electrically conductive plasma, the electric
arc, exists between the metal contacts. The thermal, aerodynamic and magnetic interactions
of this electric arc within the switching device are decisive for a safe current switch-off.
These processes are simulated using the nite element code ANSYS to solve the magnetic
eld problem and the nite volume code Fluent to solve the uid dynamics problem. Both
codes are coupled using the coupling interface MpCCI.
This work shows numerical results of an electric arc simulation using the example of a motor
protective circuit breaker from Moeller series production. The simulation results show good
agreement with experimental results.
INTRODUCTION
The electric current is omnipresent and indispensable in our life. It is delivering energy and
allows us to operate and control a lot of applications. The simplest case of controlling is the
switch-on and switch-off operation where electrical switches are required. But there are
always threatening dangers connected with the electrical current. Current consumers can be
overloaded, faulty equipment can be overheated and become sources of re. An other
danger is the electric short circuit and we have to protect people and facilities from these
effects. This protection is often provided by power circuit breakers or motor protective circuit
breakers (Figure 1). These devices are able to detect the short circuit, limit the current, switch
it off and provide save insulation before any harm can occur.
E
L
E
C
T
R
I
C
A
L

E
N
G
I
N
E
E
R
I
N
G
63
Electro-mechanical switches like the motor protective circuit breaker presented here
mechanically open or close the electric circuit. They offer advantages compared to
semiconductor devices; specically with low power losses in current carrying operation
(closed contacts) and save insulation in the open position. The presented motor protective
circuit breaker operates electro-mechanically as well. During the switch-off operation an
electrical arc occurs within the quenching chamber. This discharge is responsible for the
current limiting, for a smooth decay of the current to zero and for the safe switch off.
A special challenge is the interruption of short circuit currents because the small circuit
breaker is loaded by currents of several kA within a short time. It is very important to know
the behaviour of the electric arc inside the switch during the short circuit and to inuence it
properly.
ELECTRIC ARC SIMULATION
The electric arc
After opening of the electrical contacts the current is not interrupted immediately.
Because of the inductance L in the electric network that is always present (motor winding,
cables) only a continuous change of current i is possible. Otherwise in case of discontinuous
current changes an innite voltage would be induced:
u
L
= L
di

__

dt
(1)
To continue the current after the contact separation a conductive media, the electric arc
which is a plasma discharge is created between the separating contact pieces. Because of
the high temperatures the gas is ionized and becomes conductive. The current can be carried
by the plasma which leads to ohmic power losses and thus the temperature in the plasma
is kept on a high level and the continuity of the discharge is ensured. In case of low voltage
switchgear the medium used is air. Depending on temperature and pressure the oxygen and
nitrogen of the air is dissociated and ionized and the thermodynamic and electric properties
of the arc are strongly dependent on this. For instance the electrical conductivity (Figure 2,
we used data from [1], further data can be found in [2],[3]) is very small below temperatures
of 5000 K because no appreciable ionization occurred and thus not enough free charge
carriers are available. Until about 25000 K the conductivity is increasing due to the increasing
ionization.
Dipl.-Ing. Christian Rmpler
Fraunhofer Institute for Algorithms
and Scientic Computing SCAI
Schloss Birlinghoven
53754 Sankt Augustin
Germany
Phone: +49 (0) 2241 / 14 - 2135
christian.ruempler@scai.fhg.de
www.scai.fraunhofer.de
www.mpcci.de
C
O
N
T
A
C
T
E
L
E
C
T
R
I
C
A
L

E
N
G
I
N
E
E
R
I
N
G
Figure 2: Electrical
conductivity of air at
environmental pressure,
comparison of different
literature sources
64
Due to the difference of electric potentials and the electrical conductive path between the
contacts an electrical current ows within the arc which is generating a magnetic eld (self
magnetic eld). With the design and conguration of xed and movable contact (these carry
the current as well) also an magnetic eld is generated (called magnetic blow eld).
The resulting force acts on the arc and drives it towards the metal splitter plates (Figure 4)
where the arc is cooled, elongated and separated by the metal sheets. This results in a fast
increase of the voltage drop.
Now this quickly increased arc voltage u
b
is acting against the driving line voltage u
N
and is
therefore inuencing the current (see equation 2 with electrical resistance R ).
L
di

__

dt


= u
N
u
b
R i (2)
In order to extinguish the arc quickly and a to effectively limit the current it is necessary that
the arc voltage u
b
exceed the line voltage u
N
noticeably.
SIMULATION
The dynamic ow processes of the plasma are described by NavierStokes equations (mass
balance, momentum balance, energy balance) combined with additional equations for the
description of the electric and magnetic eld. In the literature similar approaches can be
found ([4][6]) for the simulation of low voltage switchgear. The solution of the ow problem
is done by the commercial nite volume code Fluent [8] and the electro-magnetic problem is
solved by the commercial nite element code ANSYS [9].
The equations for electric and magnetic eld are coupled with the ow equations through
the momentum source term Lorentz force density, the material property electrical conductivity
and the energy source term power loss density. Depending on pressure and temperature
the electrical conductivity is calculated via Fluent user dened function (UDF) and sent to
Figure 3 a: Calculated
temperature in
the middle plane
compared to
Figure 3 b: optical exposures,
current 1000 A, time
300 s, 350 s, 400
s, 450 s
65
the partner code ANSYS. Using this material property the electric and magnetic elds are
calculated offering the results Lorentz force density and power loss density. These values
are sent back to Fluent for the next solution step. More details of underlying equations and
assumptions can be found in [7].
The data exchange of the volume quantities between the simulation codes is done by the
software MpCCI (Mesh based parallel Code Coupling Interface, [10]). Using this method it
is possible to use specialized codes for each discipline with full functionality (e.g. nonlinear
nite element methods for ferromagnetic materials) as well as suitable discretisation of
the geometry within each code. Furthermore due to the separation of the whole task into
different tasks the computation is accelerated. Thus the high numerical effort of complex
three dimensional simulations can be handled better.
Parallel to the modelling process experiments with a simple geometry (parallel runners) have
been done for verication purposes. Experimental data (current, voltage, pressure, magnetic
eld) have been compared with simulation results to obtain inferences for the modelling
process.
APPLICATION
Motor protective circuit breaker PKZM0
Circuit breakers are often known from the fuse boxes at home. They switch off when lines are
overloaded or a faulty device causes a short circuit. Manual switching is possible as well and
provides for safe insulation. The functional principle of the motor protective circuit breaker is
the same as that of the miniature circuit breaker, but all features and device characteristics are
tuned to full the special demands of motor protection and industrial application. Actuators
are detecting the current ow continuously and in case of an overload current the contacts
are separated depending on the amount and duration of this overload. For this contact
separation a pretensioned latch is triggered. Furthermore in case of short circuit special
Figure 3 shows the simulated temperature compared with optical exposures of the parallel
runner geometry. Further results can be found in [7], [11] and [12].
Figure 4: Quenching
chamber of the motor
protective circuit breaker
PKZM0, consisting of a
connection in series of
two separate quenching
chambers
66
mechanisms enable a faster opening. The process of current switch off follows after contact
separation within the quenching chamber (Figure 4). This chamber is specially designed for
the fast switch off of high short circuit currents with prospective currents up to 50 kA.
The switch is limiting the current to an effective value of 10 kA.
The processes during the switch-off operation is as follows (see also Figure 4):
The contacts open, an initial arc is formed.
The special design of the current path leads to electrodynamic forces driving the arc
towards the splitter plates. Due to the ferromagnetic effects of the iron material of the
splitter plates the forces are amplied.
Between the metal sheets the arc is cooled, elongated and separated. The arc voltage is
increased and will act against the current, limit the current and switch off the current
(see equation 2).
Within some milliseconds this switch off operation is done where power in the order of 10
6

Watt is dissipated. Because of the temperature increase an over pressure of several bar inside
the chamber is produced, which is released through special exhaust openings. The quality of
a short circuit switch-off operation is basically inuenced by the arc movement behaviour and
the penetration of the splitter plates by the arc. Again the arc movement can be controlled
by thermal, magnetic and aerodynamic manipulation of the arc. These processes are under
examination in case of electric arc simulations.
Simulation of the quenching chamber
The simulations are done using ANSYS and Fluent simultaneously as described before. The
Fluent model represents the quenching chamber, as shown in Figure 4. Effectively modelled
is one half of a single chamber, a quarter of the geometry shown in the picture. To keep the
computation time short the model consists of only 50000 cells. We used an explicit time
discretisation scheme.
E
L
E
C
T
R
I
C
A
L

E
N
G
I
N
E
E
R
I
N
G
Figure 5: Calculated temperature distribution
inside the quenching chamber. The iso-
surfaces of temperature representing the arc.
Figure 6: Prepared motor
protective circuit breaker for
arc observation
67
On the one hand the ANSYS nite element model represents the quenching chamber, which
is geometrically identical to the Fluent model. On the other hand the surrounding air has to
be modelled for the magnetic eld calculation. Compared to the Fluent model the ANSYS
model is build up coarser with about 35000 elements. Because of the nonlinear magnetic
material behaviour of the iron parts (e.g. splitter plates) a nonlinear solution has to be
obtained for a series of stationary steps.
A xed geometry is assumed. This means that the opening process of the movable contact is
not included. The calculation starts (t = 0 ms) assuming completely open contacts. Here we
present results from a calculation of 2 ms arcing time.
The results we obtain are temperature, pressure and velocity distribution depending on time
from Fluent as well as the distribution of current density, voltage, magnetic eld from ANSYS.
From the temperature distribution, as shown exemplary in Figure 5, one can evaluate the
position and shape of the arc at different time points.
Verication
Beside the arc simulation experimental investigations have been performed. To get optical
records of the arcing process the optical access was needed. Therefore the sidewall has been
perforated like a breadboard as shown in gure 6. The holes are covered by a transparent
plate which is additionally sealed by an o-ring.
The prepared switch was used in a short circuit interruption, where the arc was lmed with
a highspeed video camera. Some pictures of this movie are presented together with the
corresponding simulation results in gure 7 exemplarily.
Figure 7: Comparison of
simulation results and
experiment
68
The simulation started at t = 0 ms with fully open contacts patching a hot zone into the uid
model representing the initial arc, whereas the arc could spread out a little bit during contact
opening phase in the experiment. Afterwards the arc moves towards the splitter plates.
Beginning at 1.0 ms the arc penetrates the splitter plates almost, where a short reignition in
front of the splitter plates can be seen on the left side.
E
L
E
C
T
R
I
C
A
L

E
N
G
I
N
E
E
R
I
N
G
SUMMARY
The short circuit interruption is a special challenge in the development process of switching
devices such as motor protective circuit breaker, power circuit breaker or miniature circuit
breaker, particularly because the tests using real devices can only be done in a late stage
of the development process. Here the arc simulation was introduced into the development
process providing hints about the arc behaviour.
69
Even though a lot of effects are included in the simulation, the arc simulation model itself is
still under development. The model is improved continuously and tested by means of adapted
experiments.
The application of the simulation model to series devices and experimental investigations
enables a realistic estimation of the signicance of the simulations for the development
process. Additionally this short link to the application pushes the modelling process ahead.
REFERENCES
[1] S. Selle, U. Riedel: Transport Coefcients of Reacting Air at High Temperatures. In:
38
th
Aerospace Sciences Meeting & Exhibit, American Institute of Aeronautics and
Astronautics, Reno, NV, USA, 10.13. January 2000.
[2] J. Yos: Revised Transport Properties for High Temperature Air and its Components,
AVCO Corporation, Technical Release (1967).
[3] M. Capitelli; G. Colonna, A. DAngola: Thermodynamic properties and transport
coefcients of high-temperature air plasma. In: 28
th
IEEE International Conference on
Plasma Science and 13
th
IEEE International Pulsed Power Conference, digest of papers,
S. 694-7, Las Vegas, Nevada, USA, 2001.
[4] Frank Karetta: Dreidimensionale Simulation wandernder Schaltlichtbgen. VDI Verlag
Dsseldorf, Fortschritt- Berichte Band 21, Nr. 205, 1998.
[5] Manfred Lindmayer, Erik Marzahn, Alexandra Mutzke, Matthias Springstubbe: Low
voltage switching arcs experiments and modeling. In: 15
th
Symposium on Physics of
Switching Arc Brno, Czech Republic, 2003.
[6] Hartwig Stammberger, Thomas Daube, Carsten Dehning, Michael Anheuser: Arc
simulations in realistic low voltage arcing chambers. In: 21
st
Intern. Conf. on Electrical
Contacts, Zrich, Switzerland, 9.12. Sept. 2002.
[7] Christian Rmpler, Frank Reichert, Hartwig Stammberger, Peter Terhoeven, Frank
Berger Experimentelle und numerische Untersuchung des Lichtbogenlaufverhaltens In:
Kontaktverhalten und Schalten, 18. Fachtagung Albert-Keil-Kontaktseminar,
Karlsruhe, 5.-7. Okt, 2005
[8] Fluent 6: Fluent Inc., 10 Cavendish Court, Centerra Resource Park, Lebanon, USA
(http://www.uent.com).
[9] ANSYS, Rev. 10.0, ANSYS Inc., Canonsburg, PA, USA, (http://www.ansys.com).
[10] Mesh-based parallel Code Coupling Interface, (http://www.mpcci.de).
[11] Christian Rmpler, Frank Reichert, Hartwig Stammberger, Peter Terhoeven, Frank Berger
Numerical study of the electrical arc movement supported by experiments In: 23
st
Intern.
Conf. on Electrical Contacts, Sendai, Japan, 6.9. June 2006.
[12] Frank Reichert, Frank Berger, Christian Rmpler, Hartwig Stammberger, Peter Terhoeven:
Experimental studies of the arc behaviour in low voltage arc rail arrangements supporting
numerical simulations In: 23
st
Intern. Conf. on Electrical Contacts, Sendai,
Japan, 6.9. June 2006.
70
MpCCI-COUPLED SIMULATIONS FOR ELECTRICAL-
DISTRIBUTION EQUIPMENT
Ian Lyttle, Benjamin Pulido Schneider Electric
MOTIVATION
MpCCI has become an indispensable tool for multi-physics simulations, allowing for
co-simulation using software tools native to each set of physics. Furthermore, the open
architecture of MpCCI offers considerable possibilities for customization, permitting optimized
solutions for specic classes of problems.
At present, MpCCI is used principally by specialists who have training and experience with
MpCCI. Greater use of the tool can be made, for example, if an everyday user of thermal-
analysis software can work with an everyday user of electromagnetics-analysis software to
make a co-simulation, without being an everyday user of MpCCI.
Much of the co-simulation community is motivated by uid-structure interaction (FSI)
modeling, where surface regions are coupled between a uid simulation and a structural
simulation. However, at Schneider Electric we are concerned with problems where
volume regions are coupled to each other. In this work, we consider problems where an
electromagnetics simulation is coupled to a thermal simulation. For electrical-distribution
equipment, the electromagnetics simulation calculates volume-based Joule (resistive) heating,
based on temperature-dependent electrical resistivity. Accordingly, the thermal simulation
calculates the volume-based temperature, based on the Joule heating.
For this work, we use FLUX (offered by Cedrat) as our electromagnetics simulation tool; we
use ICEPAK or FLUENT (offered by ANSYS) as our thermal simulation tool; we use MpCCI as
our coupling tool.
SOLUTION
The rst step in our efforts to automate the coupling procedure was to develop a naming
system (within Schneider Electric) for co-simulation. It was decided that two levels of naming
are required: the class level and the scheme level.
We dene a coupling class by the types of physics used. For example, the coupling class
magneto-thermal refers to the coupling of an electromagnetics simulation with a thermal
simulation. We dene a coupling scheme by the broad assumptions we make within each of
the physics. For the magneto-thermal example, the coupling scheme harmonic-steady-state
identies the electromagnetics problem as harmonic, and the thermal problem as steady-
state. This class-scheme combination is abbreviated to MT-HSS.
E
L
E
C
T
R
I
C
A
L

E
N
G
I
N
E
E
R
I
N
G
Figure 1: Flow of information
for the magneto-thermal
coupling class
Figure 2: Simplied owchart
for the harmonic steady-
state coupling scheme
FLUX
(EMAG)
ICEPAK
(THERMAL)
MpCCI
(COUPLING)
send
mesh
send
temperature
receive
power losses
solve
thermal problem
send
mesh
receive
temperature
solve
emag problem
send
power losses
interpolate
mesh
interpolate
data
interpolate
data
FINISH
START
Converged?
N Y
71
Ian Lyttle
Schneider Electric
3700 6th Street SW
Cedar Rapids, IA 52404
USA
Phone: +1 (0) 319 368 - 3033
ian.lyttle@us.schneider-electric.com
www.schneider-electric.com
C
O
N
T
A
C
T
72
E
L
E
C
T
R
I
C
A
L

E
N
G
I
N
E
E
R
I
N
G
Our next step was to dene the scheme further: All co-simulations of a given scheme have
the same exchange of volume data and scalar data; processes follow the same owchart.
The data ow for MT-HSS is shown graphically in Figure 1, and as a owchart in Figure 2.
From the perspective of the MpCCI project, having dened the coupling class and scheme,
the only items particular to a co-simulation are the names of the simulation les and the
names of the regions to be coupled.
Having established these denitions for the MT-HSS scheme, we wrote a set of scripts for
FLUX and ICEPAK. To automate the construction of the MpCCI project le, we developed a
simplied GUI in which the user species the following information:
Coupling class, e.g. magneto-thermal
Software, e.g. FLUX and ICEPAK
Coupling scheme, e.g. harmonic steady-state
Release of simulation software, e.g. FLUX 9.3.2, ICEPAK 4.2.8
Names of simulation les, e.g. FLUX problem, ICEPAK case
Runtime information, e.g. amount of memory for FLUX solver
Once the coupling class and software are chosen, a tabbed window appears where the
remaining choices are specied. By following a specic naming convention, a script is used
to match the regions automatically. Upon the choice of coupling scheme, specic scripts
used to control FLUX and FLUENT (the ICEPAK solver) are put into place. The simplied GUI
constructs and saves MpCCI project le and launches the MpCCI GUI; the co-simulation is
then launched from the MpCCI GUI.
By providing a simplied GUI based on the coupling class and scheme, we are able to provide
everyday thermal and electromagnetics users with a reduced set of choices as compared with
the MpCCI GUI. As well, we are able to automate the matching of the regions, which can
save time and possibility of error. For a well-dened coupling class and scheme, this lowers
considerably the threshold of MpCCI experience needed to make an effective co-simulation.

TEST CASES
Shown in Figure 3 are the results for a simple problem, where a copper wire is subjected to
an alternating current. On the left of this gure are shown the electromagnetic skin-effects;
on the right are shown the temperatures for the wire and the surrounding air.
We use this approach to solve problems with more-complicated geometries such as:
Busbar systems
Circuit breakers
Switchboard systems
Figure 3: Results for a simple MT-HSS co-simulation
73
One of the features of this automation method is that the MpCCI setup for a switchboard
system is no more time-consuming than for the copper-wire example. This may be notable,
considering that a switchboard system may have hundreds of coupled regions, while the
copper-wire has only one.
POTENTIAL IMPROVEMENTS
The developments previewed by Fraunhofer SCAI for MpCCI version 4 offer the possibilities to
make more-sophisticated coupled analyses. One of these possibilities is stronger coupling
for transient analysis; however, this depends on the partner software as well as MpCCI.
Consider the transient FSI problem using, for example, FLUENT and ANSYS. Ideally, one could
choose a time-step size for FLUENT, t
F
, according to the uid problem and a time-step size
for ANSYS, t
A
, according to the structural problem. One could also choose an MpCCI
time-step size, t
C
, denoting the physical times for making exchanges. Presumably, the
MpCCI time-step would be an integer multiple of both the FLUENT time-step and the ANSYS
time-step.
A owchart of how such a stronger coupling might be implemented is shown in Figure 4.
In this arrangement, there would need to be an iterative agreement between ANSYS and
FLUENT at the end of every MpCCI time-step. Within the ANSYS cycle, the loading would
vary linearly with time for the MpCCI time-step. Correspondingly, within the FLUENT cycle,
the displacement would vary linearly with time for the MpCCI time-step.
E
L
E
C
T
R
I
C
A
L

E
N
G
I
N
E
E
R
I
N
G
Figure 4: Outline for how stronger coupling might be implemented
ANSYS t=t
0
t=t
0
+ t
A
Preserve
displacement at
t=t
0
+ t
C
Preserve
loading at
t=t
0
+ t
C
t=t
0
+ t
C
t=t
0
+ t
C
t=t
0
+ t
F

t=t
0
MpCCI Interpolation Converged?
FLUENT
Interpolation t=t
0
+ t
C
N
N
Y
74
To implement such an algorithm, it will be necessary for the partner software (FLUENT,
ANSYS, etc.) to be able to save and recover their solver states at arbitrary time-steps, while
maintaining their connections to MpCCI. It may be important to be able to save the solver
state for an arbitrary time-step, and not just the previous time-step. This is because the user
may wish for the MpCCI time-step to be a multiple of an individual softwares time-step.
For those software for which this step-back procedure is not currently possible, we encourage
their management and developers to work with Fraunhofer SCAI to nd ways to implement
this idea.
ACKNOWLEDGEMENTS
The authors would like to thank the following people and organizations for their ongoing
technical support:
Pascal Bayrasy, Fraunhofer SCAI (MpCCI)
Rmy Perrin-Bit, Cedrat (FLUX)
Bill Wangard, Ansys Inc. (FLUENT)
Vincent Leconte, Schneider Electric
ANSYS is a registered trademark of Ansys Inc.
FLUENT is a registered trademark of Ansys Inc.
FLUX is a registered trademark of Cedrat
ICEPAK is a registered trademark of Ansys Inc.
MpCCI is a registered trademark of Fraunhofer SCAI
75
76
PRESSURE MOULDING SIMULATION USING FLUENT AND
ABAQUS
Jasper Kidger FLUENT UK
ABSTRACT
This paper presents the methodology and results of a co-simulation using MpCCI (v3.0.5),
FLUENT (v6.3) and ABAQUS (v6.6-2). The simulation concerns a pressure-moulding
application, where the interest lies in the movement of a exible plastic membrane, forced
into a mould by a high external pressure. Whereas the basic moulding process could be
simulated using the FE code alone, in this case it is important also to account for the
movement of the air within the mould cavity.

The loading is applied as a uniform external pressure in the ABAQUS model, on the top
surface of the exible membrane. The FLUENT model simulates the air region inside the
mould, which deforms as the membrane moves.
INTRODUCTION
This work concerns the prediction of the moulding process as a exible membrane is formed
into its nal shape by the action of an external pressure. The cavity of the mould (into which
the membrane is forced) has three ventilation holes of a very small diameter through which
the trapped air must pass as the exible part is formed into shape.
This paper presents work performed by Fluent Europe for Amcor Flexibles. Amcor are one
of Europes largest suppliers and market leaders in exible packaging, supplying to a wide
range of food, beverage and health care markets. Amcor Flexibles are existing users of both
FLUENT CFD software and the ABAQUS structural-nite element (FE) software. Traditionally
Amcor have simulated this process by FE alone. However in order to gain a more detailed
understanding, and improve their computational prediction capability, Amcor wished to
explore the coupling of FLUENT and ABAQUS (using MpCCI) so that the movement of the air
could be accounted for when predicting the structural response.
A sketch of the practical arrangement is shown in Figures 1 and 2. The membrane is held
across the top of a rigid mould. The cavity of this mould originally contains air at atmospheric
conditions. A high external pressure is applied to the top surface of the membrane to drive
the process. As the membrane takes up the shape of the mould, the trapped air is ejected
through three small holes on the base of the mould.
.
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
Figure 1: Sketch of Start of Moulding Process
Figure 2: Sketch of End of Moulding Process
Applied Force
(Uniform Pressure)
Flexible
Membrane
Rigid
Mould
Vent to
Atmosphere
BEFORE
Air
AFTER
Membrane
Pressure-
Moulded to
Shape
77
ABAQUS MODEL
If the pressure inside the mould cavity could be assumed to be constant (or indeed known
as a function of time), then the pressure moulding simulation could be performed using a
structural-FE code on its own, without the need for co-simulation with a CFD code.
Simulations have been performed in this manner with ABAQUS-EXPLICIT (v6.6-2). Figure 3
shows the mesh on the top membrane this is a square surface of 11,796 M34DR elements,
held encastre around the perimeter of the square. Figure 4 shows the model from below, in
which the rigid mould can be seen, along with the three vents from the mould cavity. This
rigid surface is built of 11,724 elements.
Jasper Kidger
Fluent Europe Ltd.
Shefeld Business Park
6 Europa View
Shefeld S9 1XH
United Kingdom
Phone: +44 (0) 0114 2818888
jasper@uent.co.uk
www.uent.co.uk
C
O
N
T
A
C
T
Figure 3: Top View of the ABAQUS Model Figure 4: Bottom View of the ABAQUS Model
Figure 5: Transient Prole of the Displacement
of the Centre Node of the Flexible Membrane
(ABAQUS Only)
78
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
Applying a constant pressure of 6 bar gauge on the top face (the same pressure was used
for the co-simulation discussed later), and running the ABAQUS model on its own (hence
assuming a constant, atmospheric, pressure on the lower face of the membrane) results in the
deformation behaviour shown in Figure 5. This plot shows the movement, with time, of the
node at the centre of the membrane. There is a signicant amount of initial bouncing, but of
decreasing magnitude. When the tip of the membrane reaches its nal location on the oor
of the mould, no further displacement of this node occurs, however the simulation continues
whilst the membrane expands into the outer corners of the mould. The nal deformed shape
of the membrane, coloured with von-Mises stresses, can be seen in Figure 6.
Figure 6: Deformed Shape of Membrane
from ABAQUS-only simulation (colour
contours show von-Mises Stresses)
79
FLUENT MODEL
In order to solve for the uid ow within the mould cavity, it is necessary to use a CFD code
capable not only of compressing its mesh as the domain changes shape, but also one capable
of remeshing (adding or deleting cells) at very frequent intervals, as a result of the extreme
deformation present here. Hence the unique dynamic mesh capability of the FLUENT CFD
software (v6.3) was used for this work.
At every timestep, as the new location of the deforming boundary is received from ABAQUS,
FLUENT rst performs a spring-smoothing action to compress/stretch the existing mesh to t
the new shape of the domain. Then, the quality of the resulting mesh is checked. If any cells
are found either to exceed a prescribed skewness, or whose lengthscales exceed set minimum
or maximum values, then these cells are deleted, and a new mesh grown in just that region
of the uid domain. Since the remeshing is purely a local issue close to poor-quality cells, and
not a global remeshing, this makes for a very efcient process that is affordable to perform
at every timestep during the simulation. The compute time for the remeshing work at each
timestep is of approximately the same magnitude as the compute time for solving the uid-
ow equations at that timestep.
Figures 7a-d show cross-sections of the FLUENT mesh during the solution process. A small
minimum length for the mesh is needed in order to handle the region close to the crater
lip, hence comparing Figures 7a and 7b shows that as the deforming membrane moves
downwards, a zone of compressed cells is evident below the membrane (the depth of this
region is a function of the spring-smoothing constant applied to the dynamic mesh). As the
deformation then becomes more severe (Figures 7c and 7d), it becomes clearer the extent to
which cells have been deleted as the volume of the uid region shrinks.
Figure 7a: Initial FLUENT mesh on mid-plane Figure 7b: FLUENT mesh on mid-plane
Figure 7c: FLUENT mesh on mid-plane Figure 7d: FLUENT mesh on mid-plane
80
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
The solutions here were made on a full 360 representation of the mould. Although two
symmetry planes do exist in the model, which would allow a 90 sector to be simulated,
this was found to create problems because of the added constraints on the mesh along the
common edge between the two symmetry planes.
It is a requirement of FLUENTs dynamic mesh model that the topology of the uid domain
cannot change during the simulation. As the exible membrane takes the shape of the
mould, consideration needs to be given to the way of representing the region where the two
surfaces (membrane and mould) have come into contact. There are two ways in which this
could be accommodated. The nodes along the two boundaries could be slid along as the
contact point moves. However, the approach adopted here has been to shrink the dimensions
of the rigid mould by 1 % in the ABAQUS model only. When the position of this membrane is
transmitted to FLUENT, this preserves a small, nite, gap between the membrane and mould
wall, in which a mesh can remain (as can be seen on Figure 7d).
81
Other key details of the FLUENT model are as follows:
The air was modelled as an ideal gas
The Pressure Based Coupled Solver was applied. This simultaneously solves the momentum
and pressure correction equations, and is ideally suited to applications (like this) where
compressible effects are important on the resulting ow.
Turbulence was modelled using the standard, k-epsilon turbulence model.
The mesh was originally 139k cells, which reduced to 76k cells at the end of the simulation.
Spring-smoothing was applied to the dynamic mesh model with a spring constant of 0.7.
Any cells with a skewness of greater than 0.9 were automatically remeshed, as were any
cells above or below a set lengthscale threshold.
Timesteps of 10-6s were used, with the data exchange with ABAQUS, and potential
remeshing, occurring at every timestep.
MPCCI SETTINGS
The co-simulation of ABAQUS and FLUENT was made with the use of MpCCI, version 3.0.5.
This was set for a gauge-pressure based FSI simulation, with loads being transmitted across
the deforming boundary between ABAQUS and FLUENT. Both codes were run in parallel
(job run on a dual-CPU computer), by taking the explicit transfer option.
ABAQUS (itself timestepping at 10-7s) was set to exchange data with FLUENT every 10-6s in a
loose manner. All other options within MpCCI were left at their defaults.
RESULTS
Figure 8 shows the displacement of the centre-node of the membrane (in a similar manner
to Figure 5), but comparing the results from the ABAQUS-standalone simulation (blue line)
and the co-simulation with FLUENT (magenta line). The co-simulation ran until the centre
node rst met the base of the mould, at which point the topology of the problem changes
and the middle vent became blocked. It can be seen that the duration of the event (from
Figure 8: Transient Prole of the
Displacement of the Centre Node of the
Flexible Membrane, comparing Standalone
and Coupled Simulations
Figure 9: Thickness of the Membrane, Centre-
Plane Fluid Pressure, and the CFD Mesh at a
time of 0.002 seconds
Figure 10: As Figure 9, but at the end of the
Transient Co-Simulation
82
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
start, to centre node reaching base of mould) is about 50 percent greater in the co-simulation
case than in the standalone case, and that the oscillatory behaviour of the membrane has
continued throughout this time period.
Figure 9 presents a snap-shot of the moulding process at a time of 0.002s. The two data-sets
have been combined in this image, with the contour plotting of the membrane representing
sheet thickness, and the centre-plane of the uid region coloured by pressure. The ability
of FLUENTs dynamic mesh technology to continuously deform and rebuild the mesh as the
domain changes shape is clearly evident.
Figure 10 shows the nal point of the co-simulation, as the membrane meets the mould base,
in which the nal shape of the uid mesh, and the thicknesses of the membrane, can be
seen.
The computations were performed on a 3GHz dual-CPU machine (with FLUENT and ABAQUS
each running simultaneously, with one processor per code). The total compute time was 5
days wall-clock time. Over this period FLUENT performed 3,400 timesteps, ABAQUS 34,000
timesteps, and there were 3,400 2-way communications through MpCCI between the two
codes.
CONCLUSIONS
A uid-structure interaction co-simulation has been performed for a pressure-
moulding application in which a very large change in shape of the uid region has been
accommodated. In order to perform this simulation, ABAQUS FE software was used to
simulate the response of a exible membrane to an applied force which was constant on
one side, but which varied on the other side as a result of the varying uid conditions in the
mould cavity. Simultaneously, FLUENT CFD software was used to simulate the uid ow in the
mould as the air was ejected, taking account of the moving boundary from the membrane,
83
and the very small exit apertures through which the trapped air could exit. The coupling of
these two engineering simulation packages (FLUENT and ABAQUS) was made using MpCCI
software. This enabled, at every time-step, the position of the structure to be transmitted
from ABAQUS to FLUENT, and the forces on the moving surface to be transmitted from
FLUENT to ABAQUS.
FLUENTs unique dynamic mesh capability is ideally suited to this application, since arbitrarily
large changes in shape of the uid region can be resolved. At every timestep, FLUENT applied
spring-smoothing to the mesh to accommodate the change in shape, as well as deleting
and rebuilding the mesh in regions where cell skewness was too great, or cell dimensions
exceeded prescribed thresholds.
As a result of this co-simulation, it was demonstrated that the presence of the air in the
mould cavity does have a signicant inuence on the moulding process, and therefore there
are clear reasons why one should consider adopting a coupled uid-structure interaction
technique.
ACKNOWLEDGEMENTS
We are very grateful to the following people for their assistance in this work:
Amcor Flexibles Ltd, Bristol, UK, for provision of original ABAQUS input model.
ABAQUS UK, for technical support on use of the ABAQUS software.
Fraunhoffer-SCAI for technical support on issues relating to the ABAQUS-FLUENT coupling.
84
SOLVING FLUID-STRUCTURE INTERACTION USING ABAQUS
Albert Kurkchubatsche ABAQUS US
ABSTRACT
As uid-structure interaction (FSI) gains momentum in the industry, the use of established
numerical solution techniques to enable coupled simulations is becoming relevant. We outline
various solution techniques for solving uid-structure interaction problems using ABAQUS
and provide industrial examples from various industry sectors.
ABAQUS provides signicant build-in capabilities for modeling uid. This includes ALE,
equation of state, hydrostatic uid elements, a capability to compute current and wave
loading on submerged and partially submerged structures, and acoustic-structural capability.
In cases where the ABAQUS uid capability is limited, you may use one of the approaches
described below to obtain a solution in conjunction with a computational uid dynamic (CFD)
code.
Linear FSI includes the class of application where the structural deformations are small and
the linearized displacements and uid ow computation may be solved entirely within the
CFD code. In such cases, ABAQUS can export structural stiffness, mass and damping matrices
to the CFD solver. These approaches are CFD centric, that is the ow results and structural
displacements may be readily obtained, however the structural response (e.g. stresses and
strains) are not available. For cases where structural response is important you need to use a
partitioned approach described below.
Non-linear FSI includes the class of applications where it is important to consider the effect
of structural non-linearity using nite element formulations. For non-linear structural ap-
plications we resort to a partitioned treatment of the coupled system in which the interac-
tion effects are communicated in an explicit manner via Mesh-based Parallel Code-Coupling
Interface, MpCCI. With the upcoming release of ABAQUS V6.7, ABAQUS supports a link with
FLUENT and STAR-CD.
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
V5 Simulation Products
and CAA V5 Partners
The World is
Non Linear
Open simulation
platform
Unied FEA
Realistic Simulation
85
SIMULIA AND ABAQUS, INC.
Albert Kurkchubatsche
ABAQUS US
39221 Paseo Padre Parkway, Suite F
Fremont, California 94538-1611
USA
Phone: +1 (0) 510 / 794 5891 - 103
albert.kurkchubasche@abaqus.com
www.abaqus.com
C
O
N
T
A
C
T
Automotive Crash Simulation and Airbag Deployment using ABAQUS/Explicit
6-DOF solver Linear FSI
Nonlinear FSI
Staggered Approach
Specialized
techniques
Monolithic
approach
Structure rep-
resented in the
uids code as a
6 DOF entity
Compliance ma-
trix/eigen value
approach to solv-
ing the structural
problem inside a
uids code
Structure and uid equa-
tions solved separately
with code coupling and
mapping at the interface
1. SPH: Meshless
method
2. Immersed
Boundary Tech-
niques
Single set
of equa-
tions for the
uid and
structural
domains
Suitable for rigid
body motions in
a uid.
Suitable for
linear structural
problems
Suited for weak to
moderately strong cou-
pling physics problems.
Implicit coupling well
suited for tackling unsta-
ble FSI problems
Suitable for prob-
lems where struc-
tural modelling
is too complex or
deformations are
signicant
Suited for
most all cou-
pling physics
problems
Examples: IC en-
gines, rigid valve
movement
Examples: Slosh-
ing, vortex-in-
duced vibrations
Examples: pulsatile blood
ow, tire hydroplaning,
dispensing
Examples: Air-
bag/parachute
deployment
Examples:
All
86
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
SIMULIA: The DS Simulation Brand
Vision and Mission
The SIMULIA

vision is to make simulation an integral business practice that

enables engineers and scientists to reduce dependence on physical testing
supports multidisciplinary collaboration and
drives innovation in research and development
The SIMULIA

mission is to be the leading provider of simulation solutions for engineering

and scientic simulation
powered by an open, simulation centric framework that facilitates
collaborative decision making,
manages the simulation lifecycle and
achieves cross-disciplinary efciencies
FSI FROM AN ABAQUS PERSPECTIVE
Pro-Star (CD-Adapco)
interface for ABAQUS
Analysis of Landing Gear using ABAQUS
87
FSI Technical Approaches
Sequential Coupling: Heat Transfer
File-based, sequential approach commonly used in industry via ODB API
Interface to
FLUENT
STAR-CD
third-party codes
In-house codes
Rigid-body Displacements
FSI problems involving rigid-body displacement are most effectively handled by the CFD solver
Most CFD solvers include 6-dof solver capability
The ABAQUS co-simulation approach using MpCCI can be used for problems involving
complex mechanisms using connector elements
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
A typical application: SWAGELOK pressure regulator
Courtesy Fraunhofer SCAI
Nonlinear Structural Response
ABAQUS provides an open system approach to couple with third-party CFD
software using MpCCI
any third-party CFD software compatible with MpCCI can couple to ABAQUS
Fraunhofer SCAI will support an adapter toolkit for in-house codes with the release of
MpCCI 3.0.6
ABAQUS, Inc. fully supports and qualies the following products
ABAQUS and ANSYS / FLUENT
ABAQUS and CD-Adapco / STAR-CD
88
Linear Structural Response
ABAQUS co-simulation approach using MpCCI can be used for problems with
linear structural response
Various approaches are offered by CFD partners
Limited to structural deformation only (no stress recovery)
UDEX simulation performed with ABAQUS
acoustic / structure capability
89
ABAQUS FSI PRODUCT UPDATE
Status with FLUENT
Next FSI product release in Q2, 2007
ABAQUS 6.7-1 with FLUENT 6.3.26 and MpCCI 3.0.6
ABAQUS will jointly support the FSI solution with FLUENT using MpCCI
Releases to be qualied
ABAQUS to provide frontline support for joint customers
ABAQUS/CAE FSI plugin module to be available to customers with Version 6.7-1
Status with STAR-CD
First jointly qualied product will be available with
ABAQUS 6.7-1 with STAR-CD 4.02 and MpCCI 3.0.6
ABAQUS will jointly support the FSI solution with CD-Adapco using MpCCI
Releases to be qualied
ABAQUS to provide frontline support for joint customers
STAR-CD Version 3.26 is compatible with ABAQUS version 6.5 / 6.6 via MpCCI 3.0.5
Although we have not qualied this product, we will support mutual customers
who are using these products
INDUSTRIAL FSI EXAMPLES
Industrial Use Case Overview
Highlight use cases from industry segments using ABAQUS for FSI
Automotive
Biomedical
Manufacturing
Aerospace
Consumer goods
Offshore/Energy
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
Pro-Star Interface to ABAQUS MpCCI Interface with FLUENT
90
Review how ABAQUS is currently being used to solve engineering problems involving FSI
Each industry segment will highlight use cases with different approaches to solve
the uid structure interactions.
Automotive: Coupled Heat Transfer
Coupled heat transfer analysis
File based sequential coupling approach
Open system approach via MpCCI
BIOMEDICAL: DRUG-ELUTING STENTS
Drug-eluting stents deliver medication to the surrounding tissue after implantation to help
keep the passageway from re-closing
Courtesy CD-Adapco
Model Courtesy: Vernay Labs
Computational Modeling of Drug-Eluting Stents: Modeling Device and Drug Delivery,
ABAQUS Inc., FLUENT, US FDA researchers to be presented at 5
th
World Congress of
Biomechanics, Munich, Germany, July / August, 2006 and BMES, Chicago, October 2006
The mention of commercial products, their source, or their use in connection with material
reported herein is not to be construed as either an actual or implied endorsement of such
products by the Department of Health and Human Services
Biomedical: Heart Valves
Analysis of heart valve motion
Unsteady pulsatile ow
Hyperelastic material properties for artery wall
1/6
th
symmetry model
MpCCI interface with STAR-CD
91
Manufacturing: Flow devices
Elastomeric components in uid ow:
Minimize impact of inlet pressure variation for reliable and consistent operation
Nonlinear, large deformation structural analysis from ABAQUS
Turbulent ow, cavitation effects, remeshing from FLUENT
MpCCI interface with FLUENT
Example Courtesy: General Atomics Aeronautical Systems, Inc.
Consumer Products: Packaging
Drop test failure analysis of water bottle
ABAQUS provides build-in capabilities for modeling uid
ALE
Equation of State
Hydrostatic Fluid Elements
92
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
Aerospace: ZAERO Interface to ABAQUS
Critical aerodynamic loads lead to instabilities:
Flutter analysis
Flight loads (trim) analysis
Courtesy: Bayer Material Science, LLC
93
Offshore/Energy: Oil platform
ABAQUS/Aqua
Current and wave loading on submerged and partially submerged structures
ABAQUS CAE PLUG-IN FOR FSI
Complementary product to MpCCI GUI
MpCCI GUI facilities execution of FSI simulation between a CFD and structural
analysis group
Models may be prepared by different groups with different software
Groups may be located at remote facilities
The aim of the ABAQUS CAE plug-in to allow structural users to work in the familiar
ABAQUS CAE environment to dene
the domain geometry and mesh,
loads/boundary condition,
run the coupled simulation, and
post process results all within ABAQUS CAE.
Plugin is available to ABAQUS users via an ABAQUS Answer
Plug-in currently supports FLUENT and in the future STAR-CD
Animation: ABAQUS CAE Plug-in for FSI
SUMMARY
ABAQUS is being used for FSI across all major industry segments
We showed examples from various segments
Most engineering analysis involve FSI in one form or the other
Different solution techniques are being used in industry to solve these FSI problems
MpCCI is being employed to solve the most complex FSI simulation
As part of Dassault Systmes, ABAQUS will be the foundation for multi-physics simulation
and the SIMULIA vision
We are looking forward to continue our efforts with Fraunhofer SCAI
94
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
95
96
COUPLING STAR-CD WITH ABAQUS TO MODEL FSI
INTERACTION OF A BODY ENTERING A FLUID
Philip Morris Jones CD-adapco
INTRODUCTION
CD-adapco have been solving uid ow problems for over 25 years. The software developed,
STAR-CD and STAR-CCM+, have a wide range of physical models that span a wide range
of ows, from supersonic ows and space craft re-entry to the mixing of foodstuffs such
as yoghurt. The latest versions even go beyond computational uid dynamics (CFD) to
computational continuum mechanics (CCM) and so model heat transfer in solids, stress
analysis, melting and solidication for casting etc.
As is often found, despite having such a range of models applicable to different uids and
ow regimes, the majority of the ow simulations are concerned with simple ows of air
or water. These are very common everyday uids but are very different. Air at standard
temperature and pressure has a density of 1.292 kg/m
3
whereas water is much more dense at
999.8 kg/m
3
. The whole atmosphere, a depth of some kilometers, exerts a pressure of about
100,000 Pa at sea level but just 10m of water produces the same pressure.
Many designs are aimed at containing or working with one of these uids, but a real problem
comes when an object is submitted to changing conditions that go from air to water and
back. This can lead to failure through a number of mechanisms, but commonly it can be
through peak loads exceeding the design limits or cyclic loads leading to failure by fatigue.
This paper looks at such ows where a body and a free surface (the interface between a light
and heavy immiscible uids) interact and coupling these forces with a structural model of the
body.
FLUID STRUCTURE INTERACTION WITH A FREE SURFACE
To correctly model uid structure interaction with a free surface, you rst need a number of
building blocks in place:
Model free surface ows in a stationary domain
Model free surface ows in a moving domain
Couple the ow with a structural model
MODEL FREE SURFACE FLOWS IN A STATIONARY DOMAIN
One modelling approach used by CD-adapco to model free surface ows is called the volume
of uid (VOF) approach. In this technique we solve transport equations for the volume
fractions of component uids:

i
=
V
i

__

V
(1)
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
Figure 1: Blunt body in a ow tank Figure 2: CFD mesh and free surface
Figure 4: Comparison of wave prole
along hull computed at TUHH with
experimental data obtained at Ship
Research Institute, Tokyo
Figure 3: Average water level in symmetry
plane and along hull, and average velocities
in symmetry plane in water and air
97
The properties of the effective uid (e.g. density) vary in space according to the volume
fraction of each component i.e.
=
i

i
(2)

Extensive validation studies have been performed to check the VOF implementation.
A classic experiment is blunt body ow.
Philip Morris Jones
CD-adapco
200 Shepherds Bush Road
London
W6 7NL
United Kingdom
Phone: +44 (0) 20 7471 6200
philip.jones@uk.cd-adapco.com
www.cd-adapco.com
C
O
N
T
A
C
T
N
c
i=1
Figure 5: Computation and experiment of sloshing in a tank.
98
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
MODEL FREE SURFACE FLOWS IN A MOVING DOMAIN
CD-adapco can model ows with moving meshes. In the most general case, a mesh can be
made to translate, rotate or distort in any prescribed way, by specifying time-varying positions
for some or all of the cell vertices. This type of mesh movement, which is sometimes referred
to as Arbitrary Lagrangian-Eulerian, can accommodate a wide range of moving-boundary
problems, such as the ow in a reciprocating-engine combustion chamber with moving piston
and valves.
For this type of problem, an additional equation called the space conservation law is solved
for the moving coordinate velocity components. This relates the change in cell volume to
the cell-face velocity. The simultaneous satisfaction of the space conservation law and all
other equations of uid motion facilitates the general moving mesh operations performed.
Effectively this means that in addition to a velocity term caused by the uid moving through
the mesh, you get a contribution from the mesh moving through the uid.
The following shows an example of a domain showing rigid body motion. Here the domain
moves and causes the uid to slosh within it.
Figure 6: Geometry and locations of
pressure taps
Figure 7: Comparison of simulation and
SRI experiment for sway motion,
20 % full, 60 mm amplitude, 1.74 s period,
at location P1
Figure 8: Comparison of simulation and
SRI experiment for sway motion,
20 % full, 60 mm amplitude, 1.74 s period,
at location P3
Figure 9: Water entry of a
wedge at 5 inclination
99
The next example is of a light wedge (nose angle 140) in free fall from two positions enter
into water (from PhD thesis of R. Azcueta, TU Hamburg-Harburg, 2001). This is a more
complex case because the motion is not pre-determined. Obviously the wedge initially falls
with gravity (so has an acceleration of 1g) but as soon as it starts to interact with the water
it has a strong resisting force. This type of calculation needs very strong coupling and needs
to be done within the inner iterations of the CFD solver as the problem is highly non-linear
and linearisation would produce gross overshoots.
Figure 10: Comparison of vertical acceleration
in simulation and experiment
Figure 11: Comparison of angular
acceleration in simulation and experiment
Figure 12: ABAQUS model Figure 13: ABAQUS Mesh
100
S
O
L
U
T
I
O
N
S

F
R
O
M

O
U
R

S
O
F
T
W
A
R
E

P
A
R
T
N
E
R
S
COUPLE THE FLOW WITH A STRUCTURAL MODEL
There are many options for the coupling, and many methods for each option.
The coupling can be very loose or one way with data exchange happening only once the
simulation is complete
The coupling can be two way with data exchange at a number of points within the
simulation
The coupling can be fully coupled: needed for highly non-linear problems.
The following example shows a coupling of type (2) above. The wedge case shown previously
is fully coupled as (3) above but for this case the wedge is modelled as a rigid body so its
motion is dened by Newtons laws of motion.
The structural model is build in ABAQUS and is a reinforced box meshed with hexahedra
using the ABAQUS mesh tools.
The STAR-CD model is of the surrounding uid. The mesh shows where the box starts above
the water and is initially surrounded by air. The meshes do not match.
Figure 14: STAR-CD model of
the uid domain
Figure 15: Displacements within ABAQUS
due to hitting the water
Figure 16: Pressure vs. Time for a number of
points in the ow domain
101
MpCCI is used to couple the two codes, with STAR-CD sending pressure to ABAQUS and
ABAQUS sending displacements back to STAR-CD. The deformation of the box is very small
compared to the size of the near wall cell so the whole deformation is accounted for in this
cell without the need to spread the deformation further.
Results are shown for displacement using a modied material to exaggerate the deformation
and time vs. pressure history for a number of points in the ow domain which show that the
maximum loading is around 14 or 15 bars.
CONCLUSIONS AND FUTURE WORK
It has been shown that STAR-CD has extensively validated models for free surface ows
with moving meshes. This capability has been coupled to the ABAQUS structural code using
MpCCI and a coupled result has been shown.
The next step would be to look at a much tighter coupling to further address the non-linear
aspects of the coupling.
102
R&D OF NUCLEAR POWER PLANT SIMULATOR ON ATOMIC
ENERGY GRID INFRASTRUCTURE
Noriyuki Kushida, Yoshio Suzuki, Osamu Hazama, Hitoshi Matsubara, Akemi Nishida,
Fumimasa Araya, Tetsuo Aoyagi and Norihiro Kakajima Japan Atomic Energy Agency JAEA
ABSTRACT
Center for Computational Science and E-systems of Japan Atomic Energy Agency (hereafter
referred to as CCSE/JAEA) conducts the R&D of the virtual nuclear power plant simulator to
improve security and safety. To realize such simulator, we are constructing complex structural
analysis simulation program considering the inuence from uid or thermal. At the current
state, we have established the structural analysis method for the extremely large-scale
structure called assembly structural analysis (hereinafter referred to as ASA), which enables to
treat a complex structure as an assembly of parts. ASA enabled us to analyze the whole outer
shell of the High Temperature Gas Cooling Reactor named as High Temperature engineering
Test Reactor (hereinafter referred to as HTTR). Concurrently with HTTR structural analysis,
we are trying to perform structural-thermal coupling analysis of a steam generator of Fast
Breeder Reactor (hereinafter referred to as FBR).
CCSE/JAEA also conducts the R&D the information infrastructure for atomic energy research
named Atomic Energy GRID InfraStructure (hereinafter referred to as AEGIS) including some
applications works on it. One of the most important roles of GRID is to perform large-scale
computing beyond one supercomputer. However, its slow network provides extremely long
computation for usual parallel computing method. In ASA, each part could be solved almost
independently, thus ASA could well work in spite of slow network.
In this presentation, we will present the current state of our developed simulator and
international collaboration related to AEGIS with Fraunhofer SCAI, HLRS (in Germany), IRIT
and ORNL in United States.
INTRODUCTION
Japan Atomic Energy Agency, who is the only comprehensive laboratory in Japan, desired to
ensure the long-term energy security and countermeasures to environmental problems.
To respond to such expectation, we are currently constructing entire nuclear power plant
simulator in recent days. In line with such background, entire nuclear power plant simulator
must have following features:
It can simulate seismic response of quite complex structure such as nuclear power plant
It can simulate multi-physical system that is observed in nuclear power plant
It can run on world wide GRID environment, because entire nuclear power plant simulator
requires huge computational resources
C
I
V
I
L

E
N
G
I
N
E
E
R
I
N
G

103
Deformation simulation that was related to nuclear power plant was sometimes curried out
(for example, static[1] or dynamic[2] simulation of pressure vessel were presented by Shioya
and Ogino, respectively), however, the target of them were limited to a part of entire nuclear
power plant. According to the recent report, aws that appeared in nuclear power plant
are not limited in pressure vessel but also the outer part such as piping of cooling system.
Because of complex structure of nuclear power plant, no one can predict the effect of any
nugacious aws. Therefore, we are trying to understand the behavior of entire nuclear power
plant including building. Additionally, we must analyze multi-physical effect for more accurate
simulation of nuclear power plant. As a result, simulating entire nuclear power plant requires
quite large computer resource which is far beyond a current supercomputer. To provide such
huge computer resource, we are developing Atomic Energy Grid Infrastructure (hereinafter
referred to as AEGIS) as a hub of world wide GRID environment.
In the following chapter, the features of entire nuclear power plant simulator are introduced:
Assembly Structure Analysis which is the framework to analyze complex structure
Weak coupling multi-physical analysis on GRID environment
Interoperability among different GRID systems
ASSEMBLY STRUCTURE ANALYSIS
As described in previous chapter, nuclear power plant consists of considerably complex
structure and typical it has 100 thousand to 10 million parts. The nite element method
(FEM) [3] has mainly been used in structural analyses with success. However, from our past
experiences with the method in nuclear power plant simulations, seismic response simulations
of such a huge structure in its entirety will require resources of memory and disk capacity
of the order of 100 TB to 100 PB, which far exceed the hardware limitation of modern
supercomputers. In line with this context, we hold following two issues for conventional FEM-
based structural analyses:
Difculty in generating a mesh model of a huge structure in its entirety
The mesh generation process usually requires interactive operations on a PC by a human.
The preparation of models as illustrated in Figure 1 through the conventional approach
requires four to six months because of geometric complexity. With regard to hardware,
Noriyuki Kushida
Japan Atomic Energy Agency JAEA
4-49 Muramatsu,
Tokai-mura, Naka-gun,
Ibaraki 319-1184,
Japan
Phone: +81 (0) 29 - 282 - 1111
kushida.noriyuki@jaea.go.jp
www.jaea.go.jp/english/index.shtml
C
O
N
T
A
C
T
Figure 1b: Finite element mesh data Figure 1a: A CAD model
Figure 1: Mesh generation of a geometrically complex reactor model.
104
C
I
V
I
L

E
N
G
I
N
E
E
R
I
N
G

modeling an entire nuclear power plant will require tera- to peta-byte memory. Constructing
a model of an entire nuclear power plant would require too much manpower and too
many computational resources to be practicable; this is a serious bottleneck in large-scale
simulations.
Difculty in processing gigantic data sets of a huge structure in its entirety
As illustrated by the red line in Figure 2, CPU time and memory storage consumed by the
conventional approaches increase by the order of n
2
, where n is the model size. Therefore,
they would rapidly exceed the hardware limitation of current super computing resources as
the model size n increases [4,5].
To overcome these difculties, we developed a numerical simulation framework using an
approach called assembly structure analysis (ASA). The concept of this approach is
somewhat analogous to structured programming proposed by Dijkstra [6], which helps to
reduce the complexity of large programs. ASA allows:
Scalable generation of a mesh model
To overcome the first difficulty, we developed a strategy called from parts to the whole.
This strategy allows construction of the necessary finite element mesh data of an entire
nuclear power plant in a part-wise manner. Each part is bonded together using existing
mortar schemes to form an entire structure.
Hierarchical processing of a seismic response simulation
To overcome the second difculty, we developed a hierarchical simulation strategy. This
approach reduces the amount of CPU time and memory storage from n2 down to n
log n (blue line in Figure 2) because the entire simulation of a nuclear power plant can
be considered as a linear combination of component-wise simulations: a component is
considered to be an assembly of small parts.
Figure 2: Effectiveness of from parts to the
whole strategy in ASA against conventional
approach with respect to the required
resources.
105
We present our numerical simulation framework based on the ASA approach and its
application to the seismic response simulation of major components composing an actual
nuclear power plant.
MESH GLUING
The mesh generation procedure encounters bottlenecks in the ow of nite element
simulations. Conventional analysis must generate a mesh of an entire model all at once.
This makes mesh generation very difcult as the geometric complexity and size of the
model grow. We developed a strategy called from parts to the whole in ASA that resolves
the bottleneck during the mesh generation process. This was realized by enabling the
generation of mesh data of an entire nuclear power plant part by part. To enable part-by-
part mesh generation, we incorporated penalty schemes into our strategy to mesh gluing.
The nonconforming interfaces of each part are bonded together using penalty schemes. The
from parts to the whole strategy in the ASA approach offers two promising features in the
preparation of a huge mesh:
Part-wise meshing reduces the difculties in meshing a complicated global geometry
Part-wise meshing removes the limitation to the global model size
Simulation Results
In this chapter, we show the example of our ASA simulation. In this simulation, the 20-second
real data taken from the El Centro earthquake [7] was applied as input to the primary level.
The results of analysis of the major components of a nuclear power plant were accomplished
by the developed ASA framework. The 2.3 TB result data was visualized as shown in Figure 3.
Figure 3a
Figure 3: Global views of seismic responses of (a) all major components and (b) just the reactor
complex.
Figure 3b
106
C
I
V
I
L

E
N
G
I
N
E
E
R
I
N
G

WEAK COUPLING MULTI-PHYSICAL SIMULATION ON GRID
As mentioned above, nuclear power plant must be considered as multi-physical system for
accurate simulation. For example, as you can easily imagine, heat-ow-structure coupling
system exists in pressure vessel. In present work, we developed two coupling codes as rst
step using MpCCI, heat-structure and uid-structure coupling. Since these two coupling
analysis are popular, we do not explain the detail of codes, but the way to build the GRID-
enabled MpCCI is briey introduced. In present work, older version of MpCCI was used
and we must developed GRID-enable MpCCI using our GRID-enabled MPI implementation
called STAMPI. In Figure 4, the illustration of developed system is shown. In our developed
system, both uid and structure analysis code were parallelized using vender-MPI in one
computer and each codes were set for different computers. When value exchange period
was coming, MpCCI was called with STAMPI, thus GRID-enabled coupling simulation was
achieved.
GRID INTEROPERABILITY
In recent years, a number of high performance computing centers have recognized that GRID
can play a great role to provide secure and huge computing environment. GRID is originally
developed to provide barrier-free access among the institutes or universities with strong
security. However, such barrier still remains so far, because GRID middle-wares are developed
without any standard. In line with such background, we recognized that enabling the global
GRID environment is the next issue. Up to now, JAEA is working with some foreign institutes
to provide or enhance interoperability between their GRID and ours. Especially, interoperation
with UNICORE of high performance computing center Stuttgart is the rst experience for
us. In that work, we employed connection sever style to provide interoperability which
requires no modication to GRID middle-ware. Connection server was set between two GRIDs
Figure 4: Illustration of
MpCCI-STAMPI system for
uid-structure analysis
107
and translated the messages of them. The details of connection server should be referred
in reference [8]. Following this successful result, we are currently developing interoperability
with DIET at Institut de Recherche en Informatique de Toulouse.
Concluding remarks
In this presentation, our current state of entire nuclear power plant simulator was introduced.
For entire nuclear power plant simulation, we set roughly three issues:
How to simulate large scale problem
How to simulate multi physical problem
How to obtain computational resources which satisfy above two issues
To overcome these issues, our newly developed simulation method which was named ASA,
multi-physical simulation code with MpCCI and computational environment which was build
with GRID interoperability were introduced.
REFERENCES
[1] R. Shioya, G. Yagawa; 100 million DOF analysis using hierarchical domain decomposition
method, Transactions of JSCES, No. 20010024, 2001. (written in Japanese)
[2] M. Ogino, R. Shioya, H. Kawai and S. Yoshimura; Seismic Response Analysis of Nuclear
Pressure Vessel Model with ADVENTURE System on the Earth Simulator, Journal of the
Earth Simulator, Vol. 2, pp. 41 - 54, 2005
[3] O. C. Zienkiewicz, and R. L. Taylor; The Finite Element Method (Fifth edition),
Betterworth-Heinemann, 2000
[4] ADVENTURE project. http://adventure.q.t.u-tokyo.ac.jp/
[5] GeoFEM project. http://geofem.tokyo.rist.or.jp/index.html
[6] E. W. Dijkstra,; Structured programming, In J. N. Bixton and B. Randell editors, Software
Engineering Techniques. Pages 84-87, NATA Scientic Affairs Division, 1970
[7] Vibrationdata, http://www.vibrationdata.com/elcentro.htm
[8] Yoshio Suzuki, Takahiro Minami, Masayuki Tani, Norihiro Nakajima,Rainer Keller, Thomas
Beisel; Interoperability between UNICORE and ITBL, Workshop on Computational Grids
and Clusters (WCGC2006), at The 7
th
Int. Meeting on HPC for Computational Science,
Rio de Janeiro, Brazil, July 10.-13., 2006
3ZM-GRIMEX logo
108
3-CODE COUPLING IN FLOOD SIMULATION
Jrg-Volker Peetz and B. Steckel Fraunhofer SCAI; Thomas Sommer and K. Eulitz Dresden
Ground water Research Center; N. Ettrich Fraunhofer Institute for Industrial Mathematics
ITWM; M. Mller Ingenieurbro fr Grundwasser GmbH, Leipzig
C
I
V
I
L

E
N
G
I
N
E
E
R
I
N
G

of a software system for the coupled simulation of three components: surface water,
groundwater, and sewer water. The coupled system will be used in ood risk management.
The implementation of the software system is based on MpCCI. Exchanged quantities are
water heights and water uxes between the components.
The paper describes the coupled software system and the involved application codes with
emphasis on the requirements for MpCCI. Challenges include that all application codes are
interacting with each other, the codes are 1-, 2- as well as 3-dimensional, and the time-scales
in the codes are very different. Furthermore, the paper outlines the use of MpCCI, including
the development of suitable code adaptors with a coupling driver, the time step negotiation,
the coupling algorithm, and a dummy code used for tests and as substitute.
Finally, numerical results for a generic test model will be given. Simulations of the ood 2002
at Dresden are planned for 2007.
THE 3ZM-GRIMEX PROJECT, GOAL, AND MOTIVATION
The following text gives a report about the project Development of a 3-Zone Model for
Groundwater and Infrastructure Management After Extreme Flood Events in Urban Areas
or abbreviated 3ZM-GRIMEX [1]. This project has two main goals: (1) The simulation of ood
events in the urban area of the city Dresden by coupling the calculation of groundwater,
sewerage, and overland ow. (2) The development of a holistic ood risk management.
Six institutional partners are working together in this project:
DGFZ Dresdner Grundwasserforschungszentrum e.V.
TUD Technische Universitt Dresden
Fraunhofer-Institut fr Techno- und Wirtschaftsmathematik ITWM
Fraunhofer-Institut fr Algorithmen und Wissenschaftliches Rechnen SCAI
UFZ Umweltforschungszentrum Leipzig-Halle GmbH
Umweltamt Landeshauptstadt Dresden
ABSTRACT
The ood 2002 at the Elbe River showed that besides the
apparent damages by surface waters, considerable damage
was caused by groundwater and waters from the sewer
system. Fast rising groundwater levels due to inltration of
overland ow or water from the sewer system can damage
basements and can be a serious thread to the static of
houses.
The objective of the project 3ZM-GRIMEX is the development
Figure 1: Causes of damage for the ood August 2002 in Saxony (after [2])
109
The project started in March 2005 and is expected to nish in February 2008. It is funded by
the Federal Ministry of Education and Research (BMBF) within the research focus RIMAX.
The conception of the project was initiated after the catastrophic Elbe ood event in Saxony
in the year 2002.
Dr. Jrg-Volker Peetz
Fraunhofer Institute for Algorithms
and Scientic Computing SCAI
Schloss Birlinghoven
53754 Sankt Augustin
Germany
Phone: +49 (0) 2241 / 14 - 2945
joerg-volker.peetz@scai.fhg.de
www.scai.fraunhofer.de
www.mcci.de
C
O
N
T
A
C
T
It was shown that serious damages were not only caused by the surface water but also by
the groundwater and the water from the sewer system. Nearly half of the ood damages are
attributed to groundwater and sewage. While the surface water reaches normal levels within
days or weeks after the peak of the ood, the groundwater can have signicant raised levels
as long as one year after the ood event.
Figure 2 shows a cross-section of the underground through the Elbe valley in Dresden. In
the middle the river in its bed is depicted in light blue with the water-gauge of the ood in
August 2002. The light blue line indicatesthe level of the groundwater for this ood event.
The dark blue line marks the groundwater level a few weeks later.
This illustrates that during a ood event there occurs a ood wave in the groundwater
although with a slower dynamic. Likewise, the level of the groundwater rises due to
inltration of overland ow or water from the sewer system. Conversely, groundwater can
inltrate into the sewer system if the groundwater level is above the sewage level.
Therefore, an integrated ood risk management in urban areas requires the consideration
of all relevant ow processes including surface runoff, ow through sewer systems, and
groundwater ow.
Figure 2: Flood and groundwater (after [3]) Figure 3: Participating codes
Figure 4: Surface waters,
computation areas
(Map: Dresden, Ofce for
Environment)
110
C
I
V
I
L

E
N
G
I
N
E
E
R
I
N
G

INTRODUCTION TO THE MpCCI CONNECTION IN 3ZM-GRIMEX
The strategy of this project is to use established simulation programs for each of the three
domains. These programs have been approved for their particular area of application. They
will be coupled with each other with the MpCCI software [4] (see Figure 3). MpCCI manages
the communication between the individual programs, i.e., the mapping between the different
model geometries, the time synchronization, and the data exchange.
Depending on the task, two different codes will be used for each, the overland ow (RisoSurf
[5], TrimR2D [6]) and the sewerage system (Hamoka [5], Hystem-Extran [7]). Because the
groundwater code (PCGEOFIM [8]) can calculate selected regions with ner spatial resolution,
it can be used for large scale simulations, coupling with TrimR2D and Hystem-Extran, as well
as for small scale simulations, coupling with Risosurf and Hamoka.
Figure 5: Water paths Figure 6: Example of water path
111
For the urban region of Dresden the simulated area is divided into a global model and a few
local models. The global model extends parallel to the river Elbe on each side of it (the area
with the yellow background color in Figure 4) where buildings and streets are only roughly
taken into account and the sewer system is modeled without details. Here the combination
of the programs TrimR2D, Hystem-Extran, and PCGEOFIM is applied. The local models
(illustrated by colored rectangles in Figure 4) are computed with details on the surface, such
as buildings and curbs, and with more manholes in the sewerage system. In this case the
combination of the programs RisoSurf, Hamoka, and PCGEOFIM is used.
The identication and denition of the relevant water ows between the individual
components of the ow system was an important step of the project, compare Figure 5.
The ow system includes the diverse states depending on the expansion of the ood. Each
of the individual simulation domains works with its own boundary conditions and sends one
quantity to and receives one quantity in return from each of the other component.
As an impressive example for one of the water paths in the ow system, the picture in Figure
6 shows a fountain-like outow of water from a manhole in an already awash region of
Dresden in August 2002.
In order to set up the communication in the coupled computation for each bilateral
exchange the geometrical part of the simulation model on each side and the quantities to
be exchanged have to be specied. For example, for the coupling between the surface water
program and the groundwater code the geometrical model parts are given by the potentially
ooded elements (triangles or rectangles) of the surface model and the top side of the cells
nearest to the surface of the groundwater model. The quantity sent by the surface program
is the water height above ground at cell center in meters. The quantity sent in return by
the groundwater code is the water velocity (or water ux per area) measured in meters per
second. The same procedure is applied to each of the other two combinations: surface code
with sewerage program and groundwater code with sewerage simulation.
Figure 7: Coupling period
112
C
I
V
I
L

E
N
G
I
N
E
E
R
I
N
G

In order to set up a scenario simulation of the 2002 ood, the simulation periods of the
coupled codes have to be dened and synchronized. Some of the codes need a certain
pre-coupling run-time. Also, since the surface code needs much more CPU-time for a certain
simulation time interval, it will disconnect from the coupled computation soon after the peak
of the river ood wave. On the contrary, the groundwater code has to calculate a simulation
time interval of several months after the ood peak coupled with the sewerage program in
order to observe the normalization of the underground water level. Furthermore, if a second
ood occurs within a few months from the rst, as it was the case in January 2003, the
groundwater calculation has to include both ood peaks to get a realistic starting condition
for this situation. (Compare Figure 7)
INTERFACE DESCRIPTION MpCCI DRIVER API AND CODE API
In a coupled computation the coupled codes and the MpCCI coupling server processes run
independently and (quasi) in parallel on one or several computers. For this project a driver for
MpCCI was developed, which provides a simple interface as software library for the codes to
be coupled. Nevertheless, each code has to be extended for the coupling. Such an extension
is mainly a loop around the underlying simulation, which comprises the negotiation of the
coupling time step and the exchange of the coupling quantities. Furthermore, an API for the
access on the code internal data needs to be added to each code. This can be done either by
re-programming, if the source code is available, or by a wrapper technology.
The newly developed MpCCI driver offers a specialized API for accessing MpCCI for the
simulation codes. This API includes initialization of the coupling, negotiation of the coupling
time step, data exchange, as well as disconnection from the coupled computation (compare
Figure 8). For the communication with the simulation program the driver uses the code API to
be provided by the code owners. The driver also implements the algorithms for the coupling
time step negotiation.
For the development and testing of the driver and the code extensions a so called dummy
code was developed. It serves three purposes: it provides an exemplary implementation of
a code API and the computation ow in the main simulation loop; it is used as a testing
Figure 8: MpCCI-GRIMEX-driver
113
tool for the MpCCI driver and its API; it can be used as a substitute (dummy) for any of the
participating simulation programs by reading its geometrical model and the data to be sent
from les.
The code API of each coupling simulation program has to provide functions for describing the
part of the geometrical model which takes part in the coupling. It has to identify the coupling
region or part, to supply the number of nodes and elements, their identiers, the coordinates
of the nodes and the denition of the elements. The API further needs a function to check
the negotiated coupling time step and nally functions to get and set the exchange quantities
(see Figure 8).
Blockade situations need to be avoided when more than two codes are coupled with MpCCI.
For example, code A wants to communicate with code B, B is asking C, and C tries to contact
code A. This can happen since the functions for the coupling time step negotiation and the
data exchange wait for a reply from the partner code. Therefore, we introduced a rule of
way. It states that when a code wants to negotiate the coupling time step or exchange
data with two other codes at the same time the code has to do it in a prescribed order of the
coupling parts. For the same reason each data exchange has to be preceded by a coupling
time step negotiation.
Since the three coupling regions in this project have different simulation dynamics, a
customized coupling algorithm had to be invented. The programs for the surface runoff and
the sewerage system with the faster dynamics have to couple more frequently with each
other before they both couple with the groundwater. For their coupling a parallel coupling
scheme is used, while they both utilize a sequential scheme for the coupling with the
groundwater (compare Figure 9).
As already mentioned, one method to connect a simulation code to MpCCI is the wrapper
technology. For example, PCGEOFIM, the code for the groundwater, is adapted to the MpCCI
driver in thisway. PCGEOFIM is wrapped as a Python module with the help of the tool f2py [9].
Figure 9: Exemplary Computation
Flow for 3-Code Coupling
114
C
I
V
I
L

E
N
G
I
N
E
E
R
I
N
G

This allows for a fast wrapping because most work is done automatically. The main loop
of PCGEOFIM remains; it needs only small program modications. By using a callback
-function as argument, a call back into Python is possible from any place in the code.
PCGEOFIM runs in a thread, whose state is observed from the Python-wrapper. he data
exchange is based on a FORTRAN 90 structure that holds the data to be exchanged. This
principle has proven its usefulness also in other coupling projects. It makes it possible to build
an object-oriented interface for a program that has been developed for more than 35 years.
A Python interpreter was embedded into a C program by using Pythons C-API because there
exists no native Python interface for MpCCI. The C-calls are handed over to Python and, if
necessary, the change of data formats is handled in Python.
VISUALIZATION OF RESULTS FOR A GENERIC MODEL
Test and analysis of the results of the coupled simulation have their special difculties. There
may be visualization tools for each of the simulation programs but none for the integrated
results of the coupled system. Additionally, no single person knows how to operate all of
the coupled codes, let alone how to utilize the analyzing tools for each. Therefore, MpCCI
contains a visualizer for the geometry of the coupling regions and the exchanged quantities.
Unfortunately, it can not present not-exchanged results of one of the coupled codes such as
the height of the water in the manholes of the sewer system. Hence, we decided to extract
the results of interest and convert them into a format understood by a general purpose
visualization program. We used the program called paraview, which is built upon the
visualization toolkit vtk.
The results presented here are calculated for a generic test model consisting of a plane with
a swale and two hills behind and separated from it by a river. In the pictures shown here only
the part of the plane with the swale is considered. The rst picture in Figure 10 is a top view.
It contains the dry surface of the ground in shades of brown color. The parts of the surface
with water on it are shown in shades of blue color, depending on the height of the water:
Figure 10: Top view of the generic test model Figure 11: The generic test model from below
Figure 12: Generic test
scenario after 1800 s
Figure 13: Generic test
scenario after 5400 s
Figure 14: Generic test
scenario after 18000 s
115
the darker the blue, the higher is the water. Additionally, the height of the groundwater is
indicated with semi-transparent tiles according to the top of the groundwater cells in the
height of the groundwater level. The color of the tiles represents the water ux from the
sewerage system to the groundwater (q_KzuGW).
Figure 11 shows a second view on the landscape of the generic test model now from below.
This allows for a look on the sewerage system. The water-height in the pipes perpendicular to
the surface (the manholes) is indicated by the blue color. The dry part is shown in brown.
In this scenario the sewerage system is divided into two parts by a bottleneck between the
third and the fourth manhole from the left. Due to the height of the underground water
level, the water from the underground is inltrating into the sewerage system. Therefore, in
the right part of the sewerage system the water is pushed out through the manholes and
ows onto the surface.
In the Figures 12 to 14 the development in time of this test scenario is shown. The
groundwater is transferred from the underground through the sewerage system to the
surface. On the surface the water is converging at the swale and partly running away again
through the part of the sewerage system on the left.
Up to now, we have implemented a customized MpCCI driver for the coupling in 3ZM-
GRIMEX along with a dummy code as programming example and as testing replacement
116
C
I
V
I
L

E
N
G
I
N
E
E
R
I
N
G

for each of the application programs. Three of the ve codes to be coupled in this project are
already adapted to the MpCCI-GRIMEX-driver. With this coupled system rst scenarios for a
generic test model have been calculated and the results were analyzed with the help of 3D
visualization by means of adaptation to a standard tool.
After the coupled simulation system has been completed with the last two codes, tested, and
investigated elaborately with further scenarios, computations with scenarios of the area of
Dresden have to be done in order to develop a holistic ood risk management.
REFERENCES
[1] Web site of the project: http://www.hochwasser-dresden.de/3ZMGRIMEX
[2] Huber, G.; Hiller, G. u. Braune, A.: Konzepte des Hochwasserschutzes fr die Bauten des
Freistaates Sachsen im Historischen Stadtkern von Dresden. In: LH DD / DGFZ (Hrsg.)
(2003): Hochwassernachsorge Grundwasser Dresden. Tagungsband zum Statusseminar
am 8. Oktober 2003, S. 57-61.
[3] Hochwasser August 2002: Auswirkungen auf das Grundwasser Gesamtkonzeption
zur Grundwasserbeobachtung/ -berwachung fr den Groraum Dresden, 3. Etappe.
Dresdner Grundwasser Consulting GmbH, AG: StUFA Radebeul, Ref. Grundwasser,
Dresden, 01.11.2003 (unverff. Gutachten).
[4] Web site: http://www.scai.fraunhofer.de/mpcci.html
[5] Ettrich, N. et al.: Risikomanagement fr urbane Entwsserungssysteme Simulation und
Optimierung (RisUrSim) Teilprojekt 1: Oberchenwasser und Informationsaufbereitung
Abschlussbericht. Fraunhofer-Institut fr Techno- und Wirtschaftsmathematik
ITWM, 2003. Schmitt, T. G. and Thomas, M.: Risikomanagement fr urbane
Entwsserungssysteme (RisUrSim) -- Teilprojekt 2: Kanalnetz. TU Kaiserslautern, 2003.
[6] Fulford, J. M.: Computational Technique and Performance of Transient Inundation Model
for Rivers 2 Dimensional (TRIMR2D): A Depth-Averaged Two-Dimensional Flow Model.
Technischer Bericht, U.S. Geological Survey, 2003.
[7] Fuchs , L., Scheffer, C. and Verworn, H.-R.: Modellbeschreibung: Hystem-Extran 6.
Institut fr technisch wissenschaftliche Hydrologie GmbH (ITWH), Hannover, 2004.
[8] Mller, M., Sames, D., Mansel, H. (2003), PCGEOFIM A Finite Volume Model for
More?, MODFLOW and More 2003: Understanding through Modeling. Conference in
Golden, CO, USA, September 16. -- 19.09.2003.
[9] Peterson, P. (2006): F2PY: Fortran to Python interface generator. Open-Source-Software,
part of Numpy, available from http://numpy.scipy.org.
117
Figure 1: KaHMo CAD model [9]
118
PRECONDITIONS FOR A FLUID-STRUCTURE
INTERACTION OF THE HUMAN HEART USING FLUENT,
ABAQUS AND MpCCI
Sebastian Krittian, T. Schenkel , H. Oertel University of Karlsruhe
H. Schmid University of Auckland, NZ
ABSTRACT
Statistically, heart disease has been the major cause of death in Germany in the recent past.
In this context surgical treatment mostly focuses on a clinical or structural point of view.
However, in order to characterise current cardiac conditions, the ow pattern inside the heart
can no longer be neglected. To emphasise this statement, the patient specic Karlsruhe Heart
Model (KaHMo ) has been developed at the Institute for Fluid Mechanics, University of
Karlsruhe (TH), Germany. The present development stage of this numerical CFD-approach
(STAR-CD) describes the inner ventricular wall movement by segmentation of Magnetic
Resonance Images (MRI). This allows the transient ow pattern as well as the total losses
to be precisely analysed. However, the internal ow pattern is also inuenced by numerous
external biomechanical processes. In order to gain a deeper understanding of the entire
system, the prescribed geometry within the KaHMo will now be replaced by simulating
the myocardial deformation using FLUENT, ABAQUS and MpCCI. Starting with general
information about the KaHMo, this article summarises the identied requirements towards
a coupled patient specic heart simulation. Like in other biomechanical models it turns out
that performing uid structure interaction goes along with the need for stronger coupling
schemes.
INTRODUCTION
Many uid structure interaction approaches exist around the world analyzing cardiac
conditions [2,7,12,13]. However, due to the lack of structural in vivo data, they are mostly
based on either generic data or in vitro measurements of animal hearts. Consequently, a
transformation on a patient-specic human heart model cannot be undertaken.
B
I
O
M
E
D
I
C
A
L
Figure 2: Comparison of MRI ux measurements
(left) and simulation results (right) [6]
119
For this reason, a stand-alone CFD model (STAR-CD) based on time dependent movement
of the inner ventricular wall was developed based on patient-specic MRI images. The main
objective of the so called Karlsruhe Heart Model (KaHMo) is the ambition to analyse
the transient inner ventricular ow pattern. Both, healthy and diseased hearts, can be
investigated from a uid mechanics point of view and important clinical indication criteria can
be determined.
Dipl.-Ing. Sebastian Krittian
University of Karlsruhe
Kaiserstrae 12
76131 Karlsruhe
Germany
Phone: +49 (0) 721 / 608 - 2765
krittian@isl.mach.uni-karlsruhe.de
www.uni-karlsruhe.de
C
O
N
T
A
C
T
A description of the heart anatomy itself, the segmentation procedures as well as information
about geometry and numerical model are given in Oertel et al. 2006 [9]. Figure 2 shows the
latest evaluation process comparing simulation results and MRI ux measurements [6].
Observing the continuous progress of MRI techniques, future generations of KaHMo will
have to take the structural calculation into account again. Desperately needed in vivo data
will become more and more available. The new KaHMo FSI will extend the uid calculation
of KaHMo by replacing the time dependent movement with heart wall models. However, the
main focus will still be put on uid mechanics.
Figure 3:. Shaper geometry
120
FIRST SIMULATIONS
The cardiac cycle can be divided into a passive lling and an active contraction phase whereas
the focus of this work is put on the lling phase only. To judge the ability of the chosen
software packages to perform the cardiac uid structure interaction, simplied ABAQUS and
FLUENT models have been generated.
Geometry
In order to gain a symmetrical model geometry for rst testing purposes, Chase Medical
[1] kindly provided information about so-called shaper devices which are currently used for
selected surgery methods after myocardial infarctions. The special geometrical dimensions of
this body ensure the reconstructed ventricle keeps a certain maximum volume and at least an
approximate prolate shape. This information builds the starting point of the developed end-
systolic reference geometry shown in Figure 3.
Settings
Starting from the reference geometry mentioned above, both a structural and a uid stand
alone model have been developed using ABAQUS and FLUENT, respectively.
The nite element model consists of about 1000 shell elements; triangle elements have
been used to mesh the base ( xed boundary conditions) and quad elements to mesh
the symmetrical apex (coupling surface). The structural density is
s
=980 kg/m
3
. Generic
nonlinear hyperelastic material properties and damping values have been chosen so that a
stable simulation could be performed with physiological load conditions.
The number of 10.000 tetrahedron cells for the uid domain is also kept small in light of the
testing purpose of the simulation. Due to large deformation of the structural domain, the
mesh dynamics ability of Fluent is strongly needed to provide an adequate mesh quality. As
boundary condition an atrial pressure of p = 1000 Pa has to perform a volume change from
V
0
= 50 ml up to V
1
= 150 ml. The uid is treated as laminar and Newtonian with F = 1008
kg/m
3
and = 0.0055 Pas and solver settings are chosen as segregated and implicit. The
total lling time is set to T = 0.5 s whereas the coupling time step size is t = 0.005 s.
B
I
O
M
E
D
I
C
A
L
Figure 4: left: Qualitative
pressure distribution during
lling phase; right: Streamlines
during lling phase
Simulation: Krittian & Steinhilber (2006)
121
RESULTS
As far as the nite element model is concerned, the chosen generic non-linear hyperelastic
material parameters allowed to achieve the demanded end volume with the physiological
loading conditions. In order to check the stand alone uid model, a one-way coupling
was performed using the described inlet boundary conditions. Loads were not applied by
the coupling procedure but directly to the structural model. Therefore, the FE model was
responsible to deform the uid domain to the end volume conguration. Consequently,
both, the structural and the uid stand alone model proved their ability to perform either
simulations. However, the one-way coupling is not reasonable from a uid mechanical point
of view. The uid structure interaction process denitely needs to be applied to get into the
transient ow pattern.
Performing the uid structure interaction, stability issues occurred. Reasons for diverging
simulation results are mainly caused by not achieving equilibrium conditions after each time
step due to the weak coupling scheme. Like in other biomechanical uid structure interaction
processes, stability issues occur if the density of uid and structure are of the same order and
thus inertia effects come into play [8]. Within the actual test case, stable end congurations
could be achieved by either using higher structural stiffness accompanied by higher inlet
pressure or using strongly increased numerical damping. However, both congurations are
not recommendable for analysing the patient-specic ow pattern inside the human heart.
Figure 4 shows a qualitative pressure distribution during the lling phase as well as midplane
projected streamlines.
Figure 7: Schematic of left
ventricular matrix structure and
bre reinforcement
Figure 5: left: inlet (blue), outlet (red) and coupling surface of CFD model; right: structural model
Figure 6: Schematic of left ventricular matrix
structure and bre reinforcement
122
B
I
O
M
E
D
I
C
A
L
Finally, the test cases pointed out that further basic research has to be taken into account.
First of all, advanced solid mechanical modelling methods as well as more realistic constitutive
parameters have to be applied. The executable stand alone models have to be replaced by
more complete numerical approaches. Therefore, a ner mesh discretisation has to be used
for the uid domain; the FE model needs to be represented by continuum elements. Figure
5 shows the future coupling surface in Fluent with inlet (blue) and outlet (red) boundary
positions as well as the corresponding FE model.
However, the main issue seems to lie in the weak coupling scheme itself. Therefore,
possibilities for a long-term realisation of a stronger coupling for biomechanical applications
need to be taken into account. Please note that future models of KaHMo FSI have to consider
patient-specic reference geometries out of MRI data together with quasi bre directions
for the lling phase.
iso aniso
123
ADVANCED METHODS
Composite approach
The myocardial deformation within KaHMo FSI shall be considered from a macroscopic point
of view to satisfy available patient-specic information. The decision whether an orthotropic,
a transversely isotropic or even an isotropic nonlinear constitutive behaviour is chosen for a
special application depends strongly on the application itself. Although the heart muscle is
usually considered to be nonlinearly orthotropic [5], KaHMo FSI will follow a quasi transversely
isotropic modelling approach. Effectively existing dominant directions called bre, sheet
and sheet normal direction (Figure 6) are represented by an isotropic matrix and bre
assembly (Figure 7).
Schmid et al. [11] suggested, that even selected isotropic laws may build a good rst
approximation for whole heart FE models, when phenomena like wall thickening do play a
secondary role. A major advantage in choosing the composite approach is the possibility to
include so called quasi bre directions out of MRI tissue phase mapping procedures [4] in
the future. The correlation between these directions derived from acceleration measurements
and the effect of distributed muscle bres need to be further analysed within the scope of
those analyses.
W =
c
1

__

2c
2
(e
(c
2
(I
1
3)
1) +
c
3

__

2c
4
(e
(c
4
(I
2
-3)
1) +
c
5

__

2c
6

(e (c
6
(I
4
- 1)
1) (1)
Constitutive laws
One can see that the isotropic part is represented in terms of the principal invariants of the
isochoric right Cauchy-Green stretch tensor (I
1
, I
2
) whereas the anisotropy is considered in
terms of a modied invariant (I
4
) which includes the bre direction (see also Holzapfel et al.
[3]). A note able alternative approach is the usage of ABAQUS Marlow Law which is used for
existing uniaxial stress strain test data for nearly incompressible behaviour.
In order to capture the highly nonlinear hyperelastic myocardial behaviour a simple shear
tting procedure is applied to dene the needed material parameters c
1
c
6
[11] based on
a modied least square approach. However, the simple shear test data used is still based
on animal tissue tests. Please note that for an optimal tting result less restricted laws like
exponential Fung-type or polyconvex laws need to be considered [10].
CONCLUSION
First simulations for KaHMo FSI showed the ability of FLUENT, ABAQUS and MpCCI to
simulate cardiac uid structure interactions. However, the need for stronger coupling
possibilities for biomechanical applications had to be brought into context. Future work
includes the described further development of both stand-alone models towards a more
complete patient-specic description. Therefore, more information out of MRI phase mapping
)
2
)
2
)
2
procedures will be taken into account. As long as no in vivo stress strain relationships for
human hearts are available, parameter estimation will be used but converted to human tissue
requirements.
REFERENCES
[1] www.chasemedical.com
[2] Cheng, Y., Oertel, H. and Schenkel, T.: Fluid-Structure Coupled CFD Simulation of the
Left Ventricular Flow during Filling Phase. In: Annals of Biomedical Engineering 33
(2005), S. 567 776
[3] Holzapfel, G.A., Gasser, T.C. and Ogden, R.W.: A New Constitutive Framework for
Arterial Wall Mechanics and a Comparitive Study of Material Models. In: Journal of
Elasticity 61 (2000), S. 1 48
[4] Jung, B.A., Kreher, B.W., Markl, M. and Henning, J.: Visualization of Tissue Velocity Data
from Cardiac Wall Motion Measurements with Myocardial Fibre Tracking: Principles and
Implications for Cardiac Fiber Structure. In: European Journal of Cardio-thoracic Surgery
29 (2006), S. 158 164
[5] LeGrice, I.J., Smaill, B.H., Chai, L.Z., Edgar, S.G., Gavon, J.B. and Hunter, P.J.: Laminar
Structure of the Heart. In: American Journal of Physiology 269 (1995), S. H571 H582
[6] Malve, M.: Weiterentwicklung des Modells KaHMo, Institut fr Strmungslehre,
Universitt Karlsruhe (TH), Diss., 2006
[7] McQueen, D.M. and Peskin, C.S.: A Three-Dimensional Computer Model of the Human
Heart for Studying Cardiac Fluid Dynamics. In: Computer Graphics (2000), S. 56 60
[8] Nobile, F.: Numerical Approximation of Fluid-Structure Interaction Problems with
Application to Haemodynamics, Ecole Polytechnique Federale de Lausanne, Diss., 2001
[9] Oertel, H., Spiegel, K. and Donisi, S.: Modelling the Human Cardiac Fluid Mechanics
- 2nd edition. Universittsverlag Karlsruhe, 2006
[10] Schmid, H., Nash, M.P., Young, A.A. and Hunter, P.J.: Myocardial Material Parameter
Estimation - A comparitive study for simple shear. In: Journal of Biomechanical
Engineering 128, 5 (2006), S. 742 750
[11] Schmid, H.,Wang, Y.K., Ashton, J., Ehret, A.E., Krittian, S.B.S., Nash, M. and Hunter,
P.J.: Myocardial Material Parameter Estimation II A comprehensive comparison of
orthotropic constitutive equations. In: Journal of Biomechanical Engineering (at review)
(2007)
[12] Vierendeels, J.A., Riemslagh, K. and Dick, E.: Computer Simulation of Interventricular
Flow and Pressure Gradients During Diastole. In: Journal of Biomechanical Engineering
122 (2000), S. 667 674
[13] Watanabe, H., S., Sugiura, H., Kafuku and T., Hisada: Multiphysics Simulation of Left
Ventricular Filling Dynamics using Fluid-Structure Interaction Finite Element Method.
In: Biophysical Journal 87 (2004), S. 2074 2085
124
B
I
O
M
E
D
I
C
A
L
125
126
CODE ADAPTER FOR ARBITRARY PRESCRIBED MOTION AND
DEFORMATION USING FLUENT AND MpCCI
Markus Perschall, H. Spiess, T. Schenkel, H. Oertel University of Karlsruhe
ABSTRACT
The purpose of this paper is to show an approach to the general problem of handling
large motions and deformations of uid regions such as they occur in the wide eld of
biomechanics applications.
A MpCCI code adapter has been developed for the general purpose of prescribing arbitrary
geometry deformations based on time discrete MRI or CAD data. MpCCI is supposed
to transfer nodal coordinates from motion prescribing, structured surface meshes to an
unstructured surface mesh in FLUENT. Taking prot from MpCCIs neighbourhood search and
interpolation techniques the full dynamic mesh capabilities of the receiver FLUENT can be
used to generate high quality volume meshes for each stage of motion.
In future work this approach may be used as a standard method for CFD simulations of either
ow bounded by moving biological tissues or the uids engineering design process of exible
pumping chambers. Those cavities are frequently used in sac-type blood pumps and cardiac
assist devices.
Using this method one is able to prescribe any geometry motion as long as structured surface
meshes for the corresponding geometries are available. In contrast to structured volume
meshes, these surface meshes can easily be generated by meshing software packages such as
FLUENT Gambit even for complex geometric shapes and deformations.
MOTIVATION
As severe heart disease is the major cause of death in Germany different surgical techniques
are applied to treat heart failure. Ventricular hemodynamics play an important role for the
success of any surgical treatment. Therefore, both an evaluation and a prediction of the effect
of surgical options on uid mechanical and ventricular sufciency is necessary.
In cooperation with surgeons from the University Hospital of Bonn and Freiburg the Institute
for Fluid Mechanics developed KaHMo (Karlsruhe Heart Model), a patient specic CFD-Model
of the human heart. Like in almost every biomechanical kind of problem the simulation
geometry is moving as well as deforming. Geometry data is obtained from the segmentation
of patient specic MR images and is therefore available at discrete points of time only.
Similar issues arise on the topic of animal locomotion in uids, such as swimming and ying
as well as in the uids engineering design of the sac-type ventricular assist devices. The latter
case will be considered as an application example in the following section.
B
I
O
M
E
D
I
C
A
L
127
METHOD AND APPLICATION EXAMPLE
Before describing the applied method a short review of the previous treatment and model
setup relevant for the simulation of deforming zones is given. In the past the usual approach
at ISL (German: Institut fr Strmungslehre) to numerically simulate the uid ow in a
deforming region consisted of the following steps:
Generation of the geometry surfaces for discrete stages of the motion
Generation of structured volume meshes of identical topology for each of the geometries
with the same node numbering order.
Application of an approximation/ interpolation technique on each volume mesh vertex to
account for stages of motion in between the ones provided by the given geometries
Prescription of each volume node position in time
Due to the restriction of topological identity the sketched method allows limited deformations
only because large overall deformation leads to large cell deformation and thus decreasing
accuracy. To overcome these limitations FLUENT dynamic mesh tool can be used to realize
complex geometry deformation, because only the position of surface nodes has to be
provided while an unstructured volume mesh is both automatically smoothed and remeshed
during run time.
Because an identical surface mesh topology as well as an identical node numbering order on
each of the given surfaces is mandatory for the ability of motion prescription, one has to deal
with structured surface meshes and thus quadrilateral surface elements. In consequence of
that a transition layer of pyramid cells must be created in order to connect the quadrilaterals
on the surface to the tetrahedrals of the unstructured volume mesh.
However, pyramid cells are not recommendable for the description of complex geometries on
the one hand due to their skews in regions of small angles and on the other hand because
they are less exible to use in very narrow geometric regions.
In the following the further steps are illustrated by an application example. For the purpose of
ow computation inside blood pump chambers [4] three CAD surfaces of the same chamber
at different stages of motion are provided (Figure 1).
The approach is to split up the problem into two parts: One part is the description of surface
motion represented by the node position prescription of structured surface meshes and
Dipl.-Ing. Markus Perschall
University of Karlsruhe
Kaiserstrae 12
76131 Karlsruhe
Germany
Phone: +49 (0) 721 / 608 - 2591
perschall@isl.mach.uni-karlsruhe.de
www.uni-karlsruhe.de
C
O
N
T
A
C
T
B
I
O
M
E
D
I
C
A
L
Figure 1: CAD surfaces of each stage of
motion (grey) with structured surface mesh
(yellow, meshing was performed in GAMBIT)
Figure 2: Using MpCCI to
couple two subproblems
128
the other one is the CFD simulation using a fully unstructured grid. The rst subproblem is
basically treated by interpolation and approximation techniques implemented in the in-house
developed PGM-Code (Prescribed Grid Motion).
It has to be emphasized that the terms interpolation and approximation used here are related
to the calculation of positions of each single node with respect to time (node trajectories).
It turned out that approximation curve segments of node trajectories have to show 2
nd
order
parametric steady behaviour at joint points in order to avoid pressure discontinuities in time
which are associated with acceleration discontinuities of wall node movement [3].
The two sub problems are now coupled via MpCCI, i.e. structured surface meshes provided
by the PGM-Code via interpolation/approximation between the discrete motion stages are
associated with an unstructured surface mesh in FLUENT where the CFD computation is
performed.
To make communication between the PGM-Code and MpCCI possible a code adapter has
been implemented [1]. Its task is processing coupling region data of the PGM-Code as well as
performing node and element denitions via driver functions on the one hand and passing/
obtaining data to/from MpCCI via the so called Code Coupling Manager on the other hand.
The principle is sketched in Figure 2 and the serial coupling scheme applied here can be seen
in Figure 3.
Figure 3: MpCCI coupling scheme
between the PGM-Code and FLUENT
Figure 4: Geometries of each stage of motion with unstructured surface mesh
129
In the beginning of a simulation FLUENT performs an initial exchange of quantities and the
PGM-Code does an initial receive. The only quantities exchanged are node position and
physical time. When the PGM-Code receives physical time it computes the corresponding
new position from it and gives it back to FLUENT which then performs one timestep in the
uid ow computation. The new physical time is transferred to the PGM-Code and so on.
In this manner we were able to perform the CFD calculation for the shown geometry
deformations. The model was set up for laminar ow. The valves were simplied as pressure
inlet/wall and wall/pressure outlet in the inow phase and
the outow phase, respectively. Two cycles with an overall time of 2 seconds were simulated
using 2000 time steps. The three meshes in FLUENT that correspond to the stages of motion
depicted in gure 1 are shown in Figure 4 and section slices of the volume meshes are
depicted in Figure 5. A velocity vector plot during the lling phase of the ventricle can be seen
in Figure 6.
B
I
O
M
E
D
I
C
A
L
Figure 5: Side view section slices of the unstructured volume mesh smoothed and remeshed
by FLUENT
Figure 6: Front view of velocity
vector plot during the lling phase
of the DLRVAD ventricle
130
CONCLUSION AND OUTLOOK
To prescribe extreme types of motion and deformation of surfaces structured meshes have to
be used if no other surface mapping or projection technique is available.
With the shown method it is possible to perform CFD simulations in FLUENT even for extreme
prescribed geometry deformations, because the information of surface motion is transferred
via a code adapter and MpCCI from a structured mesh to a completely unstructured
tetrahedral mesh which can better handle very small angles and narrow regions.
In the example presented here the topology of the unstructured surface mesh remains the
same for each time step. In this case it is considerable to only perform the association for the
given stages of motion via MpCCI and use a GRID_MOTION udf in FLUENT to prescribe the
remaining time approximated stages.
The sketch below shows the new software framework established at ISL which shall in future
be used modularly to either perform FSI computations [2], or simulate uid ow in regions
bounded by externally prescribed deformations. Depending on further application cases
Figure 7: Software
framework consisting
of ABAQUS, FLUENT,
PGM-Code and MpCCI to
realize different modelling
approaches
131
future developments might deal with the issue of surface remeshing of the FLUENT side
coupling region as well as the involvement of moving prism cell layers in the near wall regions
in order to reduce numerical diffusion in near wall gradient computation.
ACKNOWLEDGEMENTS
Thanks go to Thomas Schmid, DLR Institute of Robotic Systems, Wesseling, Germany for
providing the CAD geometries of the presented application example of a sac-type ventricular
assist device.
REFERENCES
[1] Fraunhofer SCAI. MpCCI 3.0.6 Programmers Guide, 2007.
[2] S.B.S. Krittian, H. Schmid, T. Schenkel, and H. Oertel. Preconditions for a uid structure
interaction of the human heart using FLUENT, ABAQUS and MpCCI. Proceedings for the
8
th
MpCCI User Forum, 2007.
[3] M. Perschall. Berechnung des Leistungsumsatzes des menschlichen linken Ventrikels durch
das kahmo-herzmodell. Studienarbeit, Universitt Karlsruhe (TH), 2005.
[4] T. Schmid, W. Schiller, K. Spiegel, M. Stock, D. Liepsch, B. Laschka, G. Hirzinger, H. Oertel,
and A. Welz. Optimization of ventricle-shaped chambers for the implantable dlr assist
device. Journal of Biomechanics, Abstracts of the 5
th
World Congress of Biomechanics,
Munich, Germany, 29 July 4 August 2006, 39(1):250, 2006.
132
133
134
E
D
I
T
O
R
I
A
L

N
O
T
E
S
Fraunhofer Institute for Algorithms
and Scientic Computing SCAI
Klaus Wolf
Schloss Birlinghoven
53754 Sankt Augustin,
Germany
Phone: +49 (0) 2241 / 14 - 2557
klaus.wolf@scai.fraunhofer.de
www.scai.fraunhofer.de
www.mpcci.de
Visual concept: Markus Mritz
Layout: Bianca Backert
ISSN 1860-6296
Copyright 2007
Fraunhofer Institute SCAI
All rights reserved. No part of this publica-
tion may be reproduced, stored in a retrieval
system or transmitted in any form or by any
means: electronical, electrostatic, magnetic
tape, mechanical, photocopy, recording or
otherwise, without permission in writing
from the publishers.
Only the authors but not the Fraunhofer
Institute SCAI can be hold liable for any
mistakes made in manuscripts of the
conference proceedings.
.
CONTACT
Fraunhofer Institute SCAI
Schloss Birlinghoven
53754 Sankt Augustin
Germany
www.scai.fraunhofer.de
www.mpcci.de