Aerospace Science and Technology 9 (2005) 641651

www.elsevier.com/locate/aescte

Associative parametric CAE methods

in the aircraft pre-design
Christof Ledermann a, , Claus Hanske b , Jrg Wenzel c , Paolo Ermanni a , Roland Kelm b
a Centre for Structure Technologies, ETH Zurich, Leonhardstrasse 27, 8092 Zurich, Switzerland
b Airbus Deutschland GmbH, Mass Properties EGW, Kreetslag 10, 21129 Hamburg, Germany
c German Aerospace Center (DLR), Institute of Structural Mechanics, Lilienthalplatz 7, 38108 Braunschweig, Germany

Received 4 January 2005; received in revised form 26 April 2005; accepted 10 May 2005
Available online 15 June 2005

Abstract
Aircraft manufacturers are facing several challenges in the pre-design of aircraft structures. This early stage of the aircraft design has a
very multi-disciplinary character. Different competence centres need input data, which is at this point in time to a large extent undefined.
Therefore, a large variety of specialised tools is used in order to estimate and predict the required data. If these tools are not compatible,
interface problems are the consequence. A permanent improvement of the applied processes with regard to the informal value as well as the
applicability remains a continuous challenge.
The objective of a collaboration project between Airbus Germany GmbH, the DLR Braunschweig, and the ETH Zurich is to find new
methods and approaches to improve accuracy, efficiency, and flexibility of data prediction for primary aircraft structures. The use of modern
CAE systems together with the integration of finite element methods into the early pre-design process is a very promising approach [F. Bianconi, P. Conti, N. Senin, D.R. Wallace, CAE systems and distributed design environments, in: XII ADM International Conference, Italy,
57 September, 2001 [2]; M. Pellicciari, G. Barbanti, A.O. Andrisano, Functional requirements for a modern CAD system, in: XII ADM
International Conference, Italy, 57 September, 2001 [9]; T. Richter, H. Mechler, D. Schmitt, Integrated parametric aircraft design, in: ICAS
2002 Congress, Institute of Aeronautical Engineering, TU Munich].
The modular and knowledge-based architecture of modern CAE systems allows to represent complex assemblies like aircraft structures by parametric associative and very dynamic models. Design knowledge can be integrated into the modelling [M. Mntyl, S. Finger,
T. Tomiyama, Knowledge Intensive CAD, vol. 2, Chapman & Hall, 1997 [8]] and different characteristics or individuals of the same structure
can be mapped through parameters.
This document presents concepts, which allow to design comprehensive digital models of novel aircraft structures whereas the level of the
modelling detail shall be variegated flexibly [D.E. Whitney, R. Mantripragada, J.D. Adams, S.J. Rhee, Designing assemblies, Res. Engrg.
Design 11 (1999) 229253 [11]; P. Aspettati, S. Barone, A. Curcio, M. Picone, Parametric and feature-based assembly in motorcycle design:
from preliminary development to detail definition, in: XII ADM International Conference, Italy, 57 September, 2001].
The strongly parameterised structures allow calculating and assessing different individuals of a given structure in a very efficient and
automated way. This makes parametric associative structures very suitable for optimisation.
After structural optimisation tasks have successfully been performed with parametric models [U.M. Fasel, O. Knig, M. Wintermantel,
N. Zehnder, P. Ermanni, DynOPS an approach to parameter optimization with arbitrary simulation software, Centre of Structure Technologies, ETH Zurich; O. Knig, R. Puisa, M. Wintermantel, P. Ermanni, CAD-entity based evolutionary design optimization, Centre of Structure
Technologies, ETH Zurich, and VGTU, Faculty of Mechanics, Vilnius, Lithuania; U.M. Fasel, O. Knig, M. Wintermantel, P. Ermanni, Using evolutionary methods with a heterogeneous genotype representation for design optimization of a tubular steel trellis motorbike-frame,
Centre of Structure Technologies, ETH Zurich], multi-disciplinary optimisations are gaining importance, since they have the potential to find
global optima instead of the discipline-dependent optimal configurations and solutions.
2005 Elsevier SAS. All rights reserved.

This article was presented at the German Aerospace Congress 2004.

* Corresponding author. Tel.: +41 44 632 31 47, fax: +41 44 633 11 25.

E-mail address: [email protected] (C. Ledermann).

1270-9638/$ see front matter 2005 Elsevier SAS. All rights reserved.
doi:10.1016/j.ast.2005.05.001

642

C. Ledermann et al. / Aerospace Science and Technology 9 (2005) 641651

Keywords: Aircraft pre-design; Parametric associative CAD models; Dynamic objects; Structural optimization; Multi-disciplinary optimisation

Fig. 1. Schematic representation of the dependencies between different disciplines.

1. Introduction
The strong coupling of different disciplines and physical
parameters leads to a relevant mutual dependency between
different competence centres. This dependency will be illustrated by means of the simplified scheme in Fig. 1.
In order to perform a structural design or calculation, the
aerodynamic loads, which have to be supported by the structure, as well as some key measures like the wing span or
the length of an aircraft need to be known. For the purpose
of providing such information the aerodynamics on its part
needs to know the total weight of the aircraft and its allocation in order to calculate the required lift, wing positions, and
profiles. Obviously the mass can only be estimated when first
design ideas, structures, and key dimensions are available.
This multi-disciplinary dependencies are kept within
bounds for late development stages or for the design of
derivatives within a given aircraft family where a lot of measures are already known. However, the impact is crucial for
the pre-design stage, especially if novel aircraft are developed. Hence, the development of such novel aircraft is a
challenging task where improvements in the highly multidisciplinary pre-design phase seem possible.
This problem can only be solved using iteration processes
beyond disciplines or departments. Thus, it is very important
to have clearly defined interfaces between different disciplines that allow the exchange of reciprocatively required
data. In practice this is not always easy to establish because
different disciplines and competence centres have different
requirements and needs regarding simulation tools and data
formats.
A common parametric-associative geometry as a basis
for various simulation domains and different disciplines is
promising to overcome these difficulties in the aircraft predesign.
The geometry is supposed to be as generic as possible in order to allow all imaginable configurations. Design
knowledge can be integrated into the geometry. The parameters representing the different configurations as well as
some measures, simulation results and configuration assessments are stored in a database. Structural analysis can then
be performed by loading or importing the components of
interest, meshing them, defining the loads and boundary conditions, and solving the FE problem. The analysis can be

Fig. 2. Knowledge-based geometry as a basis for different domains.

conducted with the help of any specialised tool (Nastran,

Ansys, CATIA, . . .) supporting a geometry format that can
be exported by the common geometry system. Similarly,
CFD simulations can be performed using any specialized
CFD tool (Fluent, StarCD, CFX, in-house code. . .) that allows meshing one of the supported geometry formats. Any
other software package, scripts or in-house code can use the
geometry for calculation.
All results of the different domains are fed back to the
database which can be used to modify and optimise the
geometry. Since the different domains use the same geometry, information can even be exchanged between different
simulations. For example mesh-coupling tools like MpCCI
[5,12] allow to exchange data between very different meshes
based on the same geometry using interpolation algorithms.
Like this, multi-disciplinary optimisations can be performed.
The following sections will describe and illustrate the
concepts and architecture of such a knowledge based parametric associative geometry model.

2. Hierarchical parametric associative aircraft model

Fig. 3 shows a hierarchical parametric associative aircraft
model. This model is a generic one and was built to illustrate certain concepts. It does not raise a claim to accurately
represent any specific aircraft type or model.
The presented model was done with CATIA V5, which
is a modern CAE system with consistent object oriented
architecture. It allows parameterising geometry, associating
different components and building hierarchical assemblies.
Due to the open architecture there are a lot of possibilities to
automate processes and to integrate in-house code. Further,
CATIA V5 provides a large variety of features which allow
to integrate design knowledge into a digital mock-up. Some
of the used terminology may be system specific. However,
basically all presented concepts are of general character and
could be transferred to similar systems with equivalent functionality.

C. Ledermann et al. / Aerospace Science and Technology 9 (2005) 641651

643

Fig. 5. Variation of the cross section of a constant fuselage section.

Fig. 3. Hierarchical parametric-associative aircraft model.

Fig. 4. Different fuselage cross sections. 1) ellipse, 2) blended double bubble, and 3) spline.

more flexibility. Most flexibility is given by a spline. However, defining a cross section by splines may be inefficient,
as basically two parameters are required to define the coordinates of each point.
The above described problem can be solved in CATIA
V5 as follows. All possible cross section types are defined
in parallel and one parameter is used to define the currently
used type. A variable of type curve is used to take over
the geometrical solution of the used cross section type. This
solution works fine, allows a large degree of flexibility, but it
also means some additional work to design and parameterise
all different cross sections in the beginning.
In general it can be stated that the geometry is supposed
to be as generic as required in order to allow all imaginable configurations of interest. But the labour of defining
the generic geometry increases by the number of parameters. Furthermore, optimisations get very costly for a large
number of parameters.

2.1. Parametric CAD-models

2.2. Associative CAD models
Complex designs cannot be made in one step. It is necessary to only represent the most important data in order to
evaluate different configurations. As soon as some key data
like fuselage length or wing span are fixed, the design can be
further detailed based on the best configuration (see 2.3.2)
[1].
To have the possibility to assess different configurations,
the digital model needs to be parameterised. If one parameter is changed, the geometry needs to adapt to the changes
accordingly.
A crucial point for the efficiency of a parametric geometry is the parameterisation. Often a trade off needs to be
made between efficiency and flexibility. This will be illustrated with the help of Fig. 4.
Designing fuselage cross sections in a geometric aircraft
model, one needs to know what kind of cross sections should
be represented by the model. An easy way to design a cross
section might be to use an ellipse. Like this, three parameters
are sufficient to define the geometry one parameter for the
major and the minor axis each and a third one for the vertical
position. Due to the symmetry of the cross section it cannot be moved horizontally. If the representation of an ellipse
is sufficient, this would be the most efficient way to parameterise for calculation or optimisation. However, it might
not be sufficient if the next aircraft would have a blended
double-bubble as cross section. Parameterising a blended
double bubble will need some more parameters but it allows

Complex CAD designs like a whole aircraft cannot be designed as one part. A huge amount of data, components and
parts have to be organised in a smart way. Basically these designs can be split into parts and assemblies where assemblies
are built of several parts.
If a value changes in one part, this can have an impact to
other parts. For example, if the position of the wings should
be modified, it is not sufficient to send this information to the
electronic assembly of the wings since the fuselage section
with the wing box also needs to move, and depending on the
level of detail that is reached, it might be necessary to move
a lot of reinforcing elements.
An associative model allows interrelating different geometric objects. If the position of the wings changes, also
the wing box, the fuselage, and the fairing are automatically
adapted to the new situation.
2.3. Hierarchy and data flow
The associativity requires a clearly defined data flow. The
more associativities exist in a geometrical model, the more
vulnerable it gets to circular references. For example if a pylon should adapt its shape to the wings profile and at the
same time the locations of the wing ribs should be influenced by the position of the pylon, this will lead to a circular
reference, which cannot be solved in general. This problem

644

C. Ledermann et al. / Aerospace Science and Technology 9 (2005) 641651

Fig. 6. Schematic representation of a parametric associative CAD assembly

consisting of different components.

can be solved by hierarchical structures though. In the above

described problem a superior component could contain the
information of the wing profile and the position of the pylon. Both the wing and the pylon as two sub-components
would then retrieve the same information from one and
the same location, and both sub-components could adapt to
value changes.
Therefore, the digital mock-up is organised hierarchically, where each component can have sub-components.
A whole aircraft will exist of many components like wings,
fuselage, horizontal and vertical stabilisers, and so on. For a
first approach this might be detailed enough, but of course,
each component can be detailed further by creating subcomponents. Like this, a wing could contain ribs and a fuselage will contain frames, stringers, and windows. The same
procedure can theoretically be continued down to the rivets.
Each component is organised in a similar way according
to Fig. 7. The design specification consists of derived published parameters and internal parameters. The derived published parameters represent a link to the inherited specification parameters defined externally in the super-component.
This makes sure that all sub-components meet the basic idea
of the super-component. For example, if the whole fuselage
has a parameter to adjust the length, all fuselage sections as
sub-components need to refer to this total length. As soon
as the total length of the fuselage is changed by the parameter setting, the fuselage sections need to react. They may
react in different ways. All of the sections could keep a certain percentage of the full fuselage length, only one section
could adapt to the changes, or the number of sections could
even be altered for big changes in length. All these possibilities represent the design knowledge or intelligence of the
sub-components.
Internal parameters are used for the component-internal
use only. Of course, each component contains geometry.
This is what the CAD designers see on the screen and what
can be exported to other applications.
Although it is important to have a clear data flow topdown in a hierarchical structure in order to make sure that all
components help to fulfil the global goals, it is not sufficient.
There is also some bottom-up information flow required
in digital mock-ups. Especially measured data like weight

Fig. 7. Composition of a single component.

could be aggregated on each level in order to have the information of the total weight available at top level. This is
done by Smart Data objects explained further down. Also
checks designed to verify if all requirements within a component can be met, if limits are exceeded or if problems
occur, should send their exception information to the super
component.
Components can also contain analysis data which is the
result of finite element analysis and helps to assess different configurations. Besides optimisation cases, configuration data, and other objects, the components also contain
published parameters or published geometry, which is visible for other sub-components. Published parameters and
published geometry are objects, which are referenced only
by their respective names. Given that, they allow exchanging components with the same name definitions.
Considering a hierarchical geometric model that works as
described above and that aggregates the mass from the base
to the top, this would mean that the total mass of the aircraft
on top is only known when the last rivets are designed in
the digital mock-up. Thus, a second organisational element
is introduced the organisation in levels of detail.
As a consequence, there are now two organisational elements:
Organizational structure with components;
Levels of detail.
Both of these two elements are important and they have
a different task. They will be described in the following two
sections.

C. Ledermann et al. / Aerospace Science and Technology 9 (2005) 641651

645

Fig. 8. Organisational structure of hierarchical aircraft model with components.

Fig. 9. Organisational matrix of a hierarchical geometry model.

2.3.1. Organisational structure with components

Organising a large and complex structure or assembly
into different components and sub-components leads to a
tree structure. This is very helpful if branches of a tree are
considered. The organisational structure allows to use concurrent engineering, to split responsibilities and to organise
access rights. Most of modern CAE systems support natively
the organisational structure with components. For large and
complex systems, concurrent engineering, a split of responsibilities or the organisation of access rights is normally administered by PDM systems.
2.3.2. Levels of detail
Levels of detail organise a parametric associative geometry model across the organisational structure with components. Together with the tree structure of components and
sub-components it leads to an organisational matrix.
Each part of an assembly is assigned to a certain level
of detail. These levels can then be used for visualisation,
optimisation, estimation or calculation of any data. Fig. 10
shows the geometry of the same aircraft model at different
levels of detail.
For illustration, the first depicted level is a wireframe representation of the aircraft. The second one shows the master
geometry and the third one contains some inner geometry.
Of course, a comprehensive digital mock-up will have more
than these three levels of detail.
The important point is that all three levels represent the
complete aircraft. They contain for example a mass estima-

Fig. 10. Geometric aircraft model with different levels of detail: (top) wireframe, (middle) master geometry, (bottom) including some inner geometry.

tion based on the available geometry. The third one might be

more accurate and more detailed. However, on the first level
of detail there is also a complete estimation made based on
the data available at that point in time. This has the big advantage, that measured and aggregated data like mass is not
only available at the very end of the pre-design.
The levels of detail are more in line with a real design
process of a large and complex project. At the beginning
there is only some key information about the final product
available. Nevertheless, an overview of the whole product
must be there. The further the project develops, the more detailed the design gets, and the more accurate calculated or
estimated data will be. But at the same time the required
computational power will increase dramatically if some of
the main dimensions are changed.
Levels of detail allow to efficiently modify geometry only
taking into account the geometry down to a certain depth.
Furthermore, it allows to integrate the concept of design
freeze in digital models. After certain key measures are fixed,
a design freeze can be made on a certain level. Like this,
sub-components one level further down can still be modified and optimised while their inherited input data will stay
fixed. This is crucial for optimisations since the required
computational power for optimisations mainly depends on
the number of free variables.
Levels of detail are of big importance to managerial activities since they allow keeping an overview of the whole

646

C. Ledermann et al. / Aerospace Science and Technology 9 (2005) 641651

aircraft at any given time. In contrast to the organisational

structure with components, this functionality is not natively
integrated in all modern CAD systems. PDM systems may
take over this functionality, but they complicate automation
or flexible optimisation processes.
A combination of the two organisational concepts allows
to perform tasks like the following ones:
Visualisation of the left wing with all its details;
Aggregation the mass of the fuselage at the level of detail 5;
Minimisation the weight of a fuselage section by finding
the best values for stringer pitch, frame pitch, and shell
thickness, meeting some stiffness boundary conditions,
and taking into account the data at the level of detail 2.
This list of tasks should just give an idea and could be
continued. The third task mentioned will be described more
in detail in section [4].
2.4. Dynamic objects
Dynamic objects are a concept to improve parametricassociative geometry. The concept of dynamic objects was
implemented with the help of Visual Basic scripts.
Fig. 11 will help to outline the idea of the dynamic objects.
Currently available parameterisation features can be divided into different parameterisation levels.
Fixed models represent the lowest level of parameterisation. This means that certain geometry has an assigned
property like for example a length. This length is saved in the
geometrical object itself, and in order to modify the value of
the property, the object needs to be opened and modified. In
fact, it may be argued whether this is parameterisation at all.
Parameters are used to define a geometrical property like
a length outside the geometric object. The parameter can be
associated with one or several geometrical properties. If the
parameter gets changed, all geometrical properties referring
to this parameter will be changed. Parameters are a kind of
persistent variables in a CAE system. They can not only be

Fig. 11. Schematic representation of different levels of parameterisation.

used to be assigned to geometrical properties but also to save

any kind of information using different data types.
Formulas interrelate different geometric properties or parameters. If one parameter or property is changed, the related
one will change as well according to the predefined rule
of the formula. Formulas represent a possibility to integrate
knowledge into a geometric model.
Object Patterns, PowerCopies or UDFs (user defined features) are more sophisticated possibilities to integrate knowledge into geometric models. The user just defines once how
a certain object looks like or how it is built. Patterns allow
then to instantiate exactly the same geometry dynamically as
many times as desired in user defined patterns. PowerCopies
and UDFs allow to apply the same design rules to a different context. However, patterns as well as UDFs have certain
disadvantages. Patterns instantiate dynamically, but all instances of patterns together are just one object and cannot
be modified independently. On the other hand PowerCopies
and UDFs are context dependent, and each instantiation represents an object on its own. Unfortunately, they are not
instantiated dynamically.
Dynamic Objects are a concept that combines the advantages of both, the dynamic object patterns and the context
dependency as well as the real object instantiation of PowerCopies or UDFs, respectively. The next two sub-sections
will explain why this concept is needed and how it can be
implemented.
2.4.1. Importance of dynamic objects
In aerospace engineering a lot of repetitive structures can
be found. Frames, stringers, ribs, and windows are some
examples. There are always groups of such repetitive structures, where each single individual is designed in the same
way. However, due to the different context the individual will
look different. In the early pre-design phase all frames may
be designed in the same way and all of them look similar.
This means a lot of repetitive work for the CAD engineers.
Nevertheless, the different frames are different due to the
context, i.e. fuselage cross sections at different locations. If
patterns are used, the resulting geometry will just be copied.
As can be seen in Fig. 12 this leads to an undesired result.
UDFs are context sensitive and will not lead to this problem. Furthermore, each UDF represents an object on its own
and can be modified independently.
But if the length of a fuselage section needs to be modified or if the frame pitch has to be changed, the supernumerary frames need to be removed or missing frames have to be

Fig. 12. Fuselage section with a pattern of frames.

C. Ledermann et al. / Aerospace Science and Technology 9 (2005) 641651

Fig. 13. Implementation of dynamic objects for the variation of the number
of frames and stringers.

created, respectively. This means a lot of work and is not in

line with the concept of associativity.
To solve this problem, Dynamic Objects were suggested
and implemented to show their basic functionality. The following sub-section explains how these objects were implemented.
2.4.2. Implementation of dynamic objects
A fuselage section was created in CATIA V5, and among
a lot of other parameters there are two parameters, which
define the number of frames and stringers. As soon as
one of these parameters gets changed an event is fired. In
CATIA V5 these events can be intercepted by so-called reactions (knowledge ware objects). The short VB script attached to the reaction executes then the required sub routines and functions of a more comprehensive Visual Basic
library.
These routines adjust the positions of the existing frames
and stringers, remove the supernumerous ones, or create
new ones if needed by instantiating UDFs from external
files.
This procedure proved to work fine, and the number of
frames or stringers can just be handled as parameters, which
is very much in line with the parametric associative philosophy. The implementation of these automation routines was
even an indispensable prerequisite for the performed optimisation described in Section 3.
Unfortunately, it is time-consuming writing such procedures and if other objects should be instantiated dynamically,
some coding work would have to be done again. Furthermore, the routines run quite slow due to the uncompiled
scripting language. Because this feature is of general use and
can save a vast amount of time, it would be of great benefit to
have this functionality implemented by the software supplier
in a compiled and, thus, faster language.
2.5. Weight estimation
Hierarchical parametric-associative geometry models
need to have a clearly defined data flow, as it was described
in the previous sections. In contrast to the design process
and the main data flow, which is always from top to bottom,

647

there is some information flow needed to go the opposite

way. This holds mainly for all measurements which have an
informal character for the top levels of the design or which
have to be aggregated.
Mass estimation is one of these examples. Because the
weight is a very important factor for the efficiency of an aircraft, it is important to have a good weight estimation of the
aircraft at each point in time of the pre-design.
Since mass estimation in parts has a different character
than mass estimation in assemblies, the two cases will be
discussed separately. The following sections are referring to
the current situation in CATIA V5.
2.5.1. Mass estimation in parts
If a geometric model is designed as solid, the mass is almost calculated for free. One only needs to assign a material
to the part whose total volume is automatically calculated.
If a surface model is used, there is some minor labour
to be done. Areas as well as curve lengths can be measured
and kept as parameters. Using simple formulas to multiply
by predefined thicknesses or cross section areas are an easy
way to get to mass measurements. If the number of elements
changes, a script needs to make sure that all changes are automatically taken into account for the mass estimation.
In a demonstrator model, a fuselage section with a variable number of stringers and frames, the lengths of all these
objects are measured each time the number changes. This is
done by a Visual Basic script for automation.
The native algorithms for curve length, area, volume or
mass estimation are very efficient. Correction terms and factors for geometry not incorporated in the model can be taken
into account by knowledge-based features and scripts.
Therefore, modern CAE systems are very suitable for
directly deriving and estimating masses from parametricassociative geometry.
2.5.2. Mass estimation in assemblies
In assemblies, the weight is not directly derived from
geometry, but it needs to be aggregated in a smart way. This
section presents the concept of Smart Data elements, which
allow aggregating calculated, measured and estimated data
like mass. This concept is not implemented in the current
version of CATIA V5.
Smart data elements are a combination of parameters and
reactions. They can be defined as formula depending on any
other Smart Data. Additionally, they have the capability to
inform others.
On each level of detail of an assembly there is a Smart
Data element, which aggregates estimated and calculated
masses. As soon as one value has changed, the element informs the one of its superior assembly. Instead of updating
all masses of all assemblies at all levels of detail after one
minor part has changed its weight, only the affected parts
will be informed and updated from bottom to top. This update mode has major performance advantages for large assemblies compared to conventional update.

648

C. Ledermann et al. / Aerospace Science and Technology 9 (2005) 641651

Fig. 14. Conventional updated compared to update with Smart Data elements.

Additionally Smart Data elements are capable of estimating rather than just aggregating. They use the lower and
upper limits of parameters. Aggregating the possible upper
and lower deviations would lead to conservative results. Instead of doing so, they use the theory of errors to retrieve
reasonable estimation bounds. The ongoing implementation
of Smart Data elements is done with Visual Basic.
2.6. Costs of parametric associative models
In a first phase of a project it is much more costly creating
parametric-associative geometry models than creating conventional models, because all the design knowledge needs
to be defined and integrated. Parameters need to be set and
related to others, possible rules and checks have to be defined, and automation routines need to be written.
On the other hand, this process fosters the conceptional
thinking and forces the designer to think about which part
will be dependent on which parameters or other parts.
The further the project gets and the more complex a
design is, the more parametric-associative models pay off.
A change of basic parameters like wing span at a relatively late stage of the pre-design phase would be disastrous
with conventional design. Depending on the quality of the
parametric-associative model, such big changes could lead
to a major computational effort only.

Fig. 15. Load case with boundary conditions.

3. Optimisation of parametric models

Parametric associative geometry models allow efficient
calculation of different configurations. This is the first step
towards optimisation. This section explains how optimisation can be performed based on parametric associative
geometry models.
As an example, a fuselage section shall be optimised. The
optimisation task can be formulated as follows.
The weight of a fuselage section has to be minimised by
changing the number of frames and stringers as well as the
thickness of the shell meeting some minimum stiffness requirements regarding bending and torsion.
In order to solve this problem it will be formulated more
precisely. The objective of the optimisation is to minimise
the weight of the fuselage section. The objective function
can be written as



min f (
x)
where f (
x ) is the weight of the fuselage and x is the vector
of free parameters. The free parameters are
x1
x2
x3

number of frames
number of stringers
thickness of the fuselage shell.

Without any stiffness restrictions, the optimisation would

obviously find a minimum for zero frames, zero stringers and
a shell thickness of zero. Therefore, the following boundary
condition is defined as stiffness restrictions.



max u(
x , T , F )
umax
Compared to conventional design, parametric-associative
design requires more involvement of CAE models in the
early stages of the project like the feasibility phase and the
concept phase. This will lower the overall design costs and
development risks. It will also result in a more balanced cost
distribution during the aircraft design process [10].
Nevertheless, one needs to weigh up how comprehensive the parameterisation of a model should be. Basically,
a model should be as much parameterised as required to
map all possible configurations that might be needed. It is
not necessarily the aim to parameterise as much as possible.

where u is the total displacement at the node with maximum

displacement, T is a given torque to define the torsional stiffness, F is a given force to define the bending stiffness and
umax is a given limit for the maximum displacement.
In order to retrieve a maximum displacement u for the
given loads F and T , a complete finite element analysis
needs to be performed. This analysis was done in Ansys because automatic meshing in batch mode with shell elements
and a variable number of geometrical objects is not possible
to achieve with release R13 of CATIA V5.
The given optimisation problem deals with discrete design variables and optimizing complex geometrical models

C. Ledermann et al. / Aerospace Science and Technology 9 (2005) 641651

649

The total weight of the fuselage section is calculated in

CATIA V5 and used as input for the evolutionary optimisation algorithm.

4. Multi-disciplinary optimisation
4.1. Problem statement

Fig. 16. Schematic representation of the optimisation procedure.

Fig. 17. Displacement representation as a result of a finite element calculation in Ansys.

As stated in the introduction, optimisation of geometry

regarding one discipline does not necessarily yield the optimum for a coupled multi-disciplinary problem. There may
be different reasons why coupled multi-disciplinary optimisations are not established in todays standard development
processes, although they would be very beneficial for the aircraft pre-design phase [6].
One of the major reasons is that different disciplines have
different needs and different approaches and, as a consequence, developed their specialized tools over the past few
decades.
This section illustrates with the help of an aeroelastic
coupling example how two different disciplines can be combined without depending much on the approaches and tools
of each discipline. Furthermore it shows, how important it
is, to have one common geometry basis for different disciplines. It also outlines that automated optimisation of such a
multi-disciplinary problem is only possible with the help of
parametric associative models.
4.2. Example: aeroelastic simulation

Fig. 18. Different individuals of an optimisation.

of real-world structures often leads to non-convex problems.

These problems are difficult to solve with mathematical approaches like gradient methods. Thus, the optimisation was
performed with the help of an evolutionary algorithm using
DynOPS [3].
Fig. 16 shows a schematic representation of the whole
optimisation loop. Starting point of an optimisation is the
geometry as depicted in the left upper edge. With the help of
a Visual Basic script the CATIA V5 geometry can automatically be exported as model file. With an APDL script the
geometry will be imported into Ansys. The defined loads F
and T are applied to the structure as shown in Fig. 15. Since
the number of frames and stringers is variable, the force and
the torque need to be equally distributed to the correct nodes.
After the mesh and the loads are defined, the solution of the
FE problem can be started. It will yield a displacement distribution as shown in Fig. 17
As can be seen in Fig. 17, maximum displacements can
be found between the stringers. The maximum displacement
is used for further optimisation. If the displacements exceed
the allowable limit of umax , the optimisation routine needs
to increase the number of stringers, the number of frames or
the thickness of the shell.

A parametric sailplane wing shall be optimised taking

into account aero-elasticity. Modifying the sweep angle or
the profile of the wing will influence the aerodynamics as
well as the structural aspects. Generally, the leaner a profile is the better it is for aerodynamics due to the reduced
drag. On the other hand, very lean profiles lead to increased stresses and strains in the structure, which need to
be compensated by additional structural elements of a certain weight.
For structural analysis, an equally spread and fine rectangular mesh on the structure itself is ideal to retrieve accurate
simulation results.
If the wing shall be used for CFD simulation, the needs
are very different. The mesh in the structure is not important
at all, but the mesh around the structure is needed. The CFD
mesh should be fine close to the wing and can be coarser at a
certain distance from the wing. While the structural meshes
for surfacial structures are ideally made of shell elements,
the CFD analysis needs volume elements.
Some approaches try to define meshes which can be handled by both kinds of applications. However, this will always
be a compromise. Neither the results of the CSM simulations
nor the ones of the CFD simulations will be ideal.
Therefore the idea of coupling independent and specialised software is much more promising. A necessary pre-

650

C. Ledermann et al. / Aerospace Science and Technology 9 (2005) 641651

Fig. 19. Parametric CAD model in CATIA V5.

Fig. 21. Coupling of two independent and domain specific simulation programs.

5. Summary

Fig. 20. Coupling of two different FE meshes.

requisite to do so is that both simulations use information

based on the same geometrical model.
Meshes can be derived in domain specific software based
on the common geometry. The simulation results of one
mesh can then be mapped to a mesh of very different character using interpolation algorithms.
MpCCI (Mesh-based parallel Code Coupling Interface)
[5,12] is a coupling library that is capable of doing this interpolation called neighbourhood search. It was developed
at Fraunhofer Institute SCAI.
Aeroelastic simulations of a sailplane wing by means
of MpCCI have successfully been realised at the Centre of
Structure Technologies coupling Fluent and Ansys [6].
4.3. Aeroelastic optimisation
Due to the above mentioned reasons aeroelastic simulations are not yet very established in aircraft pre-design
processes. Aeroelastic optimisation goes even one big step
further. In order to achieve a smooth optimisation process,
the geometrical model needs to be fully parametric and associative in order to quickly modify a complete assembly.
Furthermore, all involved meshing routines and simulation
tools need to be run in batch mode.
After structural optimisation as well as aeroelastic simulations have successfully been performed with parametric
models [3,4,6,7], multi-disciplinary optimisations are gaining interest since they have the potential to find global optima instead of the discipline specific optima. Efforts will be
made to use evolutionary optimisation algorithms also for
multi-disciplinary optimisation.

This paper illustrated how beneficial parametric-associative CAE methods are in aircraft pre-design. The knowledgebased geometry can serve as basis for different domains like
structural analysis, computational fluid dynamics, and others.
It could be shown that it is possible to build hierarchical associative structures with a well defined data flow in
order to avoid circular references. Besides the native organisational structure with components it is important to have
levels of detail. The design process runs top-down the hierarchical structure and reveals more and more detailed geometry. While the most of the data flows top-down the structural
tree to make sure that requirements at top level are met at
any level of detail, some information also needs to inform
the hierarchically superior elements. This holds mainly for
measurements or conflicts.
The concept of dynamic objects was presented, and the
feasibility of such objects is shown by implementation of a
parametric associative fuselage section with a variable number of frames and stringers. The main weakness of this script
based implementation is the required calculation time because the scripting language based on Visual Basic is quite
slow and cannot be compiled. The same concept could be
implemented much more efficient by the software supplier
itself or by using faster and compiled code.
Parametric-associative geometry is very suitable for deriving masses from it. For solid models this is practically
obtained for free while calculating the mass of surface models can be obtained by minor efforts. Correction terms and
factors for geometry not incorporated in the model can be
taken into account by knowledge-based features and scripts.
It could be shown that parametric associative geometries
are very suitable for optimisation. For example structural optimisations are very important in aircraft design since they
allow finding configurations of minimum weight meeting
all boundary conditions like stiffness requirement or design
space restrictions. The example with the fuselage section
showed how such an optimisation can be implemented.

C. Ledermann et al. / Aerospace Science and Technology 9 (2005) 641651

6. Outlook
After different concepts were successfully presented as
demonstrators, they will be implemented more detailed in
a hierarchical geometric-associative geometry model of a
generic aircraft for the pre-design.
Collaboration with software suppliers is essential to make
sure that some of the concepts can be realised more efficiently.
The concept of Smart Data elements is promising to have
major performance advantages in the mass estimation of
large and complex assemblies. The implementation of smart
elements is a further task for the future.
After structural optimisation tasks have successfully been
performed with parametric models, multi-disciplinary optimisations are getting interesting as well for parametric associative models, because they have the potential to find global
optima instead of the discipline specific optima.

References
[1] P. Aspettati, S. Barone, A. Curcio, M. Picone, Parametric and featurebased assembly in motorcycle design: from preliminary development
to detail definition, in: XII ADM International Conference, Italy, 57
September, 2001.
[2] F. Bianconi, P. Conti, N. Senin, D.R. Wallace, CAE systems and distributed design environments, in: XII ADM International Conference,
Italy, 57 September, 2001.

651

[3] U.M. Fasel, O. Knig, M. Wintermantel, N. Zehnder, P. Ermanni,

DynOPS an approach to parameter optimization with arbitrary simulation software, Centre of Structure Technologies, ETH Zurich.
[4] U.M. Fasel, O. Knig, M. Wintermantel, P. Ermanni, Using evolutionary methods with a heterogeneous genotype representation for design
optimization of a tubular steel trellis motorbike-frame, Centre of Structure Technologies, ETH Zurich.
[5] M. Hackenberg, P. Post, R. Redler, B. Steckel, MpCCI, Multidisciplinary applications and multigrid, in: European Congress on Computational Methods in Applied Sciences and Engineering, Proceedings
ECCOMAS 2000, Barcelona, 1114 September, 2000.
[6] F. Hrlimann, Multi-disciplinary optimisation of a sailplane wing,
semester thesis IMES-ST 04-114, Institute of Mechanical Systems,
Centre of Structure Technologies, ETH Zurich, 2004.
[7] O. Knig, R. Puisa, M. Wintermantel, P. Ermanni, CAD-entity based
evolutionary design optimization, Centre of Structure Technologies,
ETH Zurich, and VGTU, Faculty of Mechanics, Vilnius, Lithuania.
[8] M. Mntyl, S. Finger, T. Tomiyama, Knowledge Intensive CAD,
vol. 2, Chapman & Hall, 1997.
[9] M. Pellicciari, G. Barbanti, A.O. Andrisano, Functional requirements
for a modern CAD system, in: XII ADM International Conference,
Italy, 57 September, 2001.
[10] T. Richter, H. Mechler, D. Schmitt, Integrated parametric aircraft design, in: ICAS 2002 Congress, Institute of Aeronautical Engineering,
TU Munich.
[11] D.E. Whitney, R. Mantripragada, J.D. Adams, S.J. Rhee, Designing
assemblies, Res. Engrg. Design 11 (1999) 229253.
[12] K. Wolf, B. Steckel, Mesh-based parallel code coupling interface, Fraunhofer Institut SCAI, Institut fr Algorithmen und Wissenschaftliches Rechnen, GMD Forschungszentrum Informationstechnik GmbH, ISSN 1435-2702.