J. Design Research, Vol. 5, No.

3, 2007 333

Enabling distributed multi-disciplinary design

of complex products: a knowledge based engineering
approach

G. La Rocca* and M.J.L. van Tooren

Faculty of Aerospace Engineering,
Delft University of Technology,
2629 HS, Kluyverweg 1, Delft, The Netherlands
E-mail: [Link]@[Link]
*Corresponding author

Abstract: In this paper, it is discussed how knowledge based engineering has

been exploited to develop a flexible design system, able to integrate a
heterogeneous set of distributed discipline-specific design and analysis tools
into a modular design framework. This system, called Design and Engineering
Engine (DEE), demonstrated its capability to support designers in performing
what-if studies and accelerate Multi-disciplinary Design and Optimisation
(MDO), through the automation of those lengthy and repetitive activities
typically hampering the design process. Design quality and innovation are
also supported by enabling the use of high fidelity analysis tools in the early
design phase.

Keywords: knowledge based engineering; design and engineering

engine; DEE; object oriented modelling; high-level primitives; agent-based
communication framework; aircraft design.

Reference to this paper should be made as follows: La Rocca, G. and

van Tooren, M.J.L. (2007) ‘Enabling distributed multi-disciplinary design
of complex products: a knowledge based engineering approach’, J. Design
Research, Vol. 5, No. 3, pp.333–352.

Biographical notes: Gianfranco La Rocca received his MSc in Aerospace

Engineering from the University of Pisa, Italy. He is currently finalising
his PhD work at the Faculty of Aerospace Engineering of TU Delft,
The Netherlands, where he has an appointment as Assistant Professor in the
group of Aircraft and Rotorcraft Design and Operation. He teaches aircraft
design and his main research activities are in the field of knowledge-based
engineering and methodologies to support MDO of aircraft.

Michel J.L. van Tooren obtained his PhD at Delft University on Composite
Fuselage Design in 1998. He is Head of the Department of Aerospace
Design, Integration and Operations at the Faculty of Aerospace Engineering
of TU Delft. He is a member of the Advisory Council of the Dutch National
Aerospace Laboratory (NLR), a member of the Technical Committee MDO,
the American Institute of Aeronautics and Astronautics and Core Professor
of the Delft Engineering Design Centre.

Copyright © 2007 Inderscience Enterprises Ltd.

334 G. La Rocca and M.J.L. van Tooren

1 Introduction

In their Aeronautic Blueprint and Strategic Research Agenda issues, NASA (2002)
and the Advisory Council for Aeronautics Research in Europe (ACARE, 2002),
respectively, have addressed the technological challenges for the next 20 years of
aviation. It is foreseen that in 20 years time the aeronautic systems will differ from
today’s systems at least as much as the actual systems differ from those of 1930.
The aeronautic community will have to take on such a challenge in a problematic
socio-economic scenario, where the availability of economic and intellectual resources is
shrinking upfront the increasing complexity of demanded products.
In this paper, Knowledge Based Engineering (KBE) is proposed as suitable
technology to improve designers’ productivity and free time for creativity and innovation,
at the expenses of the repetitive and non-added-value activities that largely hamper the
actual design process.
Although this paper mainly discusses the development of a KBE design tool for
aircraft design application, the proposed methodology can apply to other engineering
domains. The EC-sponsored projects PEGASUS and UPWIND aim, among other things,
at further developing and applying the approach presented in this paper, to support
distributed design of high-tech automotive components and very large scale wind
turbines, respectively.
This paper is structured as follows. In Section 2, the KBE design methodology is
introduced and its impact on the design process is discussed in perspective with the
traditional design approach. Section 3 focuses on the aircraft designers’ specific
needs and available design tools. Sections 4–6 detail how KBE technology has been
exploited to develop an advanced aircraft modelling tool, called Multi-Model Generator
(MMG). The MMG facilitates designers generating diverse aircraft configurations and
configuration variants and preparing dedicated models to feed the various analysis tools
involved in the MDO process. Section 7 introduces the concept of DEE, a modular
software system to address complex multi-disciplinary problems, exploiting engineering
skills and analysis tools of not geographically collocated (distributed) teams of
specialists. The pivotal role of the MMG within the DEE is illustrated. The European
project MOB is discussed as a relevant study case validating the proposed methodology.
Section 8 describes the prototype of an innovative agent based software framework,
developed to facilitate communication between the various DEE components and control
the overall design process. Finally, in Section 9, the proposed design approach is
discussed in relation with other design methodologies.

2 The impact of knowledge based engineering on the design process

In the design process of a complex product, like an aircraft or a car, a diverging and
a converging phase can be distinguished (La Rocca et al., 2002). During the first
(diverging) conceptual phase, many potential solutions are synthesised to comply with
the list of set requirements. The broader the amount of proposed solutions, the higher the
chance to have included the most appropriate ones. These design solutions can be
either variants of one concept, or completely different product configurations. In both
cases they need to be analysed (and possibly optimised, according to available time and
resources) in order to start the trade-off process. This initiates the converging phase of the
 Enabling distributed multi-disciplinary design of complex products 335

design process, where the best solutions are selected to access the next levels of design
refinement. During this diverging/converging process, many different models need to be
(re)generated and exchanged across a large number of discipline specialists, often not
geographically collocated, for analysis and assessment. The outputs generated by one
design/analysis tool typically need to be re-processed in order to become usable inputs
for others. The complexity of the design process appears in its actual magnitude when it
is realised that the level of detail and both number and typology of required tools and
expertises vary with the maturity phases of the design (van Tooren et al., 2003).
The traditional design approach shows inherent limitation in handling such
complexity with efficiency and effectiveness:
• The models required by the various disciplines involved in the design and analysis
process must reflect the very different views that different discipline specialists have
on the same product. The compilation of such models is not trivial (Giesing and
Barthelemy, 1998).
• Since the design/drafting and analysis processes are generally carried by different
actors with different tools, the design scenario is often populated by many
non-synchronised and non-relational models, leading to a high risk of inconsistency
(Cooper et al., 2001).
• Whenever major changes are required in the product configuration, new models
must be generated and new computations must be performed. Due to the large extent
of involved manual preprocessing, more time is used to set-up the various analysis
applications rather than for runtime execution (La Rocca and van Tooren, 2006;
Malone and Paydarfar, 2002). Besides, manual iterations can be easily affected by
human errors.
As results, designers get frustrated with the large amount of lengthy and repetitive
activities (Belie, 2002) hampering the design process, often not in balance with the
creative and fun part of design. The process inertia is such that fewer design iterations
and what-ifs scenarios than necessary can be investigated. Mature in-house knowledge
ends prevailing over unproven and risky innovative concepts. Promising ideas and
potential breakthrough designs are often discarded before any fair assessment.
Quality and innovation are the first victims. A real MDO approach is almost impractical
(Giesing and Barthelemy, 1998).
The schema of Figure 1 represents an alternative design paradigm through the
implementation of KBE. Here the product (or generative) model plays a pivotal role.
This product model can be interpreted as kind of rule base (Cooper et al., 2001) where
relevant knowledge of the design team (i.e., the engineering rules and reasoning
mechanisms to generate the product architecture and manipulate relative data
and information) has been recorded using a programming language. This is typically
an object-oriented programming language, enriched with sets of commands to drive an
integrated parametric CAD engine. Indeed, KBE technology is often addressed as a
fusion of Artificial Intelligence and Computer Aided Design; a KBE system can be
figured out as a sort of Expert System (Russel and Norvig, 1995) provided with the
advanced geometry manipulation capabilities (Chapman and Pinfold, 1999). Indeed,
Rule-based design, object-oriented modelling and parametric CAD represent the
cornerstones of KBE technology.
336 G. La Rocca and M.J.L. van Tooren

Figure 1 The knowledge based engineering design process

A KBE application basically operates as follows: first, the values of the parameters used
in the definition of the product model rules must be assigned, either by reading them
from a kind of input file (edited by the designer or automatically generated by some
software tool) or by inserting them via some kind of user interface. Then, the product
model is instantiated; the rules recorded in the rule base are systematically evaluated and
applied by the KBE system. Finally, geometry, or some other models possibly not
including any geometrical entity, are generated with little or no human interaction.
(See Figure 2 from Cooper et al., 2001). By recording in the product model all the
required rules for generating both the drafts of the product (for visualisation purpose) and
the models needed for the multi-disciplinary analysis, the bottlenecks between the
design/drafting and analysis phases can be finally resolved. The following benefits are
extremely significant:
• The expert does not have to assemble manually new analysis models when
changes occur in the product configuration. His/her level of frustration decreases
(Belie, 2002). The design process is accelerated, and the occurrence of human errors
is reduced.
• More cases can be evaluated during the time normally required by a single manual
iteration. More what-if scenarios can be investigated. More time can be dedicated
to creative design.
• The consistency of the multi-disciplinary analysis process can be guaranteed.
All the analysis tools can be fed with separate but relational models derived
from a unique definition of the product.
• Rules and best practices can be systematically applied.
 Enabling distributed multi-disciplinary design of complex products 337

Figure 2 The product (or generative) model

The next section will provide a critical analysis of the designers’ needs and the
limitations of available design tools in the field of aircraft design. These considerations
provide the justification for the design approach and the relative tools development
proposed in this paper.

3 Designers’ needs and available tools in the aircraft design process

CAD systems are by far the most common and widely used tools employed by designers
to record their design intent, already since the conceptual phase of the design. Current
generation CAD systems are mainly feature-based, which means they have a standard set
of parameterised primitives (points, lines, solid volumes, holes, chamfers etc.) that can be
adjusted and combined together to record a design instance. Actually, CAD primitives
have very poor knowledge recording capabilities and provide conceptual designers with
rather inadequate means to support the level of design abstraction that is typically
required in the early design phase. For a CAD system an aircraft wing will always be
a set of curves, surfaces and solids, never an assembly of functional elements able,
for instance, to generate lift, accommodate payload, provide control and generate thrust.
However, designers should not be hampered by the manipulation of infinite sets of lines
and points, but be provided with a limited set of High Level Primitives (HLPs), easier
to adjust and assemble. These primitives should be inherently able to fulfil specific
functionalities (e.g., generate lift and accommodate payload) and capable to record and
338 G. La Rocca and M.J.L. van Tooren

reuse chunks of relevant knowledge for supporting the analysis phase (e.g., capable to
automate the generation of models to feed structural and aerodynamic analysis tools).
Furthermore, there is the need off design tools and methodologies that are capable of
generating, earlier in the aircraft design process, more accurate knowledge about
the aircraft performances. Nowadays, in order to handle the complexity of the design
process, aircraft configurations are largely generated using semi empirical and statistics
based models. About 80% of the total life cycle cost of an aircraft turns out to be
determined on the base of low fidelity models (Staubach, 2003), whose reliability is
often arguable. Besides, semi empirical models and statistics are inherently inadequate
for novel aircraft concepts, for which there are no reference data available. Indeed,
the conditions to enable a proficient use of high fidelity analysis tools in the early
stage of the design process must be created. Actually, the automatic generation of
multi-disciplinary models, both of high and low level of fidelity, still represents an open
challenge.
In addition, there are some relevant non-technical issues (Belie, 2002) associated
with the development of distributed multi-disciplinary design system, often not properly
considered. First, discipline specialists should be given the possibility to participate in the
design process with their own trusted analysis tools (Morris et al., 2004). Second, they
should be facilitated in exploiting these tools in the most proficient way, hence limiting
as much as possible the associated preprocessing overhead. These issues call first for
design systems with a modular and reconfigurable architecture such that it should be
possible to plug in both proprietary and Commercial Of The Shelf (COTS) analysis tools.
Second, they put demands both on the format and on the level of preprocessing of the
data to be exchanged between design and analysis tools. Albeit it is possible to account
on a number of accepted standard data format, the second aspect, as already mentioned in
Section 2, is still an open issue.
Nevertheless, there are continuous attempts by industry and academia to develop
complex integrated design tools to cover the whole aircraft design cycle, from drafting
to high fidelity multi-disciplinary analysis and optimisation (Butler et al., 1998).
This ambitious approach has often led to an irresolvable level of integration
between design and analysis components. The utilisation of such complex tools within
distributed, collaborative design programs appears always very problematic. The eventual
substitution of one of the integrated analysis functionalities with an external analysis
tool becomes easily an overwhelming problem. Similarly, when only some of the
‘super integrated tool’ functionalities are needed, it might result impossible to
disintegrate its monolithic structure into separately usable bits. In the end, such design
systems often consists of large black box-like, monolithic pieces of code, very difficult
and often uneconomical to debug, maintain and develop further.

4 High Level Primitives (HLPs) to capture product similarities

Given a set of requirements, a designer would like to investigate diverse aircraft

configurations, such as a traditional airliner and a blended wing body aircraft, for
example. These two very different aircraft actually contain several elements of similarity:
both feature a number of similar architectural components designated to generate lift,
supply thrust, prove control and stability, accommodate payload, etc. In an
object-oriented modelling perspective (Rumbaugh et al., 1991), these components can be
 Enabling distributed multi-disciplinary design of complex products 339

abstracted as classes, instantiated multiple times and assembled in different

configurations to capture the specificity of the diverse products. These classes are the
HLPs already addressed in Section 3.
A commercial KBE system has been used to create a software implementation of the
HLPs concept. The provided object-oriented programming language has been exploited
to generate a basic suite of HLPs classes, namely the wing-trunk, the fuselage-trunk, the
engine-part and the connection-element, then used as main components for the advanced
aircraft modelling tool described in Section 6. The HLPs can be envisaged as empowered
LEGO® blocks that can be individually morphed, thanks to their parametric definition
and then assembled into a potentially infinite range of aircraft configurations and
configurations variants (Figure 3).

Figure 3 Generation of different aircraft configurations and variants of configurations using

the High Level Primitives (HLPs) modelling approach (for colours see online version)

Indeed, the flexibility of the HLPs, intended as the capability to represent objects
within a large range of typicality (Sully, 1993), is provided by the type and number of
parameters used in their classes definition. For example, the aerodynamic shape of any
wing-trunk HLP instantiation can be controlled through a large number of shape-specific
parameters, such as type, number and position of airfoils, chords length, sweep, twist and
dihedral angles, etc. Furthermore, designers can put together any number of wing-trunk
HLPs, each one individually modelled, to build up wings of whatever complexity
(La Rocca and van Tooren, 2006).
The definition of the HLPs is not limited to the parametric description of the external
aerodynamic surfaces, but includes the internal structure as well. Operating on a set
of structure-specific parameters, the designer can modify both position and orientation of
the internal structural elements (e.g., ribs and spars in the wing), as well as the topology
of the structure configuration (i.e., change number and type of structural elements). Rules
can be codified inside the HLP primitives in order to change automatically the number
and the type of structural elements according to the evaluation of certain conditions of
figures of merit.
340 G. La Rocca and M.J.L. van Tooren

Since the internal structure of each HLP is defined associatively to the outer surface,
when the shape of the latter changes (because, for instance, wing sections of different
shape are implemented), the former is automatically regenerated and adapted to the new
external contour (La Rocca et al., 2002). This is extremely advantageous because changes
in the aerodynamic design can be automatically cascaded down to the structural design.
The HLPs’ capability to handle, not only shape deformation but also large
topological variations, sets a significant difference between classic parametric CAD and
the KBE approach. Indeed, defining the HLP using a scripting language, rather that in
CAD fashion, i.e., by clicking and dragging features with the mouse, offers the possibility
‘to teach’ the HLPs how to manage autonomously constraints, even during the drastic
changes enforced by topological variations.

5 High Level Primitives (HLPs) to capture process similarities

Apart from the architectural similarities mentioned in the previous section, another
element of similarity links the different aircraft configurations of Figure 3. Indeed,
the analysis process of these very different aircraft configurations calls for very similar
methods and procedures. Whatever is the aircraft configuration under consideration,
specific discipline models must be generated to assess, for instance, the aerodynamic
performance or the quality of the structural design. As mentioned in Section 2, the
compilation of these dedicated models is typically very laborious and time expensive.
However, it consists mainly of repetitive activities, which are suitable to be formalised in
sets of explicit rules and algorithmic procedures. Again, the KBE programming approach
offers the possibility to record these rules as HLPs internal methods (Rumbaugh et al.,
1991). In other words, the HLPs ‘are taught’ to apply these rules systematically and
autonomously, relieving the designers from the burden of manually compiling a range of
dedicated models to feed the various analysis tools. Examples are provided in Section 6.
This HLPs capability to capture this kind of procedural knowledge (Rhem, 2006) has
been exploited also to increase the robustness of the geometry generative process, hence
to manage some known bugs and limitations of the CAD engine (Staubach, 2003)
integrated in the used KBE system. The programming approach has been used to ‘teach’
the HLPs to correct certain kind of geometry manipulation errors when occurred, for
example, due to some peculiar surfaces intersection operation; otherwise, to avoid the
occurrence of certain error prone situations and use appropriate workarounds.

6 From High Level Primitives (HLPs) to a flexible Multi-Model Generator

(MMG) to support aircraft multi-disciplinary analysis and optimisation

As anticipated in the previous section, the concept of HLPs has been implemented, using
a commercial KBE system, to develop an advanced aircraft modelling system, called the
MMG (La Rocca et al., 2002). In object-oriented parlance, the MMG can be defined as a
generic aircraft superclass (Rumbaugh et al., 1991; Booch et al., 2005). Based on a set of
parameters values provided by the designer via the so-called MMG input file, this generic
aircraft superclass is automatically instantiated into a specific aircraft configuration.
On its turn, this aircraft configuration is the result of the multiple automatic instantiations
of a number of HLPs.
 Enabling distributed multi-disciplinary design of complex products 341

The relevant feature is that the generated aircraft model inherits as a whole
the characteristics and the capability of its singles HLPs/components. For example,
because the wing-trunk HLP is able to ‘translate’ its outer surface into a specific model
for a given aerodynamic analysis tool, the whole wing, which is an assembly of
wing-trunk primitives, is able to generate from its complete surface a model for the
abovementioned analysis tool.
As represented in Figure 4, the structure of the MMG consists of two main functional
blocks:
• the product model, which contains the definition of the various HLPs and the rules
to assemble them in different aircraft configurations
• the report writers block, which contains the set of utilities to extract from the aircraft
product model the data and information necessary to generate the specific models to
feed the various analysis tools (in KBE parlance, these output models are called
reports).

Figure 4 The MMG structure (product model and report writers) and links with the analysis tools

As shown in Figure 4, the report writers provide the actual link between the MMG and
the various analysis disciplines. The user can indicate via the MMG input file the list
of output reports to be generated. As discussed in the previous sections, the capability
to generate sets of specific models to feed the various analysis tools is of paramount
importance in order to support the MDO approach (Morris et al., 2004; Giesing and
Barthelemy, 1998; La Rocca et al., 2002).
342 G. La Rocca and M.J.L. van Tooren

Specificity and consistency are driving requirements for the compilation of these
models. They actually represent the specific views the various analysis specialists have on
the aircraft under consideration. As shown in examples provided in the following two
subsections, the report writers are able to distil from the product model just those features
that are relevant to the targeted analysis tool, which can be either commercial-of-the-shelf
or in-house developed analysis tool, in any case a tool external to the MMG.
Furthermore, these generated models are actually ready to be processed by the recipient
analysis tools. The MMG takes the burden to perform the model preprocessing.
The MMG also makes sure that when a design change occurs on the aircraft
configuration, the effects of this change are automatically propagated to all the models
compiled for the various analysis tools. For example, if the designer changes the length of
the wing (in the MMG input file) such modification is automatically reflected in the
generation of the outer surface model for aerodynamic analysis, as well as in the models
for structural analysis and in the mould models used for toolings design (van der Laan
and van Tooren, 2005).
In the next two subsections, the links between the MMG and the structure and
aerodynamic disciplines are briefly discussed, as examples of the adopted approach used
to integrate the KBE modelling engine with external analysis services.

6.1 Link with structural analysis

One of the most time consuming activities in the preparation of a FE analysis consists
of the model preprocessing. In order to generate a proper structural mesh, the whole
aircraft model must be split into sets of meshable surfaces, i.e., surface elements with
adequate aspect ratio and skewedness, a maximum of four edges, each one matching
with one and only one edge of the adjacent surface elements. Knowledge acquisition
techniques (Shreiber et al., 2000; Stokes, 2001) have been used to capture into sets of
explicit rules and logical mechanisms the way FE experts would perform this surface
segmentation by hand. These rules and mechanisms have been recorded inside the HLPs
such that, every time a new model for FE analysis is required, the MMG generates
automatically a consistent set of meshable surfaces, for whatever shape and topology of
the current aircraft. One report writer is responsible to export the generated meshable
surfaces to the FE analysis tool (PATRAN/NASTRAN in this case) by means of standard
geometry data formats. Another report writer is responsible to collect (by means of the
so called FEM-table) and transfer all the non-geometric meta-data (e.g., materials and
thicknesses) generated by the MMG and required to set up the FE model (Figure 5).
Another piece of software has been written to drive PATRAN automating the last steps of
the FE model generation and finally run the solver (Nawijn et al., 2006).
Other specific report writers have been developed to link the MMG to other structural
analysis tools requiring, for example, a lumped masses and beams representation of
the aircraft (see Figure 6(a)) or a solid elements discretisation of the entire structure
to exploit advanced FE formulation based on p-elements (Figure 6(b), from Lisandrin and
van Tooren (2005)). Such a diversity of models demonstrates how different the same
product can appear to the eye of different discipline specialists, operating with different
tools.
 Enabling distributed multi-disciplinary design of complex products 343

Figure 5 Knowledge capturing from FEM and CAD experts required to develop tools
able to automate the pre-processing activities to build FEM model (for colours see
online version)

Figure 6 Two structural models generated by the MMG for different analysis tools:
(a) a masses and beams model and (b) a solid elements model

(a) (b)

6.2 Link with aerodynamic analysis

Similarly to the structural analysis case, a seamless integration has been created between
the MMG and external aerodynamic analysis tools, either COTS and proprietary codes.
The most basic form of link is based on the export of the aircraft outer surface generated
by MMG through standard geometry exchange format. Since certain aerodynamic
analysis tools (typically in-house developed analysis tools) do not support any standard
data exchange format, the HLPs have been programmed to ‘translate’ their external
surfaces into clouds of points, as shown in Figure 7(left) (La Rocca et al., 2002).
The coordinates of the points are automatically calculated by the MMG and then
formatted into files that are readable to the recipient aerodynamic tool. The MMG offers
the possibility to control the density of the cloud of points, by tuning some specific
parameters in the MMG input file. Hence, the cloud points can be used either as mesh
seeds for very refined high fidelity analysis grids (Qin et al., 2002), or as corner points
for a panel method coarse discretisation (Figure 7; Cerulli et al., 2005). In other cases,
344 G. La Rocca and M.J.L. van Tooren

they can be used as constraints for re-splining the aircraft surface into another
preprocessing environment (Laban et al., 2002).

Figure 7 Generation of consistent models for high and low fidelity aerodynamic analysis
(for colours see online version)

6.3 Operational characteristics of the Multi-Model Generator (MMG)

Some of the most relevant features of the MMG are listed below:
• The MMG can be operated interactively, through the MMG user interface.
• The MMG can be operated in batch mode (i.e., remotely and hands off). The MMG
input file can be edited by a remote user (either a designer or another software tool)
and the MMG fully controlled via web without opening any graphical interface.
• Because of the point above, many non-geographically collocated users can use the
same MMG and submit their customised version of the input file, plus their list of
required output reports.
• For a given version of the product model, each input file defines univocally one
aircraft configuration/variant. It is possible to store copies of the product model
and versions of the input file rather than large amount of output reports.
• The MMG code has a modular architecture, which facilitates maintenance and
incremental development. New modules (e.g., new HLPs or new report writers)
can be easily added to provide the MMG with extra functionalities.

7 Development of Design and Engineering Engines (DEEs)

In this section, the paradigm of an innovative, integrated modular design environment is

proposed, addressed as DEE. As represented in Figure 8, the DEE system consists
of a set of interconnected toolboxes, either COTS or proprietary software modules,
with the final goal of automating, as far as possible, MDO of aircraft configurations
 Enabling distributed multi-disciplinary design of complex products 345

(van Tooren, 2004). In an object-oriented perspective, the whole DEE system can be
interpreted as a class, where the various toolboxes represent the various methods to
perform conceptual and preliminary design of aircraft. In order to participate to the DEE
the various toolboxes must be able to operate as stand-alone modules and be accessible
via a clear input/output interface. In this way, they can be easily plugged in and out,
endowing the DEE design system with a high level of flexibility and maintainability.

Figure 8 The client/server architecture of the DEE software framework

The MMG represents the core unit of the DEE. Actually, the development of the MMG
and the DEE has progressed in parallel and the former has been developed in view
of supporting the envisioned DEE paradigm. The other DEE main components are
the Initiator, the Converger and Evaluator, the communication framework system and
the given set of discipline specific analysis tools. The Initiator is a preliminary
sizing software tool in charge to initialise the parameters contained in the MMG input file
(van Tooren et al., 2006). The Converger and Evaluator toolbox makes sure the
convergence of a given analysis process is reached and checks if the calculated
specifications of the aircraft meet the requirements; when necessary it will trigger some
design iterations. The role of the communication framework, which will be further
discussed in Section 8, is to enable the data and information exchange between the
various DEE modules of control their execution during the overall design process.
346 G. La Rocca and M.J.L. van Tooren

A successful validation of the DEE concept has been provided by the EC sponsored
project MOB – MDO of Blended Wing Body Aircraft Configurations. The purpose
of the project was setting up a distributed design system to exploit the engineering
skills of many specialists scattered across the European industry and academia, for
designing an innovative Blended Wing Body aircraft configuration (Morris et al., 2004).
In Figure 9, the position of the MMG within the MOB DEE is represented. The MMG
was responsible for the generation of the Blended Wing Body aircraft and had to supply
the various analysis tools operated by the consortium partners with sets of specific
but consistent models. For example, low and high fidelity models for aerodynamic
analysis; 2-D platform models, inclusive of movables surfaces for aeroelastic analysis;
structure models for FE analysis and optimisation; fuel tanks and systems masses
distribution for weigh and balance assessment. The MMG provided also the capability to
focus on a specific structural detail of the aircraft, a door cut-out in this case, supporting
the application of a multi-level analysis and optimisation strategy (La Rocca and
van Tooren, 2006). Once the DEE was set up, the MOB consortium was able to address
the design of an innovative complex aircraft configuration, for which no reference data or
experience existed. Figure 10 shows the result of a MOB design and optimisation study,
where more than 50 aircraft variants (including topological variations) have been
evaluated during one weekend, totally hands off, making use of high fidelity analysis
tools (including CFD and FE) running on a number of computers distributed across the
multi-national consortium. At the end of the design and optimisation process, the results
have been presented to the design team in form of response surfaces, showing the effect
of the various optimisation parameters on the aircraft performance. Without the use of the
DEE, such design study would have taken months!

Figure 9 Role of the MMG in the MOB project design engine (for colours see online version)
 Enabling distributed multi-disciplinary design of complex products 347

Figure 10 Results of the analysis and optimisation process of a blended wing body aircraft
configuration visualised by means of response surfaces (for colours see online version)

In van Tooren (2004) other examples of DEEs are discussed to show how the
presented approach resulted suitable to a number of different study cases, ranging from
the investigation of the ground effect of sport cars, to the design of fuselage panels
including piezo-elements for active sound damping (van Tooren et al., 2003). From loads
calculations on large commercial airliners (Cerulli et al., 2005), to manufacturing studies
and costs estimation of aircraft components (van der Laan and van Tooren, 2005).

8 Development of the Design and Engineering Engine (DEE)

agent-based communication framework

In this section an innovative agent-based communication framework (Berends and

van Tooren, 2006) is presented, which is supposed to enhance the flexibility and
reconfigurability of the DEE, with respect to the traditional framework systems available
on the market, including the one used in MOB (Malone and Paydarfar, 2002;
Laban et al., 2002). Figure 11 from Berends and van Tooren (2006) shows how the
DEE paradigm previously illustrated in Figure 8, is actually a formal and simplified
representation of the organisational structure of human Design Build Teams (DBT).
A DBT is a heterogeneous set of designers and discipline specialists operating their
multi-disciplinary design and computational tools, typically installed on different servers
and computers, possibly running different operating systems and belonging to separate
networks. The DBT structure, where the various human actors work and interact
in a flexible service oriented approach rather than in a rigidly predefined procedural
(throw-over-the-wall) scheme, has inspired the development of the agent-based
communication framework.
348 G. La Rocca and M.J.L. van Tooren

Figure 11 (a) DEE process flow to support MDO and (b) DEE translated into a DBT layout
using a multi-agent system to integrate the various tools (for colours see online version)

Each specialist’s tool, in order to participate in the DEE, is wrapped by an agent

(Russel and Norvig, 1995), which enable the communication with the other specialists’
tools. This interface role played by the agents is fundamental to develop a plug-and-play
design system: each specialist’s tool can be replaced by a different one (or an upgraded
version of the same tool) only if the interface is kept consistent.
A sort of peer-to-peer/server-client architecture has been adopted. One of the DEE
agent-wrapped tools is appointed server, while all the others agent-tool combinations are
the clients. Each agent-tool combination, in order to join the DEE, must register to the
server, from which receives the updated list of all the other available registered clients
(with relative contact address, such as host name and IP number and status of activity).
The role of server is played by the ‘most expert’ agent (i.e., the longest active agent
in the DEE), but in case of necessity each agent in the DEE can equally function as
server. Hence, if the formerly appointed server agent disconnects from the network, the
second eldest agents takes over. This approach, translated from the human DBT model,
guarantees the integrity of the DEE structure and the continuity of the design process, as
far as a minimum number of specialist tools is available in the DEE.
Contrarily to traditional integration framework is no need of a static, predefined
description of the process flow, as in most of the commercially available process
integration software. Each agent is able to initiate the design process when the output of
its wrapped tool is requested either by another agent or by the designer. Since each agent
is aware of the input needed to start its internal tool, it broadcasts the request for
necessary input to all the agents presently alive in the DEE. Each agent is also aware
of the output that its wrapped tool is able to generate. In case an agent is able to satisfy
the broadcasted request, it will start a peer-to-peer connection with the requesting
agent. The required input data will be generated and their storage location will be
communicated to the requesting client for retrieval.
 Enabling distributed multi-disciplinary design of complex products 349

The big advantage offered by this approach is that the design process will be able to
self-configure on the fly. When a different design problem has to be tackled, the agents
will automatically find their required order of intervention according to a demand driven
(or service oriented) approach. The DEE user does not have to reconfigure the new
process flow by hand.
Cerulli et al. (2005) describe the application of an agent based DEE to support the
redesign of a large transport aircraft vertical tail and automate some parametric studies of
the new proposed configuration. This design assignment, involving about seven different
toolboxes, running on different machines with different operative systems, takes about
two weeks of traditional semi-manual work. When performed by the DEE, it takes less
than 45 min and, during each iteration, allows the use of more accurate loads estimations
than those affordable with the traditional approach.

9 Discussion

The applications of KBE and agent-based technology discussed in this paper represent
a valuable use of Artificial Intelligence (AI) in design. AI is used here for automating the
non-creative and most tedious and repetitive parts of design process, where designers
would like to have the most support from computers. This is in contrast to other
examples of AI application in aircraft design found in literature (Rentema et al., 1998),
where case based reasoning and Expert Systems are used to select automatically what,
from historical data, results to be the best aircraft configuration for the given
requirements. By definition, this approach can never suggest the designer an innovative
aircraft configuration, as the BWB aircraft elaborated in the MOB project and does not
support the designer in the expensive multi-disciplinary analysis that follows the
configuration selection.
A fundamental characteristic of the methodology discussed here is that the engineer is
always used as role model. KBE is used to capture and reuse engineers’ best practice
and experience by means of the HLPs. The agent-based framework is used to mimic
the organisation of a DBT. Actually, agent-tools aggregation and human designers can
collaborate in the DEE. Further framework developments can be envisioned, where
software agents can directly contact a human specialist in the organisation, when no
software clients in the DEE are able to satisfy certain requests.
It is interesting to look at the MMG and DEE system definition with respect to the
Axiomatic Design (AD) approach proposed by Suh (1990). The design parameters used
to define the HLPs are actually selected to avoid or limit the coupling between the
different functional requirements of the primitives, hence complying with the first AD
axiom. For example, the definition of all the structure-specific parameters is such that
they do not affect at all the aerodynamic characteristics of the aircraft. On the contrary,
the HLPs internal structure is influenced by any change in the external aerodynamic
shape. However, in this case, this form of coupling is extremely beneficial and
deliberately pursued (the associativity mentioned in Section 4). The modular structure of
the DEE is also compliant with the minimum information axiom of AD. Apart from the
issues discussed at the end of Section 3, the integration of all design and analysis
functionalities in one monolithic application would have enormously reduced its
probability of success (Staubach, 2003).
350 G. La Rocca and M.J.L. van Tooren

Interestingly, the BWB aircraft considered in the MOB project and developed with
the support of the modular DEE (hence an AD-compliant system) is a classic example of
strongly coupled design; it is difficult to set up, requires many design iterations and any
small change in the single area of aerodynamics, propulsion or structure affects all the
others. As matter of fact, whilst the design of DEE and MMG is driven by modularity,
flexibility, maintainability and reconfigurability (where AD applies successfully), the
BWB is a purely performance driven aircraft configuration; as such not suitable for AD
(Hölttä et al., 2005).

10 Conclusions

In order to meet the technical challenges set for the next future of aviation, new design
systems are required to increase engineers’ productivity. In the current design approach
of a complex product like an aircraft, too much time is wasted in lengthy and repetitive
activities; not enough time remains available to investigate more product configurations
and better exploit designers’ skills and creativity. In this paper, KBE is proposed as a
suitable technology to help designers to reduce time and cost for engineering applications
by automating repetitive design tasks and supporting the systematic application of design
best practices.
The inherent capabilities of KBE to integrate object-oriented rule based design with
parametric CAD have been exploited in the development of the MMG, an advanced
aircraft modelling tools based on the HLPs concept. The HLPs, with their modular
parametric structure and knowledge recording capability, provide designers with the level
of flexibility and robustness needed to model and investigate a large number of aircraft
configurations and configurations variants, including novel concepts. The MMG is a
working concept and demonstrated its capability to automate the lengthy and repetitive
activities required to set up the different models for various analysis tools, both low and
high fidelity. In particular, the enabled use of advanced high fidelity analysis tools in the
early stages of the design process satisfies a fundamental need of the design community.
It both increases the level of confidence in the evaluated characteristics of the designed
product and allows studying innovative product configurations.
The EC sponsored project MOB has demonstrated the effectiveness of the MMG
tool in supporting MDO of a non-conventional aircraft configuration, in a distributed
computational environment. The concept of DEE represents a possible answer to the need
of the MDO community of open and modular design and analysis systems, able to take
advantage of specific expertises, often located outside the single company’s walls.
The application of the DEE in the design of complex product such an aircraft or
an aircraft component has shown how is possible to reduce the length of the design
process from months to days, from weeks to minutes. Consequently, more time becomes
available to perform what-if studies and analyse more configurations, such that the risk
associated with the development of a new product can be drastically reduced.
The flexible and modular nature of the DEE system (and its components) is the key to
guarantee a prompt integration of new and different analysis capabilities and facilitate the
reconfiguration/adaptation of the design process to different design cases. Furthermore,
it provides optimal conditions for system maintenance and incremental development.
 Enabling distributed multi-disciplinary design of complex products 351

Acknowledgement

Financial support to the research topics discussed in this paper has been provided
partially by the European Commission for the research project “MOB – A Computational
Design Engine Incorporating MDO for Blended Wing Body Configuration” and
by the Dutch Technology Foundation STW for the research project “Parametric
Modelling and Meshless Discretisation Methods for Knowledge Based Engineering
Applications”.

References
Advisory Council for Aeronautics Research in Europe (ACARE) (2002) Strategic Research
Agenda, Vol. 1+2 and Executive Summary, [Link]
Belie, R. (2002) ‘Non-technical barriers to multidisciplinary optimisation in the aerospace
industry’, 9th AIAA/ISSMO Symposium of Multidisciplinary Analysis and Optimisation,
Atlanta, Georgia, AIAA-2002-5439.
Berends, J.P.T.J. and van Tooren, M.J.L. (2006) ‘An agent system co-operating as a design build
team in a multidisciplinary design environment’, 44th AIAA Aerospace Science Meeting and
Exhibit, Reno, Nevada, AIAA-2002-5439.
Booch, G., Rumbaugh, J. and Jacobson, I. (2005) Unified Modeling Language User Guide, 2nd ed.,
Addison-Wesley Professional.
Butler, R., Lillico, M., Hansson, E. and van Dalen, F. (1998) ‘Comparison of MDO codes for use in
conceptual and preliminary aircraft wing design’, 7th AIAA/USAF/NASA/ISSMO Symposium
on Multidisciplinary Analysis and Optimization, St. Louis, MO, AIAA-1998-4857.
Cerulli, C., Berends, J.P.T.J., van Tooren, M.J.L. and Hofstee, J.W. (2005) ‘Parametric modeling
for structural dynamics investigations in preliminary design’, International Forum on
Aeroelasticity and Structural Dynamics, Munich, Germany.
Cerulli, C., Schut, E.J., Berends, J.P.T.J. and van Tooren, M.J.L. (2006) ‘Tail optimisation and
redesign in a multi agent task environment’, 47th AIAA/ASME/ASCE/AHS/ASC Structures,
Structural Dynamics and Materials Conference, Newport, RI, AIAA-2006-2241.
Chapman, C.B. and Pinfold, M. (1999) ‘Design engineering – a need to rethink the solution using
knowledge based engineering’, Knowledge-based Systems, Vol. 12, pp.257–267.
Cooper, S., Fan, I. and Li, G. (2001) Achieving Competitive Advantage Through Knowledge
Based Engineering, White Paper for the Department of Trade and Industry, Cranfield,
UK.
Giesing, J.P. and Barthelemy, J.M. (1998) ‘A summary of industry MDO applications and needs’,
7th AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis and Optimization,
St. Louis, MO, AIAA-1998-4737.
Hölttä, K., Suh, E.S. and de Weck, O. (2005) ‘Tradeoff between modularity and performance for
engineered systems and products’, ICED 05 International Conference on Engineering Design,
Melbourne, Australia.
La Rocca, G. and van Tooren, M.J.L. (2006) ‘Development of a modular reconfigurable software
modelling tool to support distributed multidisciplinary design and optimisation of complex
products’, 16th CIRP International Design Seminar, Kananaskis, AB, Canada.
La Rocca, G., Krakers, L. and van Tooren, M.J.L. (2002) ‘Development of an ICAD
generative model for blended wing body aircraft design’, 9th AIAA/ISSMO Symposium on
Multidisciplinary Analysis and Optimisation, Atlanta, GA, AIAA-2002-5447.
352 G. La Rocca and M.J.L. van Tooren

Laban, M., Arendsen, P., Rouwhorst, W.F.J.L. and Vankan, W.J. (2002) ‘A computational design
engine for multidisciplinary optimisation with application to a blended wing body
configuration’, 9th AIAA/ISSMO Symposium on Multidisciplinary Analysis and Optimisation,
Atlanta, GA, AIAA-2002-5446.
Lisandrin, P. and van Tooren, M.J.L. (2005) ‘High-order finite elements reduced models for use in
a flutter design tool’, Journal of Aircraft, Vol. 42, No. 3, pp.748–754.
Malone, B. and Paydarfar, S. (2002) ‘On the financial impact of MDO on the corporation’,
9th AIAA/ISSMO Symposium on Multidisciplinary Analysis and Optimisation, Atlanta, GA,
AIAA-2002-5495.
Morris, A.J., Arendsen, P., La Rocca, G., Laban, M., Voss, R. and Hönlinger, H. (2004) ‘MOB – a
European project on multidisciplinary design optimisation’, 24th ICAS Congress, Yokohama,
Japan.
NASA (2002) The NASA Aeronautic Blue Print – A Technology Vision for Aviation,
[Link]
Nawijn, M., van Tooren, M.J.L., Berends, J.P.T.J. and Arendsen, P. (2006) ‘Automated finite
element analysis in a knowledge based engineering environment’, 44th AIAA Aerospace
Science Meeting and Exhibit, Reno, Nevada.
Qin, N., Qin, N., Vavalle, A., Le Moigne, A., Laban, M., Hackett, K. and Weinerfelt, P. (2002)
‘Aerodynamic studies for blended wing body aircraft’, 9th AIAA/ISSMO Symposium on
Multidisciplinary Analysis and Optimisation, Atlanta, GA, AIAA-2002-5448.
Rentema, D.W.E., Jansen, F.W. and Torenbeek, E. (1998) ‘The application of AI and
geometric modelling techniques in conceptual aircraft design’, 7th AIAA/USAF/NASA/ISSMO
Symposium on Multidisciplinary Analysis and Optimization, St. Louis, MO, AIAA-1998-4824.
Rhem, A.J. (2006) UML for Developing Knowledge Management Systems, Auerbach Publications,
Boca Raton, FL.
Rumbaugh, J. (1991) Object-Oriented Modeling and Design, Prentice-Hall, Englewood Cliff,
New Jersey.
Russel, S. and Norvig, P. (1995) Artificial Intelligence: A Modern Approach, Prentice-Hall,
Englewood Cliff, New Jersey.
Shreiber, G., Shreiber, G., Akkermans, H., Ansewierden, A., de Hoog, R., Shadbolt, N.,
van de Velde, W. and Wielinga, B. (1999) Knowledge Engineering and Management:
The CommonKADS Methodology, MIT Press, Cambridge, MA.
Staubach, J.B. (2003) ‘Multidisciplinary design optimisation, MDO, the next frontier of CAD/CAE
in the design of aircraft propulsion systems’, AIAA/ICAS International Air and Space
Symposium and Exposition, Dayton, OH, AIAA-2003-2803.
Stokes, M. (2001) Managing Engineering Knowledge – MOKA: Methodology for Knowledge
Based Engineering Applications, Professional Engineering Publishing Limited, London and
Bury St. Edmunds, UK.
Suh, N.P. (1990) The Principles of Design, Oxford University Press, New York.
Sully, P. (1993) Modelling the World with Objects, Prentice-Hall, Englewood Cliff, New Jersey.
van der Laan, A. and van Tooren, M.J.L. (2005) ‘Parametric modeling of movables for structural
analysis’, Journal of Aircraft, Vol. 42, No. 6, pp.1605–1613.
van Tooren, M.J.L. (2004) ‘Aircraft design support using knowledge based engineering and
optimization techniques’, ICCES’04, Madeira, Portugal.
van Tooren, M.J.L., La Rocca, G., Krakers, L. and Beukers, A. (2003) ‘Design and technology in
aerospace. Parametric modeling of complex structure systems including active components’,
13th International Conference on Composite Materials, [Link], CA.
van Tooren M.J.L., Schut, E.J. and Berends, J.P.T.J. (2006) ‘Design ‘Feasilization’ using
knowledge based engineering and optimisation techniques’, 44th AIAA Aerospace Science
Meeting and Exhibit, Reno, Nevada, AIAA-2006-731.