Semantic Knowledge-Based-Engineering: The Codex Framework
J. Zamboni
a
, A. Zamfir
b
and E. Moerland
c
Institute of System Architectures in Aeronautics, German Aerospace Center, Hamburg, Germany
Keywords:
Knowledge-Based Engineering, Object-Oriented Programming, Semantic Web Technologies, Collaboration,
Complex-system Development, Collaborative Engineering.
Abstract:
The development of complex systems within multi-domain environments requires an effective way of cap-
turing, sharing and integrating knowledge of the involved experts. Modern Knowledge-Based Engineering
(KBE) systems fulfill this function, making formalized knowledge executable by using highly specialized
environments and languages. However, the dedication of these environments to their domain of application
poses limitations on the cross-domain integration of KBE applications. The use of Semantic Web Technolo-
gies (SWT) delivers a domain-neutral way of knowledge formalization and data integration which promises
to drastically reduce the effort required to integrate knowledge of multiple domains in a single representation.
Especially within the complex field of aeronautical vehicle design the authors are working in, characterized by
several individual disciplines having to be considered simultaneously, the combined usage of KBE and SWT
technologies seems an attractive approach for the continued digitalization of the design process. In this paper,
the COllaborative DEsign and eXploration (Codex) framework is presented which aims at merging these two
technologies into a single framework that can be used to create domain-specific knowledge-bases and inte-
grate these into a single model of the overall product. Formalizing and executing this model will lead to a
more transparent and integrated view on complex product design.
1 INTRODUCTION
Designing the next generation of aeronautical vehi-
cles requires a system of systems approach in which
a large amount of disciplinary domains need to be in-
tegrated in an overarching product development pro-
cess. Due to the current trend towards digitalization of
processes, an increasing amount of data is generated
during the design of these complex products. This
data must be aggregated and processed into compre-
hensible information before decisions can be based
upon it. These often manually executed tasks place an
increasing burden on the low number of experts avail-
able, in particular when the processes are coupled and
the tasks have to be repeated many times. The amount
of data and ever-changing landscape of requirements
and decisions can also lead to the emergence of errors
and inconsistencies among the different models.
Knowledge-Based Systems (KBS) are developed
with the aim of capturing and formalizing the ex-
perts’ knowledge so that the data processing, require-
a
https://orcid.org/0000-0003-3207-4138
b
https://orcid.org/0000-0001-7579-9628
c
https://orcid.org/0000-0003-4818-936X
ments analysis, and decision making can be (par-
tially) automated by using a computer for their execu-
tion (Milton, 2008). A KBS automatically propagates
user decisions to the relevant requirements, compo-
nents, and parameters, cascading these changes and
re-validating the model. In this way, true concur-
rent engineering in which several disciplinary anal-
ysis can be performed at the same time becomes pos-
sible. These approaches promise to drastically reduce
the time between each design iteration, ensure con-
sistency with the formalized experts’ knowledge and
free up resources to assess the impact of specific de-
cisions on the results.
In literature (La Rocca, 2012), Knowledge-Based
Engineering (KBE) is defined as the merging of dif-
ferent disciplines such as Artificial Intelligence, Com-
puter Aided Design and Object-Oriented Program-
ming (OOP). It can be considered as an extension of
KBS with a stronger focus on tasks that have a large
impact on the engineering design process, namely
solving computational problems, dealing with geome-
tries and configurations. KBE application develop-
ment is still a fairly new discipline and it has seen a
limited use only in specific engineering sectors such
242
Zamboni, J., Zamfir, A. and Moerland, E.
Semantic Knowledge-Based-Engineering: The Codex Framework.
DOI: 10.5220/0010143202420249
In Proceedings of the 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2020) - Volume 2: KEOD, pages 242-249
ISBN: 978-989-758-474-9
Copyright
c
2020 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
as aerospace and automotive. This can be traced to
the fact that these specific sectors deal with the de-
sign of safety critical and complex systems that need
to adopt new technologies to remain competitive. All
this while under the pressure to use less resources,
to shorten the time-to-market of the products, and
to satisfy a large and continuously evolving set of
requirements. KBE proves to be a good methodol-
ogy for front-loading knowledge in the design pro-
cess, thereby creating a better foundation for design-
decisions that have to be made early in the product
life-cycle, which often have significant influence on
the product performance (Kulkarni et al., 2017).
Managing knowledge in multi-domain environ-
ments, such as product development, requires a col-
laborative framework that is well aligned with the
knowledge engineering process. Creating such a
framework for knowledge representation is one of
the fundamental aspects in designing a KBE applica-
tion (Sanya and Shehab, 2014). The KBE tools avail-
able on the market all adopt OOP as the core mod-
eling approach. We argue that OOP has its limita-
tions when it comes to fully capturing and integrating
experts’ knowledge as it limits possibility for meta-
modeling to frame-based structures. Our research
proposes a novel approach to knowledge formaliza-
tion in KBE applications, in which a more generic
knowledge structure using Semantic Web Technolo-
gies (SWT) (World Wide Web Consortium, 2015) is
adopted in an effort to remove these limitations.
The Codex framework is currently being devel-
oped at the German Aerospace Center (DLR), aim-
ing to overcome a large part of the aforementioned
burdens. It targets to ease the creation of knowledge-
based engineering tools, which can be easily inte-
grated in a digitally coupled overarching product de-
velopment process for aeronautical vehicle design.
Codex is a continued development of the model gen-
erator (Zamfir et al., 2018) and it aims to improve the
accessibility, extendability, and ease of cross-domain
knowledge reuse.
The following section will highlight the chal-
lenges of collaboration in multi-domain environments
and present possible solution to these. Section 3 fo-
cuses on human-machine interaction, the importance
of the meta-modelling environment and its impact on
knowledge formalization. Thereafter, section 4 will
present the current state of the Codex framework and
provide an example of multi-domain integration. This
paper concluded with a summary and an outlook on
future goals of this framework.
2 MULTI-DOMAIN
COLLABORATION
One reason for the aforementioned low adoption of
the KBE methodology can be be found in the large
difference between the abstract, high-level language
used in these applications compared to the highly
specific domain languages and tools generally used
within each discipline. Specific domain languages
and environments allow for a high level of expres-
siveness, since the semantics are tailored to enable
very precise modeling while using a set of vocabu-
lary and concepts that the domain-expert is already
familiar with. In contrast, to allow for integration
and high-degree linking of knowledge from all stake-
holders within a coherent knowledge base, a very ab-
stract modeling language is required, which makes no
assumptions on the domain it is used for and there-
fore features low expressiveness for specific domains.
Therefore, the choice of the modeling language for
a collaborative KBE application requires a trade-off
between applicability to different domains (abstrac-
tion/generality of the language) and expressiveness
within the domain itself.
Two examples of KBE tools used in the aerospace
sector are Pacelab APD (TXT Group, 2020) and
ParaPy (ParaPy B.V., 2019). The choice of mod-
eling languages are C# for APD and Python with
a specific flavor of annotations for Parapy, which
makes knowledge-formalization straight-forward for
programmers with experience in these languages. In
both tools the meta-model is defined in a frame-
like structure (Minsky, 1974), resulting from the ne-
cessity to create the model in the particular object-
oriented programming language used by the applica-
tion. Moreover, the model is defined in a hierarchical
way as this structure is commonly used when mod-
eling a product and its sub-parts within the engineer-
ing domain. Describing relationships among parame-
ters from the same discipline or in the same system of
components is fairly easy thanks to the inherent con-
nections of this type of structure (e.g. parent and child
relationships).
An issue that may arise from a frame-based and
hierarchical structured knowledge capture approach
is that higher complexity, especially via cross-branch
connections, tends to increase rigidity of the knowl-
edge base and makes it more complicated to use. This
over-coupling diminishes the potential of knowledge
re-usability since the coupled knowledge is not en-
capsulated anymore and its usage introduces many
more (unwanted) dependencies. In fact, the hierarchi-
cal structure often does not even reflect the naturally
emerging structure of collaboration, which resembles
Semantic Knowledge-Based-Engineering: The Codex Framework
243
a graph where many cross-branch connections ex-
ist (Alexander, 1965). This unconstrained structure
yields a much higher degree of possible complexity
in designing knowledge while decreasing the friction
of cross-domain knowledge transfer.
Although this frame-like and hierarchical way of
modeling has been proven to be effective in practice,
within the Codex framework we choose a modeling
approach that focuses on avoiding the emergence of
conflicts that commonly arise in multi-domain envi-
ronments. This approach is based on semantically ex-
pressing the precise meaning between models of dif-
ferent domains, which we assume to be key to create
a highly collaborative KBE application.
SWT have the potential to tackle some of the
challenges just mentioned. The models created us-
ing SWT are based on the formal semantics of
RDF Schema (RDFS) and Web Ontology Language
(OWL) (World Wide Web Consortium, 2015) which
are unambiguous, precise, can be automatically val-
idated for consistency and can be understood by
experts from other domains. The increased meta-
modeling capabilities allowed by these formal se-
mantics enable the integration of common knowledge
shared by multiple domains while still persisting all
domain-specific interpretations.
Codex utilizes a hybrid modeling approach
where the internal knowledge representation is as
generic as possible (using SWT), while the mod-
eling interface used by the engineer is as close
as possible to his/her Domain-Specific Language
(DSL) (i.e. the language commonly used within
that specific domain). Examples of such DSLs
within the field of aerospace are the Systems-
Modeling-Language (SysML) (Object Management
Group, 2020) and Common-Parametric-Aircraft-
Configuration-Schema (CPACS) (Alder et al., 2020).
Codex is able to translate these (and several other)
DSLs to a generic knowledge representation, allow-
ing different disciplines to unambiguously link parts
of their knowledge effectively creating an inter-
domain knowledge representation of the overall prod-
uct. The following section will elaborate on this as-
pect.
As depicted in Fig. 1, this mechanism enables
coupling multiple disciplines across the entire life-
cycle of an aircraft configuration. Already in the
early stages of development, this coupling is automat-
ically managed by Codex and takes away the burden
of having to manually communicate inter-domain de-
pendencies. This mechanism ultimately results in the
availability of more time for creative, value-adding
design tasks to be performed by the engineers and
largely reduces the amount of mistakes made in in-
terdisciplinary communication.
Data interchange: RDF
Taxonomies: RDF-S
Syntax: XML
Ontologies: OWL Rules: RIF/SWRL
Identifiers: URI Character Set: Unicode
Querying:
SPARQL
Root Knowledge representation
CPACS
XSD
CPACS
Aero
CPACS
Structures
CPACS
[X]
SysML
UML
Custom
Language B
Custom
Language A
DSL
DSL
Custom
Language A-B
Figure 1: Coupling knowledge of different aerospace disci-
plines through language layers in Codex.
3 CHALLENGES IN MODELING
FOR KBE APPLICATIONS
The following section highlights the challenges that
arise when formalizing knowledge in KBE applica-
tions by first focusing on the interaction between hu-
mans and machines. Thereafter, the importance of the
meta-modeling environment is discussed by compar-
ing frame-based structures with a knowledge graph
approach.
3.1 Human-machine Interaction
Effective collaboration among machines or software
that should produce deterministic results demands a
fixed communication protocol. Changes to that pro-
tocol often require making changes to the machines
and software as well. When multiple tools/knowledge
bases must be integrated, the problem of language
barriers arises from the nature of having multiple
domains, each with their own DSL. Each DSL is
expressed through different software-systems, file-
formats, and ontologies. When two or more domains
have to share knowledge (collaborate), they often ex-
press the same (or at least similar) meaning of this
shared subset of knowledge in their own DSL. To then
integrate knowledge from a domain A and another
domain B a Model-to-Model Transformation (M2M)
needs to take place, which could manifest itself as a
file-based export into a common format understood
by both A and B or expressing the semantic mean-
ing between A and B via ontological statements. The
first method of integration is most commonly used
but does not scale well with the number of domains
involved (n), since for every additional domain the
KEOD 2020 - 12th International Conference on Knowledge Engineering and Ontology Development
244
Tool A
AàSchema
Schema
Definition
Specification Documents
SchemaßB
contributes contributes
develops
develops
Tool B
Machine
Language
Based on
Tool A Developer Tool B Developer
maintains
maintains
Schema Developer
uses
defines
(a) Human-machine protocol definition using a schema.
Codex
Tool A
develops
develops
Tool B
Tool A Developer
Tool B Developer
maintains
maintains
Knowledge Integrator
Domain
Ontology
A
Domain
Ontology
B
I/O I/O
integrates
(b) Human-machine protocol definition using Codex.
Figure 2: Comparison of legacy and Codex approaches.
number of potentially necessary M2M grows expo-
nentially (n (n 1)).
A solution to this problem is to extend the com-
mon format to all the involved domains. Such an ap-
proach has been taken within the aerospace commu-
nity with CPACS, which was initiated by the German
Aerospace Center and is nowadays continually devel-
oped as an open-source format within the aerospace
community. This schema is required to include all
knowledge that must be shared by at least two in-
volved domains and forces these domains to agree on
how this shared knowledge is expressed. This makes
integration of tools within a multi-disciplinary design
optimization workflow easier. Some successful im-
plementations of this central data exchange format ap-
proach for multi-domain and cross-company collab-
oration have been shown in (Moerland, E., Pfeiffer,
T., B
¨
ohnke, D. et al., 2017) and (Ciampa and Nagel,
2018). Utilizing a central schema results in a positive
network-effect – the usefulness of the schema and the
tools that utilize it increases with the number of do-
mains and people participating.
This removes the M2M scaling issues, but often
lacks reflection of the aforementioned graph struc-
ture of collaboration. This process demands the tool
developer to provide a mapping between the com-
mon exchange language and the internal application
model. In fact, effective collaboration among people
demands a flexible communication protocol that can
capture subtle semantic differences in meaning, al-
lows for a variety of meanings about concepts to coex-
ist, and supports the distributed nature of knowledge
acquisition by enabling knowledge engineers to ad-
just their models at any time without having to know
all or any of the dependencies on their model.
Moreover, the fixed structure of a schema, to-
gether with the high number of domain dependencies,
makes modifications to it a labor intensive task, es-
pecially in a collaborative environment. This often
results in a reluctance to introduce changes and there-
fore decreases development speed. These problems
intensify with the number of domains and participants
involved, which counteracts the positive network-
effect the more domains and people make use of
the schema, the more rigid it becomes.
Figure 2a depicts the legacy approach, in which
the communications protocol between people is based
on natural language (or specification documents writ-
ten in natural language) and the communication pro-
tocol for the machines (tools) needs to be manually
derived from these specification documents by the
schema developer.
With Codex we aim to provide the collaborators
with a framework to add semantics to their exchange
protocol and the tools with an application program-
ming interface (API) to communicate their semantic
model to the multi-domain environment, as shown in
Figure 2b. These enhancements add formal meaning
to the protocol and remove the necessity for a central
schema while still allowing all tools to communicate.
This also removes the previously mentioned negative
network effect, since the labor intensive task of con-
tinuously keeping the schema-based inter-tool com-
munication protocols updated, has been simplified.
The task is now handled by the knowledge-integrator
that can independently define the initial shared mean-
ing between the different domain ontologies by us-
ing the formal semantics of RDFS and OWL. This
shared meaning can be updated and modified with-
out the need for the different domains to agree on
a common representation, effectively creating less
communication-overhead. Manual work is still re-
Semantic Knowledge-Based-Engineering: The Codex Framework
245
quired from the tool developer to maintain the seman-
tic model of his/her tool whenever the tool internals
are modified.
3.2 Meta-modeling Environment
In the OOP paradigm the class frame takes a central
role as other constructs such as objects, methods and
values cannot exist outside of their class. Informa-
tion in one class can be re-used in others by inheri-
tance and changes made to the classes are reflected in
the objects instantiated from them. If a class is modi-
fied, all objects that depend on it become invalid. This
strict separation of meta-model and model is resolved
in OOP languages by ”freezing” the former while
the latter is being built and used, thus ensuring that
issues due to meta-model incongruities cannot arise
during run-time. This is usually done by compiling
the classes defined in the source code.
In KBE tools based on OOP languages the strict
separation of meta-model and model is implicitly
taken and becomes part of how the knowledge en-
gineer interfaces with the knowledge base. A posi-
tive result of having the meta-model in compiled form
is that performance during the knowledge execution
phase can be improved. On the other hand, a di-
rect drawback of this choice is that it takes time to
compile the meta-model and switch to the modeling
tasks. This can negatively impact the experience of
the knowledge engineer as he/she would not be aware
of the impact of the changes on the model until the
time-intensive process needed to make such changes
available is finished; in other words, rapid iteration
of the KBE system is limited and the expert is en-
couraged to make as many changes as possible before
compiling and assessing the changes. Furthermore,
having to compile the meta-model before it can be
used makes it less practical to have a KBE applica-
tion deployed on the web, which is a major advantage
when it comes to collaboration and adoption and one
of the future goals of the Codex framework.
In SWT standards every entity is a resource with
equal importance. Classes, individuals, object- and
data-properties can be linked together but their ex-
istence is not dependent on the presence of a spe-
cific entity, as is the case for objects and classes in
OOP. Moreover, there is no strict separation between
classes and individuals as, depending on the state-
ments made, a resource can be both. For example,
an expert could see an A320 as an individual of type
Aircraft while a different (and not conflicting) view
of another expert is that A320 is a class for a family
of aircraft that shares a specific set of basic proper-
ties. Thus A320 can be simultaneously a class and an
individual depending on the specific point of view.
We believe that the proposed approach within
Codex can improve knowledge re-usability when
compared to OOP based KBE tools. One reason for
this is that only the resources that are actually needed
can be re-used, leaving out unnecessary information
that might actually overcomplicate the derived knowl-
edge base. Another reason for this improvement can
be found in the ability to freely mix classes and indi-
viduals; any model can become the starting point for
a new meta-model without much overhead. Friction
of implementing new knowledge is also decreased as
the engineer does not need to worry about having to
model his/her ideas in a strictly hierarchical language.
In Codex we adopt more flexible modeling capa-
bilities allowed by SWT when building knowledge
bases. Doing so makes the boundary between meta-
model and model blurred, when compared to the OOP
approach, leading us to decide not to compile any of
the user-defined knowledge. In this way, the expert
is allowed to modify and extend the meta-model and
model simultaneously during run-time. This choice
should allow for a faster iteration of knowledge for-
malization leading to a higher speed of innovation.
Deployment of the complete knowledge base on the
web becomes also possible allowing multiple users to
collaborate concurrently on the same project.
4 THE CODEX FRAMEWORK
With the Codex framework we provide an implemen-
tation of a KBE application where semantic knowl-
edge formalization plays a central role. The frame-
work currently consists of several modules, that allow
the different domain experts to solve many of their
knowledge-engineering tasks in a holistic way. Codex
is written in the Kotlin programming language (Kotlin
Foundation, 2020), as it has the ability to leverage the
syntax of the language to create so-called type-safe-
builders, which are a way of creating custom DSLs
within the language. This allows for the provision of
an implementation of the aforementioned hybrid ap-
proach to knowledge formalization not just to users
of our Graphical User Interface (GUI), but also to do-
main experts that like to express their knowledge in
code, as well as developers of plugins for the Codex-
framework.
Within Codex, functionalities are provided in the
form of plugins, each possibly providing its own
DSL and ontology. The core module of the frame-
work is the codex-semantic module that makes use
of existing and well-established standards for knowl-
edge representation (Resource Description Frame-
KEOD 2020 - 12th International Conference on Knowledge Engineering and Ontology Development
246
work (RDF) (World Wide Web Consortium, 2015),
OWL) and utilizes mature open-source libraries,
such as those provided by the Apache Jena Frame-
work (Apache Software Foundation, 2020). This
module takes care of the creation and manipulation
of OWL-models.
The modular plugin architecture of Codex also al-
lows for existing tools, languages, and technologies
to be connected to it. A plugin that has already been
implemented is the codex-parametric module that al-
lows users to semantically define and solve systems
of parametric constraints with a DSL that closely re-
sembles common mathematical notation. This plugin
contains an execution engine that provides informa-
tion about parameter dependency (e.g. over/under de-
termination of the system of constraints) and solving
capabilities.
Another plugin that is currently under develop-
ment provides the ability to import and export models
defined using the aforementioned CPACS schema. It
also exposes this schema as an OWL ontology and ad-
ditionally provides a custom DSL that can be used by
engineers already familiar with CPACS to create and
modify models. This allows Codex to seamlessly in-
tegrate with the already existing, vast eco-system of
CPACS-enabled tools.
The knowledge asserted by the domain expert
can easily be enhanced by simply adding the formal-
semantics of RDFS and OWL as a set of rules that
will infer additional knowledge. Moreover, the user
can utilize this same mechanism (and the provided
Rule-DSL) to add custom rules to the knowledge-
base. While rules expressed through this mecha-
nism are sufficient for semantic-inferences, they do
not cover the definition of production-rules (Russell
and Norvig, 2016), that demand additional expres-
sions to be possible, such as the definition of a pri-
ority of rule-execution. These production-rules are an
essential part of any KBE application, since they al-
low for the dynamic, rule-based creation of a large
number of interconnected individuals. To satisfy this
need, a dedicated module that allows such production-
rules to be conveniently expressed via a custom DSL
and executed using state-of-the-art algorithms and li-
braries is being developed. These user-defined rules
will not be stored within the OWL model, but a more
appropriate standard such as the Rules-Interchange-
Format (World Wide Web Consortium, 2013).
4.1 Multi-domain Integration Example
In this subsection, a simplified example of how multi-
ple domains can be integrated with the help of Codex
is discussed. The following provides some exam-
ples of parameters and rules present within multiple
aerospace design domains, each represented by their
individual namespace:
a: Aircraft Sizing Domain.
(a : Wing rd f : type owl : Class)
m: Mission Analysis Domain.
R ·
g·T SFC
V
·
C
D
C
L
= ln(
M
end
M
init
)
M
end
= M
init
M
f uel
or
M
f uel
= Mission Tool(R, C
L
, C
D
, T SFC, . . . )
t: Fuel-tank Sizing Domain.
Fuel Density =
Fuel Mass
Fuel Volume
Tank Volume = Fuel Volume · Factor
r: Requirements Domain.
Range 6150km
”All fuel must be stored in the wing”
Through the use of the codex-parametric module’s
DSL and ontology, the above domains’ knowledge
can be captured; for example, an equation is mod-
elled as an expression tree where each node is ei-
ther of type operator, function (e.g. sin(x)), param-
eter, or constant. This expression tree is stored as a
RDF graph. External parametric tools, such as ”Mis-
sion Tool”, can be included in the knowledge graph as
they behave like functions having a local name within
the ontology, requiring input parameters and provid-
ing an output. During modeling, either of the defini-
tions of m : M
f uel
(equation or external tool call) can
be utilized to compute its value, as the constraints in
codex-parametric are automatically reversed by use
of numerical convergers.
During the knowledge integration phase, param-
eter equivalency between different domains can be
quickly achieved with RDF statements such as:
(m : M
f uel
owl : sameAs t : Fuel Mass)
(m : R owl : sameAs r : Range) .
This ensures that the two parameters are treated as
the same entity by the execution engine, and conse-
quently their calculated values will be equivalent.
Additionally, knowledge expressed through natu-
ral language, such as the requirement that ”all fuel
must be stored in the wing”, can be integrated by RDF
statements such as:
(r : f uel r : storedIn r : wing)
(r : f uel owl : sameAs t : Fuel Volume)
(r : wing rd f : type a : Wing) .
This example is visually represented in Figure 3
where the unconnected knowledge graphs are mean-
ingfully integrated through the use of semantic state-
ments. With this example we show that a schema is
not necessary to bridge the gap between the different
Semantic Knowledge-Based-Engineering: The Codex Framework
247
Aircraft
Mission
Fuel-tank
Requirements
Knowledge Integration
Figure 3: Multi-domain knowledge integration.
domains. Maintenance is only necessary if a change
within a domain alters the meaning of the integrating
statements.
5 CONCLUSIONS
Utilizing formal knowledge definitions in the engi-
neering domain enables making the expert’s knowl-
edge reusable, modular, and machine executable.
However, the reliance of KBE on OOP introduces
limitations that might be a cause for the low adop-
tion rate of this methodology. The choice of using a
generic language (OWL) that makes no assumptions
on the domain it is being applied to, enables a way to
include other more specific approaches to system en-
gineering such as the widely adopted SysML. Codex
aims to augment these approaches by providing the
ability to apply custom rules to these models, effec-
tively making them executable. Moreover, it is one of
our goals to leverage the inference mechanism com-
bined with the model integration capability to infer
new knowledge from the models described through
these well established approaches.
The overall aim of the Codex framework is to pro-
vide a holistic knowledge-formalization and execu-
tion environment that reflects and supports the graph
structure of an effective collaborative environment.
Although Codex is at an early stage of development
and the main expected improvements have yet to be
validated, the framework can already be utilized to
define, link, and use ontologies for the creation of exe-
cutable KBE tools. A possible obstacle to its adoption
is that semantic modeling might prove to be difficult
for the experts involved in the product development
process and the benefits do not outweigh the required
learning curve. Another pitfall could arise when users
do not follow semantic modeling best practices or
specialize their ontologies too much, lowering their
potential for re-usage within other domains.
The Codex framework is currently being devel-
oped to create a framework for knowledge digitaliza-
tion and integration for usage within aeronautical ve-
hicle design. When the framework is fully matured
and has proved its usage within the design process,
the opportunity arises to further open-up the design
space by incorporating a larger amount of engineer-
ing and scientific domains, e.g. from other means of
transport. With this, the basis is created for the over-
all design of more efficient, fully integrated mobility
solutions.
Future developments include further integration
with existing standards and tools, implementation of
a rule-based production system, and the creation of
a data-analytics and visualization environment that
leverages the data-integration and exploration capa-
bilities of SWT.
REFERENCES
Alder, M., Moerland, E., Jepsen, J., and Nagel, B. (2020).
Recent advances in establishing a common language
for aircraft design with CPACS. In Aerospace Europe
Conference, Bordeaux.
Alexander, C. (1965). A city is not a tree. Architectural
Forum, 1965, 122.
Apache Software Foundation (2020). Apache Jena. https:
//jena.apache.org/. Online; accessed Jul 27, 2020.
Ciampa, P. D. and Nagel, B. (2018). AGILE the Next
Generation of Collaborative MDO: Achievements and
KEOD 2020 - 12th International Conference on Knowledge Engineering and Ontology Development
248
Open Challenges. In 2018 Multidisciplinary Analysis
and Optimization Conference. American Institute of
Aeronautics and Astronautics.
Kotlin Foundation (2020). Kotlin Programming Language.
https://kotlinlang.org/. Online; accessed Jul 27, 2020.
Kulkarni, A., van Dijk, R., van den Berg, T., and Rocca, G.
(2017). A knowledge based engineering tool to sup-
port front-loading and multi-disciplinary design opti-
mization of the fin-rudder interface. In Aerospace Eu-
rope 6th CEAS Conference, volume 680.
La Rocca, G. (2012). Knowledge based engineering: Be-
tween AI and CAD. Review of a language based tech-
nology to support engineering design. Advanced En-
gineering Informatics, 26(2):159–179.
Milton, N. (2008). Knowledge Technologies. Polimetrica,
Monza IT.
Minsky, M. (1974). A framework for representing knowl-
edge. MIT AI Memo 306.
Moerland, E., Pfeiffer, T., B
¨
ohnke, D. et al. (2017). On the
design of a strut-braced wing configuration in a col-
laborative design environment. In 17th AIAA Aviation
Technology, Integration, and Operations Conference,
page 4397.
Object Management Group (2020). OMG Systems Model-
ing Language (OMG SysML). https://www.omg.org/
spec/SysML/. Online; accessed Jul 27, 2020.
ParaPy B.V. (2019). ParaPy - Knowledge Based Engineer-
ing Platform. Online; accessed Jul 27, 2020.
Russell, S. and Norvig, P. (2016). Artificial Intelligence:
A Modern Approach, Global Edition. Pearson Educa-
tion Limited.
Sanya, I. and Shehab, E. (2014). An ontology framework
for developing platform-independent knowledge-
based engineering systems in the aerospace indus-
try. International Journal of Production Research,
52(20):6192–6215.
TXT Group (2020). Pacelab APD. https://www.txtgroup.
com/markets/solutions/pacelab-apd/. Online; ac-
cessed Jul 27, 2020.
World Wide Web Consortium (2013). RIF Production Rule
Dialect (Second Edition). https://www.w3.org/TR/rif-
prd/. Online; accessed Jul 27, 2020.
World Wide Web Consortium (2015). Semantic Web. https:
//www.w3.org/standards/semanticweb/. Online; ac-
cessed Jul 27, 2020.
Zamfir, A., Jepsen, J., Moerland, E., and Nagel, B. (2018).
Development of a modular knowledge-based model
generator for the preliminary aircraft design process
of the future. In 2018 Aviation Technology, Integra-
tion, and Operations Conference. American Institute
of Aeronautics and Astronautics.
Semantic Knowledge-Based-Engineering: The Codex Framework
249