Numerical Simulation for Early EMC Design of Cars

Flavio CANAVERO1 , Jean-Claude KEDZIA2 , Philippe RAVIER3 , Bernhard SCHOLL4

Abstract
Current innovative trends in the car industry are essentially based on more and more sophisticated
and dense electric and electronic systems, leading to still growing electronic noise levels, and increasing the
the risk to miss the EMC standards, as well as to perturb the functional integrity of the systems. This makes
more and more acute the need for numerical prediction as early as possible in the design stages, to avoid the
limited technical and economic efficiency coming from late stage management of EMC issues.
This paper describes the approach to early EMC design of cars experimented under the AutoEMC
European Project, granted by the European Community, which gathered car makers, software providers and
researchers to address this topic.
An original coupling procedure between 3D Finite Elements, Transmission Line propagation and
circuit modeling is described and validated on basic cases of fundamental interest for automotive
applications. The application of this approach to realistic full car modeling is then presented, with a short
description of the software adaption performed to reach a correct level of industrial efficiency. The question
of the sensitivity of EMC properties of the assembled system to uncertainty on local details like random
bundling is addressed, and a conclusion regarding the practical use of numerical simulation at early stages
of EMC design of cars is proposed.

Practically, the major challenge for numerical

Introduction simulation of such complex systems is to be able
to deal with very different relevant geometric
Mastering the EMC-related features of a full scales, related to the three main parts of the
car in its early design phase is becoming one of problem : the car body, the harness and the
the major technical issues for automotive equipment.
manufacturers. Even if all sub–systems were This was the starting point of the AutoEMC
developed following an EMC compliant design, European Project, which gathered car makers,
the integration may create many sources of software providers and researchers in view of
potential hazards for the overall electromagnetic addressing this topic, up to the design of effective
behavior of the complete system, hazards methodologies and related tool sets. The
detected only once the first complete prototype is approach developed under this project is
available, and whose resolution is usually, at this essentially based on the close association of three
point, difficult and expensive. complementary levels of modeling into a single
The numerical prediction of this behavior can simulation environment : full-wave 3D resolution
lead to signif icant benefits [1], by avoiding late at the car body level, transmission line
re–design and modifications of equipment propagation at the harness and bundles level, and
implementation. circuit formulation for equipment of negligible
size with respect to the wavelengths of interest.

1
Politecnico di Torino - Corso Duca degli Abruzzi, 24 - I 10129 TORINO
2
ESI S.A. - 20, rue Saarinen - SILIC 270 - F 94578 RUNGIS
3
PSI - 20 rue Saarinen - SILIC 303 - F 94588 RUNGIS CEDEX
4
BMW AG, Dep. EE-131 - Knorrstr. 147 - D 80788 MÜNCHEN
In order to predict the overall response of full distinction : usual structural CAD for car body,
systems, such a multi-level approach implies the usual electronic CAD for equipment, and (more
cooperation of corresponding specialized codes recently) specialized CAD representation of
[4] : this paper describes how the AutoEMC wiring (e.g. CATIA/E3D).
project implemented such a strategy, based on
loose coupling of specialized solvers sharing The multi-level approach retained in this
common CAD information. Project was therefore to keep the use of
The main features of this multi-level approach specialized solvers at each level (rather than
are presented in Chapter 1 for the four main trying to target a global resolution), and to
problems in car design : cross-talk, emission, loosely couple them by allowing 2-ways transfer
immunity and antenna design. Chapter 2 presents of relevant data and results between each solver
some of the fundamental validations performed pair, with the convenient reacting of this
to check the relevance of this approach by information when required.
isolating some critical articulations of the loose For all the four problems of major interest for
coupling paradigm. The application to industrial automotive EMC applications, this may be
problems is then discussed in Chapter 3, where symbolized in the following way :
this numerical strategy is applied on a realistic
vehicle with internal harness, while Chapter 4 Crosstalk Transmission
Line
Network
Simulator
adresses some specific issues related to industrial Solver

predictiveness.
Emission Transmission
Finally, the most salient conclusions of this Line
Network-
Simulator
Field
Solver
Solver
work are drawn in Chapter 5, with emphasis on Geometrical
Modelling
the difficult question of reachable numerical
Transmission
accuracy (formal predictivity) versus industrially Immunity Field
Solver
Line
Network
Simulator
Solver
relevant accuracy (product variability), and a
suggestion is made about optimal use of
numerical prediction in this area. Antenna Field
Solver

Fig. 1 - Multi-level coupled approach

1 - The Multi-level Coupling Strategy
This approach essentially follows the
theoretical and practical outlines fully described
To be useful, numerical prediction has to
in [2]. This choice leads naturally to a "loose
bring answers to the usual questions of an EMC
coupling" (essentially a data transfer, involving
engineer in his every day work. The main
direct and inverse Fourier transforms) between
questions addressed in this work are : crosstalk
the three levels of modeling because of purely
(coupling between wires inside a harness),
practical reasons : 1) time domain simulations are
emission (radiation from the harness towards the
more efficient for extended frequency ranges in
environment or an antenna), immunity (of
3D; 2)most industrially referenced transmission
equipment against external electromagnetic
line solvers work in the frequency domain; 3)all
aggression).
circuit simulators work in the time domain to
The three main data of the problem have
deal with behavioural non- linearities. For
totally different geometric characteristics : large
instance, the commercial software5 used in the
3D structures (car body and free space), long but
AutoEMC Project were :
thin - almost 1D, "1,5D" when ground plane
- for 3D fields : CEM3D (finite differences)
effects are taken into account - cables (harness),
and PAM-CEM™ (finite elements) in the
and small equipment components (essentially
time domain;
0D). Of course, this scattering of geometric
- for networks : CRIPTE and CableMod™, in
scales corresponds to different physical
the frequency domain
behaviours, usually leading to different modeling
approaches (hence, different specialized
software) : full 3D PDEs, Multiconductor 5
CEM3D is a property of ESI-Group; PAM-CEM™
Transmission Line (MTL) equations, and ODEs. is a product of PSI; CRIPTE is a property of ONERA,
It is worth noticing that the standard organization distributed by ESI-Group; CableMod™ is a product
of CAD repositories echoes this scale-based of SimLab Software GmbH; Saber™ and Saber™-
Harness are properties of Analogy Inc.
- for circuits : Saber and Saber-Harness, in the antenna, the frequencies of interest covering the
time domain. FM band, from DC up to 120MHz. For
measurement accuracy, the whole experimental
Not surprisingly, the "pivot" scale of the test set-up was located inside an anechoic
harness plays a major role in this modelization chamber.
strategy : that is why a specific emphasis was put The coupling methodology consists in
on the validation of its articulation with both the building a MTL model of the harness (including
3D field resolution and the equipement level ground planes), computing the currents generated
modelling. The next chapter sketches the along the harness path, and making these currents
fundamental validations which were performed radiate as dipoles in the 3D simulation. The
to assess the methodology at this level. coupling only requires a data format transfer and
a Fast Fourier Transform (FFT).

2 - Basic Investigations

The first methodology aspect to be validated

is the effectiveness of the loose coupling between
3D field computations and network simulations
(using MTL equation). This was performed by
numerical / experimental confrontation on the
radiation (EMR) of the following device :

Fig. 3 - EMR : 3D/MTL vs Measurements

The figure 3 above shows the experimental

results and the simulated Transfer Function for
two different modelings of the metallic
connectors : short circuit wires into CRIPTE
(simulation 1) and metallic surfaces within the
three–dimensional model (simulation 2). The
agreement is excellent between experimental
measurements and simulation results in the whole
frequency range whatever the modeling of
Fig. 2 - EMR : test set-up. connectors is.
However, this validation already exhibits an
The harness lies over a metallic plane (1.85 important feature for mastering the predictivity of
meter long, 0.95 meter large), assumed to be numerical EMC : all relevant details (here, the
perfectly conductive. From left to right, the first precise modeling of the contacts) are of
section of the harness includes two parallel wires importance.
located 5cm above the metallic table. In the
middle of the ground plane, they separate : the The second articulation to be validated stands
height of the first wire remains unchanged while between the MTL and the Circuit modeling. The
the other one rises to 10 cm. Each te rmination of two cases of emission and susceptibility led to
those wires is connected to a thin metal piece different coupling approach, and to specific
through 50Ω terminal loads. This harness radiates validations. In both cases, the objective of the
towards a 80 cm. long wire antenna (radius 3 coupling is to associate the harness (linear, with
mm.), connected to the ground plane through interference effects between wires) and the
50Ω loads. circuit (0D, but non linear) in an integrated
For this ElectroMagnetic Radiation (EMR) simulation model.
configuration, the validation criterion is the For the susceptibility analysis, the MTL code
Transfer Function between the input voltage is used to generate an "active" (i.e. including a
applied to the harness and the voltage drop representation of the sources) Z-model of the
induced on the 50Ω load at the bottom of the harness inside the car, which takes into account
all wires coupling and ground planes. This Z- built for EMC analysis. To circumvent this
model is then used in the circuit simulator as a difficulty, a specialized pre-processor (PRE-
blackbox component, which allows to represent CEM) has been built, to offer convenient CAD
the realistic EMC behaviour of the aggressed repair and cleaning tools. To take also into
harness when associated with its non linear account the rarity of available geometry
terminal loads. representations at this early stage, PRE-CEM can
In the reverse way, for the modeling of the input both CAD models in IGES format, and pre-
emission from an equipment, the MTL code is meshed models, typically in NASTRAN format.
used again to generate a lumped model of the
harness alone, still used in the circuit simulator as
a blackbox component, able to represent the
realistic signal propagation along the harness
when excited by the associated non linear
circuit 6 .

Fig. 5 : Typical early car geometry definition

(Crash Mesh)

When it comes to the harness modeling, the

MTL code needs to know the location of the
reference structures to be taken into account to
get a realistic modeling of the line parameters
(capacitance and inductance). This led to the
development of a "parser", able to extract
Fig. 4 : MTL/Circuit : Phase delay of current automatically partial cuts of the 3D structure
signal along a CAN bus acting as ground planes, at user prescribed
locations along the harness path. Finally, PRE-
3 - Realistic Applications CEM is also able to input specific parts of the
geometry files as representing wires, antennas or
Once the basic principles of the multi-level harness mean path : this information will be
strategy have been validated on academic cases, shared by the 3D and MTL solvers in coupled
the question of its practical exploitation in simulations.
realistic industrial context arises : this chapter
will essentially focus on complementary practical
aspects of the methodology, essential from the
point of view of industrial effectiveness, and
which justified the work performed under
AutoEMC to connect all required solvers into an
efficient integrated tool set.

The first step to go through is the acquisition

of the geometrical data, usually stored in CAD
repositories.
For 3D simulations, a mesh has to be
generated from the geometry of the car body :
when addressing the early stages of the car
design, the 3D geometric modeling has to face Fig. 6 : Parsing of ground planes for harness from
the problem of partial, approximate, and often 3D CAD data
topologically ill defined, geometrical
descriptions, which, however, have never been The next step is mesh generation for 3D
simulations. The project used a commercial
6
This has become a standard feature of Saber-Harness unstructured, automatic mesh generator (PAM-
GEN3D™) currently used for CFD applications. CRIPTE and PAM-CEM™ using their respective
Its adaption to EMC models was limited to the GUIs.
development of two specific features, dedicated
to the meshing of thin surfaces and of wires.
Thin surfaces : EMC models of cars have to Using PRE–CEM and all the related tools
deal with the thin metal sheets defining the car described above, the complete modeling (CAD
body; these metal sheets are physically perfectly repair & cleaning, meshing, parsing, definition of
conducting, thus avoiding the need to model physical parameters and simulation conditions)
explicitely their thickness. Therefore, for the sake of a typical car can be performed within a couple
of CPU efficiency, a specific treatment of days.
corresponding to infinitely thin surfaces was
integrated into the advancing front algorithm of As an application, we shall now briefly
the mesher. describe the simulation of the radiation of
Wires : 3D field solvers usually use specific internal harnesses of a real car towards one
local models around the wires to accurately receiving antenna located on the back hood of the
describe their EM radiation in emission vehicle. A single prototype cabling (CD audio
simulations; the numerical counterpart of these network) was considered, running through the
specific modeling of emitting wires is the whole car from the front end to the back hood,
requirement for a semi-structured mesh around with some additional parts around the front
them, approximating correctly the natural near- window (connected to the receiving FM
field symmetry of revolution around the thin antenna).
wires. In this case, the generic multi-level coupling
The discretisation is performed in two steps, procedure reads as follows :
allowing indifferent use of FD or FE resolution
algorithms : first, a surface mesh is built. In a FD
Electrical Geometr. Data
context, this surface mesh is thus intercepted by Component Structure Harness Antenna
the structured FD mesh, while in FE context,
2D-Parser
after semi-structured meshing of the emitting Equivalent
Circuit Cross Sections
wires, the mesh generator will continue to Transmission
line analyser
produce a complete tetraedral unstructured 3D Transmission Line Parameter
mesh. Network
Behaviour Models Simulator
Crosstalk and Distributed Current

3D-Field-Solver

Voltage at the + Transfer

Base of the Antenna Function

Fig. 8 : Multi-level approach for full car.

The CRIPTE analysis of the harness is

performed in a way similar to what has been
described in chapter 2 above, as is the coupling
between the equipment and the harness.
At the end of this simulation process, the
entire electromagnetic environment of the car is
Fig. 7 : Surface + wire mesh available for visualization of radiated fields
around the vehicle (as depicted in Fig. 9 below,
The last information needed to complete the using isolines display in a cutting plane), as well
computational model is the set of physical as far field radiation pattern, induced voltages
parameters. A specific direct interface between and currents on the receiving antenna (leading
CATIA/E3D and Saber-Harness has been directly to the evaluation of the Transfer
developed : however, the electric information Function), and currents along the harness. The
provided by CAD sytems is usually poorly relevance of those various outputs, which can be
convenient for direct acquisition into a simulation used for building transient movies, depends of
model, and has to be recast in a convenient way course on the type of problem under investigation
by the user. These parameters can be defined for (emission, interference, etc…).
 wiring of bundles) and at the level of system
integration (e.g. exact path of harness).
Starting from this fact, the AutoEMC project
developed also two specific actions, focusing the
issue of making, through numerical simulation,
early EMC design decisions which would not be
over-sensitive to the uncertainty occurring in the
real-life process of the product building and
integration. The first task addressed the problem
of the sensitivity of EMC behaviour of harnesses
to the dispersion of their main geometric and
physical parameters, and their statistical
handling. The second task focused on best
practice rules able to ensure that early numerical
Fig. 9 : 3D field and harness currents EMC design decisions would still hold at the
level of the complete, deliverable, car.
In this specific example, simulated and
A very simple device, made of a pair of
experimental Transfer Function were compared
parallel wires running over a ground plane to
at the level of the back-hood receiving antenna.
which they are connected, was used to
This comparison evidenced the frequent
numerically analyse the dependency of crosstalk
necessity to take into account geometric details :
and transmission properties on geometric
here, to obtain a good fitting between simulation
parameters (wires diameter, distance to ground)
and measurements, the slots located above the
and on wire permittivity. The main result is that
back hood had to be taken into account (while the
the major factor of influence is the distance to the
initially given mesh defined for crashworthiness
ground, the other effects, while not totally
did not represent them).
negligible, decreasing quickly with this distance.
For early EMC design, when the exact
Using the same device, and allowing the wires
geometry is still not totally fixed, this
to stand at a wavy distance from the ground,
requirement clearly brings some practical
numerical experiments were performed, using a
limitation to the theoretical predictivity of such
random-wire model developed by Politecnico di
numerical analysis : an approximate geometric
Torino [6]. The figure 10 below shows the results
description may be fairly satisfactory for the bulk
that can be obtained for crosstalk by different
car body, while some specific parts (slots,
methods :
shelters, …) have mandatorily to be precisely
- bold : experimental result
known to ensure the reliability of the
- dotted: uniform distance to the ground
conclusions.
(standard numerical simulation)
- light : results for 100 random realisations
- solid : one typical "random distance" result.
4 - Complementary Investigations

All the validation tests sketched in the

previous chapters demonstrated that the
predictivity of the proposed approach to
numerical EMC enjoys a fairly satisfactory level
of formal predictivity. But it is of importance to
notice that, for obtaining a precise fitting between
numerical and experimental results, the latter had
to be obtained in strictly controlled conditions.
Indeed, the industrial effectiveness of this
formal predictivity is not straightforward, since
cars, as real industrial products, are subject to a
lot of local uncertainties, both at the level of
individual components design (e.g. detailed
Fig. 10 : Random Wire Model
 The "random distance" wire is, practically, a applications, as well as elements of best practive
series of concatenated wire sections, each of rules.
these sections having a specific "distance to
ground", determined from a pseudo-random
method based on fractal algorithms. Conclusions
It appears clearly that, while a "theoretical"
numerical modeling based on uniform parameter A multi-level, loosely coupled procedure has
hypothesis gives poor experimental fitting for the been presented, validated on fundamental cases,
overall crosstalk level throughout the whole and proven to be efficiently applicable to realistic
frequency range, averaging "random wire" full car problems. This procedure can be
models brings noticeably better overall results. implemented without major difficulties on the
The fact that the crosstalk level is still not correct basis of standard relevant simulation software. In
using Random Wire modeling would need a addition, some insight is given on the effect of
deeper analysis in two directions : modeling of the natural dispersion of bundles characteristics
the wires connections, of course, but also on EMC properties, and the application of a
reproducibility of the experimental device and specific "Random Wire" model has been
related measurements. demonstrated.
This same Random Wire modeling, which
may be applied to any set of parameters attached With respect to the use of numerical
to the wires and/or bundles physics, may also be simulation for early EMC design of cars, some
used for the description of the consequences of conclusions appear clearly.
random bundling, which is the other major effect 1- For 3D field calculations, geometric modeling
on the dispersion of cross-talk, as illustrated by accuracy is a critical factor in sensitive areas,
the histogram below, computed for different where diffraction may occur. This, combined
realisations of a 9-wires bundle (using a with the need for results across wide
traditional, deterministic simulation method) : frequency ranges, promotes the choice of
Finite Elements methods in this applicative
area.
2- The formal predictivity of the numerical
simulation may be demonstrated on well
mastered situations; taking into account some
more features (welding points, geometry
defects, etc…) could still improve this
predictivity.
3- However, as regards early design, this formal
predictivity cannot be fully exploited, because
of the still inaccurate and moving definition of
EMC relevant details (harness paths, small
shielding metal sheets, …). Of course, this is
strongly related to the organization each car
maker applies in the design process, but it
Fig. 11 : Effect of random bundling on crosstalk seems to be today's state of the industrial
design art.
On the front of methodology, rational 4- In this context, it has been clearly evidenced
guidelines for effective use of numerical that numerical EMC can noticeably help EMC
simulation in early EMC design were looked for engineers to choose between different design
[5]. options, in order to avoid late and expensive
This work reviews the peculiarities of EMC design modifications, and to assess from the
problems in the car industry, the related beginning the EMC properties of a full
standardisation and legal requirements (with vehicle. As of today, the main limitation for a
special focus on the 95/54/EC directive), and the full exploitation of numerical predictivity
existing software tools. Some proposals are made seems to be the limited precision in car
regarding a set of standard basic test-cases able to making and assembly : it is still unclear
qualify the simulation codes for automotive EMC whether - and when - car industry will need to
evolve in this direction.
 [3] B. Scholl, W. Kühn, Ph. Malnoult, H. Luzet,
J.C. Kedzia, "ElectroMagnetic Interferences,
Acknowledgments Conducted and Radiated Emissions of a
All results were obtained under the Auto– Harness towards a Receiving Antenna :
EMC project (BRPR–CT97–0592) supported by Comparison of Numerical Results with
the European Community under the BRITE Measurements", Proc. of the Euro-PAM'98
EURAM program BE–4523. Conference,Tours, France, October 1998.
[4] F.G. Canavero, "The Challenge of Numerical
References EMC", Proc. of the Euro-PAM'99 Conference,
[1] L.T. Kisielewicz, J.C. Kedzia, J. Dubois, Darmstadt, Germany, October 1999.
"Benefits from Computational Electro- [5] PoliTo, "A contribution to the standardistion
magnetics in the Perspective of the European process of EMC simulation techniques",
EMC Directives", Proc. of PAM Users AutoEMC Technical Report, May 2000.
Conference in Korea HanPAM'98, Nov. 1998. [6] S. Salio, F. Canavero, J. Lefèbvre, W.
[2] L. Paletta, Ph. Dumas, J.P. Parmantier, Tabbara, "Statistical Description of Signal
"Tangential Electric Field as Excitation in Propagation on Random Bundles of Wires",
Topological Electromagnetic Problems", Proc. of the International Symposium on
Proc. of PAM Users Conf. in Asia PUCA'96, Electromagnetic Compatibility, Zurich,
Nov. 1996. Switzerland, February 1999.