Enabling Sustainable Interoperability for Enterprise Applications with
Knowledge Links
Artur Felic
1,2
, Felix Herrmann
1
, Christian Hogrefe
1
, Michael Klein
1
and Birgitta K¨onig-Ries
2
1
CAS Software AG, Karlsruhe, Germany
2
Friedrich-Schiller-Universit¨at Jena, Institut f¨ur Informatik, Jena, Germany
Keywords:
Semantic Web, Knowledge Links, Semantic Interoperability, Sustainable Interoperability, Enterprise Interop-
erability.
Abstract:
In complex and collaborative ecosystems like business networks, different partners need to share their re-
spective expert knowledge in order to be successful. Due to the diversity of business applications and highly
customized application suites used by business partners, knowledge exchange and the establishment of inter-
operability between enterprise applications is extremely difficult. Different representations and meanings of
enterprise data lead to incomprehensibility between partners inside collaborative environments. This paper
presents an model-driven approach to support sustainable interoperability for enterprise applications in col-
laborative environments. Based on an event-driven architecture, Knowledge Links enable dynamic modelling
of knowledge transformations between knowledge domains. They keep background consistency between the
connected domains, thus making enterprise applications interoperable. Knowledge Links can be created and
modified at any time, which enables sustainable interoperability. Business partners are able to rely on their
enterprise applications and don’t need to switch to another system. Sensitive data stays covert due to the nature
of modular ontologies. The presented approach is exemplified in the context of the OSMOSE Project and will
be evaluated by the proof-of-concept scenarios in this Project.
1 INTRODUCTION
Enterprise Applications can be described as highly
customized applications or application suites that
serve individual enterprise needs. They often com-
prise a set of expert tools for specialist capabilities.
According to Fowler (Fowler, 2003) Enterprise
applications are about the display, manipulation, and
storage of large amounts of often complex data and
the support or automation of business processes with
that data”. Enterprises usually compose their set of
applications by applications from different competing
vendors, which do not want to reveal sensitive data
in order to be integrated. Since the paradigm shift
from individual businesses to Complex and collabo-
rative ecosystems like business networks with many
business partners, the situation became even more
complex. Competing and globally spread enterprises,
having many different systems installed, have to
collaborate in order to create business value. This
circumstance leads to a problem from which a whole
branch of science called Enterprise Interoperability
emerged.
According to the IEEE Standard Computer Dic-
tionary (IEEE, 1991) Interoperability can be defined
as the ”ability of systems or components to exchange
and use information”. Complementary, the ISO
1610 standard defines interoperability as the ”ability
to share and exchange information using common
syntax and semantics to meet an application-specific
functional relationship through the use of a common
interface”. Enterprise Interoperability concerns data,
services, processes and internal business. Thus,
different data models, various services and processes
as well as different levels of organization are aimed to
be interoperable. Many emerging Enterprise Interop-
erability Frameworks like EIF (IDABC et al., 2004),
FEI (Chen and Daclin, 2006), C4IF (Peristeras and
Tarabanis, 2006) and AIF (ATHENA Project, 2007)
aim at enabling Enterprise Interoperability. Although
the implementation of Enterprise Interoperability in
a collaborative ecosystem is desirable, the dynamics
inside the ecosystem also need to be taken into
account to be successful. The implementation is not
a one-off activity, but a continuous process that needs
to be sustainable in order to allow the entrance of
new business partners or changes inside the existing
597
Felic A., Herrmann F., Hogrefe C., Klein M. and König-Ries B..
Enabling Sustainable Interoperability for Enterprise Applications with Knowledge Links.
DOI: 10.5220/0005369605970607
In Proceedings of the 3rd International Conference on Model-Driven Engineering and Software Development (MDE4SI-2015), pages 597-607
ISBN: 978-989-758-083-3
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
ecosystem.
Sustainability and interoperability are two inher-
ently linked and inseparable aspects that need to be
addressed together (Dassisti et al., 2013). Sustainable
Interoperability can be described as ”novel strate-
gies, methods and tools to maintain and sustain the
interoperability of enterprise systems in networked
environments as they evolve with their environments”
(Jardim-Goncalves et al., 2012). Detection of new
enterprise systems, learning capabilities, adaptability,
analysis of the internal behaviour as well as com-
munication inside the network is addressed by this
research area. Interoperable collaborative ecosystems
with many different business partners running many
different enterprise applications must be seamlessly
configurable.
The Object Management Group (OMG)
(Object Management Group, 2014b) proposed
Model-Driven Architecture (MDA) (Object Man-
agement Group, 2014a) as a strategy towards
interoperability of heterogeneous systems (Singh and
Sood, 2009). The separation of computation indepen-
dent models (CIMs), platform independent models
(PIMs) and platform specific models (PSMs) enables
interoperability between enterprise applications by
allowing transformations between CIMs, PIMs and
PSMs. Long term flexibility, incorporating models
from different viewpoints, technology independence,
multi platform model support and easier integration
and interoperability are some of the major benefits of
MDA. Cost for the development life cycle is reduced
whereas software quality can be improved.
MDA can be extended by Semantic Web tech-
nologies, i.e. ontologies, to create unambiguous
domain vocabularies (W3C, 2006). Gruber defines
ontologies as ”an explicit specification of a concep-
tualization” (Gruber, 1993). Knowledge sharing is
fundamental in collaborative ecosystems. Ontologies
can be used to structure knowledge and knowledge
properties in order to make knowledge understand-
able for machines and humans. There are many
ontologies defined for different purposes to create
shared knowledge structures and to set up a common
knowledge base.
In highly dynamic collaborative ecosystems,
experience has shown that it is unrealistic to assume
commitment from every current and future business
partner to an integrated ontology covering all relevant
aspects of all business applications. A new business
partner entering the collaborative ecosystem or the
substitution of an enterprise application could lead
to the necessity to change the whole knowledge
base. Upper ontologies like SUMO (Niles and Pease,
2001) are purposed to structure only very general
concepts in order to support semantic interoperability.
Following the idea of modular ontologies (Ben Abbs
et al., 2012) large monolithic ontologies can be
divided into smaller domain modules in terms of
interchangeability. As the name suggests, domain
ontologies are often not comprehensible outside
the domain, but the knowledge structured inside is
of importance for other domains. Therefore, links
between those ontology modules are necessary in
order to provide different perspectives of knowledge
to the business partners that require it.
In this paper, we propose a model-driven ap-
proach using Knowledge Links (Felic et al., 2014).
Knowledge Links are used to connect the knowledge
domains and transform knowledge from one to
another. Their specification constitute a Platform
Independent Model (PIM). Depending on the un-
derlying Platform Model (PM), i.e. the platform,
enterprise applications or database of business part-
ners involved in the Knowledge Link, the modelled
Knowledge Link is further transformed into events
and business processes by model transformations into
a Platform Specific Model (PSM).
By connecting different knowledge domains in a
black box manner, business partners of a collabora-
tive environment can rely on their knowledge struc-
tures. We present an event-driven and service ori-
ented architecture approach in which each enterprise
application constitute a separated knowledge domain
with its own ontology. Furthermore, a tool is pre-
sented with which knowledge links can be created
and maintained at any time during system runtime and
thus enable sustainable interoperability. During sys-
tem runtime, the complex event processor of the mid-
dleware is aware of keeping background consistency
by analysing events that are associated with Knowl-
edge Links and reacting according to the modelled
behaviour.
In the OSMOSE Project (OSMOSE, 2014), the
aim is to interconnect the real, digital and virtual
world for sensing liquid enterprises. In the real
world physical devices, objects and actors are present,
whereas their virtual models and hypothetical scenar-
ios are contained in the virtual world. The digital
world includes data sets about objects and their mod-
els as well as metadata and multimedia content. The
sensing liquid enterprise is composed of two enter-
prise paradigms: the sensing enterprise and the liq-
uid enterprise. The sensing enterprise is a cooperative
nervous system where virtual and physical smart ob-
jects, equipment and infrastructure form an active net-
work. In the liquid enterprise, boundaries are fuzzy
and it is hard to distinguish the inside from outside
in terms of human resources, markets, products and
MODELSWARD2015-3rdInternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
598
processes due to the prevailing cloudiness. An event-
drivenand service-oriented middleware automatically
keeps background consistency between these worlds,
i.e. the shadow ambassadors of things in each of the
world. Each world has there own applications and
data structures, while the middleware has to be aware
of the meaning of data in order to decide which data
has to cross world borders to transform and deliver
data to the other worlds. The situation is quite similar
to business networks and enterprise applications. The
three worlds together with the middlewarecan be seen
as an collaborative environment, whereas the worlds
themselves are business partners with their installed
enterprise applications.
After this introduction, in Section 2, we will ex-
plore related work to this paper. In Section 3, the
modular and hierarchical knowledge base concept is
introduced. The proposed Event-Driven Architecture
is described in Section 4. The two latter sections
constitute the foundation of Section 5, in which the
sustainable interoperability approach is illustrated. In
Section 6 the paper is concluded and summarized and
future research directions are specified.
2 RELATED WORK
Enterprise Interoperability has received growing at-
tention in the last decade. (Lampathaki et al., 2012)
identified the lack of scientific foundations and for-
mulated several scientific areas of enterprise interop-
erability based on technological trends and enterprise
interoperability background knowledge. The main
scientific areas the present paper addresses are data,
software, knowledge and ecosystems interoperability.
Many sustainable interoperability approaches in
literature arise from the scientific fields of Semantic
Web, Model-Driven Architecture (MDA) or Event-
Driven Architecture (EDA). While one of the main
objectives of semantic web technologies is to induce a
common meaning about knowledge for human beings
and machines, MDA and EDA are concerned with
platform independence and heterogeneity of systems
besides other things. The following scientific findings
and publications give an overview about related work
in these scientific areas.
(Chen et al., 2008) analysed enterprise interoper-
ability architectures from 1985 to 2000 and states that
most effort has been put in framework approaches
rather architectures of enterprise systems. Further-
more, the authors identified the lack of an enter-
prise architecture ontology, the absence of scientific
methodologies and interoperability between existing
architectures as well as weak impact in industry. The
authors identified conceptual barriers as one of the
main category of interoperability barriers that are
”concerned with the syntactic and semantic differ-
ences of information to be exchanged”, which highly
motivates our work. Moreover, they describe the in-
teroperability of data as a category of interoperability
concerns that ”deals with finding and sharing infor-
mation from heterogeneous data sources”. According
to the categorization of interoperability approaches,
we present a federated approach, where partners do
not impose own models on the enterprise architecture.
In our approach, we allow that each partner can rely
on its own knowledge structured in an ontology, al-
though partners may share knowledge in a common
ontology. Semantic mapping is handled with Knowl-
edge Links by interconnecting the ontologies appro-
priately.
The authors of (Chungoora et al., 2013) utilize
a model driven architecture approach and ontologies
to enable knowledge sharing for manufacturing sys-
tems interoperability. The Manufacturing Core On-
tology constitutes the heart of the Platform Indepen-
dent Model (PIM), whereas other domains extend the
Manufacturing Core Ontology to specify domain spe-
cific models. To enable knowledge sharing of the
further transformed Platform Specific Model (PSM),
knowledge verification mechanisms are specified at
PIM level. However, the approach is limited to the
specification of constraints on PIM level and does
not allow appropriate knowledge propagation with
transformations in order to provide partners with new
knowledge build from their knowledge base.
Kadar et Al. (Kadar et al., 2013) propose a multi-
agent system to support sustainable interoperability
for networked organizations. Rule based negotiation
at business level by agents in a distributed environ-
ment enable the re-establishment of interoperability
between partners. Although the work of Kadar et Al.
is inspiring and knowledge negotiation is demand-
ing, the need to transform knowledge between differ-
ent knowledge domains and the creation of complex
knowledge relationships among different partners in
terms of sustainable interoperability is not covered.
Knowledge Links could therefore complement this
work.
(Ni and Fan, 2008) present an ontology based
approach. Domain ontologies are connected by
special collaboration ontologies, which are semi-
automatically created. These collaboration ontolo-
gies are described by OWL-S (W3C, 2004) in order
to ”enables automatic service discovery, invocation,
composition, interoperation, and execution monitor-
ing”. The OWL-S descriptions from these collab-
oration ontologies are further transformed to BPEL
EnablingSustainableInteroperabilityforEnterpriseApplicationswithKnowledgeLinks
599
and WSDL to allow automated execution of ontol-
ogy mappings. Thus, an business process execution
engine is able to react appropriately to requests ac-
cording to data changes. Although the approach is
very closely related to the knowledge link approach,
it is strongly dependent on the mechanisms behind the
semi-automatic creation process of the collaboration
ontologies. Additionally, complex links with opera-
tions and transformations cannot be created by hand.
The authors of (Agostinho et al., 2011) propose
to enhance information systems by traceability func-
tionalities to enable sustainable interoperability. Us-
ing model morphisms that allow altering and non-
altering model mapping, the relationship of models
from different knowledge domains can be formalized.
These mappings are translated in executable code in
order to exchange mapped or transformed data be-
tween business partners. The work of Agostinho et
Al. is closely related to the Knowledge Links due to
the support of morphisms between different knowl-
edge domains. Mappings are created on ontology
level, therefore Knowledge Links could complement
this work by allowing end users to model the exe-
cutable connections with given tools for end users.
By using Knowledge Links, the presented approach
could be enhanced by reusability of mappings. The
result of a mapping could further be used to define
new mappings.
The approach presented in this paper combines
ontology-driven, event-driven and model-driven ap-
proaches and is based on three pillars: A modular and
hierarchical knowledge base, an event-driven archi-
tecture for collaborative environments and a model-
driven knowledge link approach with a business pro-
cess execution engine. In the following sections, we
will take a look at these in turn.
3 KNOWLEDGE BASE
STRUCTURE
The knowledge base structure is following the work
of (Ben Abbs et al., 2012) about modular ontolo-
gies. Major advantages of the modularization ap-
proach that are of particular importance are knowl-
edge reuse across application domains, distributed
ontology engineering over different knowledge do-
mains and effective management of ontology mod-
ules. Business partners in collaborative environments
often have different applications running in their sys-
tem. Business networks in which business partners
have to comply to a centralized knowledge base and
specific enterprise applications in order to collabo-
rate are hard to establish and nearly impossible to
maintain. Furthermore, they build barriers and ob-
stacles for new business partners, which is fatal for
highly dynamic environments. Therefore, modular
ontologiesallowbusinesspartners to rely on their own
knowledge structures and maintain their own knowl-
edge base while connecting their knowledge structure
to a shared common denominator. Additionally, the
modular approach allows business partners to struc-
ture their inner knowledge base similarly.
The knowledge base structures can be categorized
in different hierarchy levels. The highest level is con-
stituted by a common ontology structuring necessary
knowledge about the business network. At the second
hierarchy level the business partners’ ontologies are
located. Each business partner structures its knowl-
edge separately. The third and lowest hierarchy level
comprise the enterprise applications’ ontologies. Fig-
ure 1 illustrates the hierarchy levels.
BP
BP BP
BP
1
st
Level:
Business
network
EA
3
rd
Level:
Enterprise
Application
BP
EA
EA
2
nd
Level:
Business
partner
Figure 1: Knowledge Base Hierarchy.
Ontologies of lower levels inherit the ontologies
of the next higher level and enrich the higher level
concepts with new specific concepts or add specific
concepts underneath higher level concepts. Accord-
ing to (Ben Abbs et al., 2012), this pattern of imports
establishes an unidirectional correspondence between
the lower level ontologies and the higher level ontol-
ogy. More precisely, the highest level ontology con-
stitutes an upper ontology for the underlying ontolo-
gies. According to (Healy et al., 2010), an upper on-
tology is a high-level, domain independent ontology
whose concepts generic and basic and ”from which
more specific ontologies can be derived”. Mid-level
ontologies between the lowest and highest level of the
ontology hierarchy ease the mapping between these
levels, whereas the domain ontologies on the lowest
level reuse the concepts from the levels above and ex-
tend them with domain specific concepts.
Upper ontologies are helpful in terms of interoper-
ability. They are often used to describe generic, plat-
form independent concepts in order to generalise the
MODELSWARD2015-3rdInternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
600
meaning of domain specific concepts at a higher level.
There are many initiatives that aim to define standard
upper ontologies, for instance, the Suggested Upper
Merged Ontology (Niles and Pease, 2001). Domain
ontologies that are compliant with the upper ontolo-
gies are able to interoperate by shared terms and defi-
nitions and mapping to these terms.
In order to describe resources like docu-
ments, websites or parts of them in a computer-
understandable manner, resources of enterprise appli-
cations need to be semantically annotated. In (Oren
et al., 2006), annotations of various domains are anal-
ysed and a unified formal model for semantic annota-
tions is presented. Using such an annotation model al-
lows enterprise application resources to be structured
and processable by our presented event driven knowl-
edge link approach.
In our approach, we use Apache Jena (Foundation,
2014) as storage and interface for ontologies. Jena is
a semantic web framework consisting of three mod-
ules. An OWL (W3C, 2012) and inference API al-
lows interaction with ontologies and reasoning over
data. The built-in triple store enables persistence and
allows querying over SPARQL (W3C, 2008) and http.
An RDF (W3C, 2014) API allows users to create, read
and manipulate triples and handles serialization.
The knowledge base structure of the OSMOSE
Platform follows the schema described above. Like
depicted in Figure 2, the OSMOSE Upper-Ontology
constitutes a common upper ontology for the whole
platform at the highest hierarchical level. It contains
generic concepts about events, entities, services and
processes as well as platform specific concepts like
’OSMOSE World or OSMOSE Process’. This com-
mon knowledge base is imported by each of the three
worlds allowing the worlds to extend these concepts.
The Intra-World Ontologies are ontologies specific
for each of the world, acting as a mid-level ontology.
Inner world applications use own ontologies by im-
porting the Intra-World Ontology and adding specific
concepts. Thus, they define the lowest level of on-
tologies, the domain ontologies. By semantic annota-
tions, the inner-world applications can map their data
to their ontologies.
To give a brief example, let us assume that the
OSMOSE Upper-Ontology defines the concept of a
generic entity. Inside the real world, the concept ’hu-
man actor’ extends the generic ’entity concept. For
a physical machine having its own ontology, ’human
actor’ is extended by concepts for administrators and
users.
Middleware
(Membranes)
Digital
World
Ontology
Real
World
Ontology
Virtual
World
Ontology
OSMOSE Upper-Ontology
OSMOSE Platform Model
Entity Ontology
Event Ontology
Process Ontology
Service Ontology
Common Ontology
Human
Actor
Entity
Figure 2: OSMOSE Knowledge Base Hierarchy.
4 EVENT-DRIVEN
ARCHITECTURE APPROACH
An Event-Driven Architecture is a style of Service-
Oriented Architecture, where the occurrence of an
event can trigger the invocation of services, which
in turn can generate new events. (Michelson, 2006)
Event-driven systems are designed to process events
as they occur, allowing the system to observe, react
dynamically to and issue personalized data depend-
ing on the recipient and situation.(Etzion and Niblett,
2010). Event-driven architectures are deployed in
many areas and have proofed as a competitive advan-
tage for a lot of organizations. Thus, they are becom-
ing increasingly popular. (Grabs and Lu, 2012)
Events can be seen as kind of messages that are
generated by resources in domains and recognised by
information systems. Events can occur and be con-
verted from anywhere in an environment (Stojanovic
et al., 2011), indicating hat an occurrence has been
observed in a system (Grabs and Lu, 2012). Luckham
defines events as objects that are records of activities
in a system and can have a relation to other events
with a timestamps as properties (Luckham, 2002).
According to McGovern, ”an event is a change in the
state of a resource or request for processing” (McGov-
ern et al., 2006).
Complex Event Processing (CEP) is a specialized
field of event processing and part of event-driven ar-
chitectures dealing with complex events. Complex
events are combinations of multiple events consist-
ing of all the activities that the aggregated events
signify (Luckham, 2002) (Robins, 2010). Complex
Event Processing deals with the identification and
handling of interdependent events across organiza-
tions, analysing their impact and taking subsequent
action in real time into account. It is an emerging
technology which allows to find real time relation-
ships between different events using elements such
as timing, causality, and membership in a stream of
data in order to extract relevant information (Perro-
EnablingSustainableInteroperabilityforEnterpriseApplicationswithKnowledgeLinks
601
chon et al., 2001). Luckham states that ”Complex
Event Processing (CEP) is a defined as a set of tools
and techniques for analysing and controlling the com-
plex series of interrelated events that drive modern
distributed information systems” (Luckham, 2008).
According to (Vernadat, 2007), loose coupling
among services and applications is the key to building
interoperable enterprise systems. Message-Oriented
Middleware (MOM) or their extension, the Enterprise
Service Bus (ESB), enable loose coupling using mes-
sage queuing techniques. They implement the Event-
Driven Architecture pattern and provide neutral mes-
sage formats with which applications are able to ex-
change messages independently and asynchronously
using simple transport protocols. Furthermore, it en-
ables communication between different business units
with heterogeneous platforms and environments.
Our approach utilizes the WSO2 Enterprise Ser-
vice Bus (WSO2, 2014) together with the RabbitMQ
Message Broker (Pivotal Software, 2014) in order to
take advantage of concepts explained before. The
WSO2 Enterprise Service Bus is characterized by
WSO2 as the highest performance, lowest footprint,
and most interoperable SOA and integrated middle-
ware today. High connectivity and compliance to
standards are additional key benefits. Additionally,
it provides features for routing, mediation and trans-
formation of messages as well as monitoring, secu-
rity and service management features. The high-
performance and lightweight ESB is advertised as
highly available, scalable and stable. ESBs support
distributed environments. Each business partner in
a business network may have its own ESB running
or even multiple ESBs that are connected to a main
ESB of the business partner. The ESBs of each busi-
ness partner are in turn connected to the business
networks’ ESB. RabbitMQ is capable of propagating
messages sent between business partners. RabbitMQ
is described as robust, highly available, easy to use
and platform independent message broker. It supports
a variety of protocols and allows flexible routing and
clustering. In order to utilize Complex Event Process-
ing, Esper (EsperTech, 2014) is used. Esper supports
management of Data Windows for fine-grained event
expiry, event series analysis, and differentevent repre-
sentation types. Additionally, it is based on SQL stan-
dards, scalable and offers full API control. Addition-
ally, it provides its own Domain Specific Language
(DSL) called Event Processing Language (EPL). Ad-
ditionally, we’ve added access to the knowledge base
to combine the power of semantic web technologies
with complex event processing. Following the work
of (Schaaf et al., 2012), events and their relations to
each other can be enriched by semantics in order to
utilize knowledge inference about simple and com-
plex events as well as everything that is related to this
event, for instance the event publisher. Figure 3 illus-
trates the components of the middleware. Events that
are sent have to proceed the middleware, i.e. the event
channel. During this process, events and entities re-
lated to them are analysed by the Context Manager
and the complex event processor, i.e. the Esper CEP
Engine. The Context Manager encompass the Con-
tect Store and the Reasoner component, implemented
by Apache Jena. Additionally, the knowledge link en-
gine, which will be explained in more detail, handles
activation of modelled knowledge links.
Middleware
Event Channel
Rabbit MQ
WSO2 Enterprise Service Bus
Context Manager
Esper
Event Stream Connectors
Esper CEP Container
Historical Data
Context Store
Reasoner
Knowledge Link Engine
Figure 3: Middleware components.
WSO2 ESB, RabbitMQ and Esper are used to-
gether in the OSMOSE Project in order to allow com-
munication and analysis about events between the
real, digital and virtual world. Autonomous compo-
nents deploy services in a WSO2 ESB. Thereby, they
can react to events, call services or negotiate with
other autonomous components. The communication
is handled by RabbitMQ and analysed by the Esper
CEP Engine.
5 KNOWLEDGE LINKS
In the previous two main sections, we defined the
foundation of our approach for sustainable interop-
erability. We have defined a knowledge base con-
cept with which structuring of knowledge from dif-
ferent knowledge domains of business partners is en-
abled. Thus, business partners can rely on their data
structures that are connected to upper ontologies in
terms of compliance to the business networks’ generic
structures. The event drivenarchitecture approach en-
ables business partners to connect their enterprise ap-
plications to an ESB in their environment, while those
environments of all business partners can be con-
nected together to a business network environment by
utilizing ESB and message broker technologies.
MODELSWARD2015-3rdInternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
602
In this section, the application of knowledge links
is explained as a linkage of knowledge concepts be-
tween knowledge domains that runs on the event
driven architecture. Due to the fact that knowledge
links are used to link structured knowledge between
enterprise applications and business partners, they are
suitable to address interoperability concerns. Further-
more, they can be defined and maintained during run-
time of the business network. Thus, interoperability
can be sustained whenever business partners enter the
environment or change enterprise applications.
Knowledge links are connections between knowl-
edge concepts of different knowledge domains in a
black box manner. Inside the black box, operators
and transformations are applied to convert knowledge
from one or more source domains appropriately to the
target domain. They constitute a morphism of how
knowledge in one domain can be constructed by us-
ing knowledge from other domains.
Knowledge Links are intended to be modelled by
end users or domain experts. Modelling should be as
easy as possible to reduce the effort to share knowl-
edge to a minimum. Therefore, we’vechosen a graph-
ical way. In order to model knowledge links graphi-
cally, we implemented the Knowledge Link Config-
urator. This graphical modelling tool allows users
to import knowledge structures from different knowl-
edge domains and create knowledge links between
them. Creating a new knowledge link or loading an
existing one opens a graphical editor that allows drag-
ging and droppingof imported knowledgeconcepts as
well as operators from a palette. By drawing connec-
tions between dropped knowledge concepts and op-
erators, knowledge links can be defined. Currently,
we have defined a set of numeric (Sum, Subtraction,
Average...), text (Concat...) and boolean (AND, OR,
XOR...) operators. After the user finished with mod-
elling, knowledge links can be stored and uploaded to
the server. The uploaded knowledge links are further
translated and pushed to the Knowledge Link Engine
in the Context Store of the middleware.
In Figure 3, modelling of knowledge links with
the Knowledge Link Configurator is illustrated. On
the top right corner of the screenshot, the different
knowledge base structures are imported. On the lower
right side the knowledgelink definitions can be found.
Creating a new knowledge link definition will cre-
ate a rule sheet on which the graphical modelling is
done (squared area in the middle). The operators can
be found on the left side. In the knowledge link on
the screenshot, the set point of a machine part is sub-
tracted from it’s measurement in order to calculate the
deviation of the machine part. The measurement and
set point is taken from the real world machine and
machine part, more precisely, from the real world on-
tology. The deviation is a digital world concept and
thus described in the digital world ontology. The de-
scription of the knowledge link can be interpreted as
follows:
1. First two nodes, ’measurement’ and ’set point’:
Retrieve ’measurement’ and ’set point’.
2. Subtraction node, ’SUB’:
Subtract the set point from the measurement.
3. Equals node, ’==’:
Store the result ...
4. Last node, ’deviation’:
...as ’deviation’ in the ontology
The translation of knowledge links is done in two
steps. First, the knowledge concepts of the source do-
main are identified that participate in the knowledge
link. For all of those knowledge concepts, a listener in
the CEP Engine is registered that listens to changes of
the data behind that knowledge concept. It is assumed
that for all changes to data that are relevant for the
knowledge base semantic annotations are used and a
change of data results in an event that is sent to the
middleware having the knowledge concept as subject.
Thus, when knowledge concepts of the source do-
main are recognized in events that represent changes
of the data behind, the knowledge link will be exe-
cuted in order to keep background consistency. After
registering the listeners, an event automata is created
that handles requests on middleware level. When re-
questing knowledge that is defined by a knowledge
link, events for gathering knowledge from each do-
main and for calculation need to be executed.
For the latter case, we use business processes to
perform actions modelled with palette operators. Dur-
ing the second step of translation, the modelled be-
haviour is translated into process-files that are associ-
ated with the knowledge link. If a listener recognises
an event that is relevant for a knowledge link, the data
from knowledge bases that refers to the knowledge
link is requested by events and passed to the pro-
cess. The process is built from predefined building
blocks that correspond to the palette operators. For
instance, the ’Subtraction’ palette operator has a pre-
defined building block for business processes. When
used, the two values that should be subtracted are
loaded by event requests from the appropriate knowl-
edge bases and passed to the business process for cal-
culation. After the business process has finished, the
result is passed back to the event automata in order
to proceed with another business process or to push
the data in the appropriate knowledge base by using
events. The translated process file knows about the
execution order and the input parameters that are nec-
EnablingSustainableInteroperabilityforEnterpriseApplicationswithKnowledgeLinks
603
Figure 4: Knowledge Link Configurator.
essary. Because of the flexibility of the predefined
building blocks for processes, sequences of them can
be reused when similar parts of knowledge links are
modelled.
5.1 Model-Driven Architecture
Approach
Knowledge Links follow the Model-Driven Architec-
ture (MDA) (Object Management Group, 2014a) ap-
proach by separating business and application logic
from platform-specific software of dynamic business
networks. According to (Truyen, 2006), MDA distin-
guishes between three different MDA Models. The
Computation Independent Model (CIM) is also called
business model and describes system expectations,
i.e. the output of the system, independently from
implementation. It bridges the gap between domain
experts and information technologists. The Platform
Independent Model (PIM) abstracts out technical de-
tails by defining appropriate services and exhibiting a
sufficient degree of independence to enable mapping
to platforms. Contrary, the Platform Specific Model
(PSM) combines specifications of the PIM with plat-
form specific details.
From their demand to their specification, mod-
elling and translation, knowledge links undergo sev-
eral model transformations between MDA Models.
Figure 5 illustrates the different MDA Models and
the transformations that are explained below. Vertical
arrows represent CIM to PIM or PIM to PSM trans-
formations respectively. Horizontal arrows represent
CIM to CIM or rather PIM to PIM transformations.
In some cases, one-to-one transformations may not
be suitable. Instead, many-to-one, one-to-many or
even many-to-many transformations will be applied.
This is illustrated by multiple consecutively arranged
boxes as input or output for the arrows in Figure 5.
5.1.1 Computation Independent Model (CIM)
Models of the CIM are business requirements that
are expressed in natural language. They describe the
knowledge demand that business partners have. Fur-
ther, they can be transformed or consolidated to other,
higher-level CIM models to express the knowledge
demand inside the business network. These descrip-
tions are computation independent.
For instance, a quality manager of a business part-
ner inside a business network would like to know
about the deviation of a measurement from a product
component, which is manufactured by another busi-
ness partner, from its set point. The access to the
knowledge base could be restricted or the value that
is demanded is not comprehensible. Therefore, the
quality manager could express his or her demand in
natural language. Other business partners could in
MODELSWARD2015-3rdInternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
604
turn read the description and provide a solution for
the demand and further rewrite the description.
Up to now, there is no appropriate automated
methodology described to transform business require-
ments to PIM models. Thus, the transformation from
CIM to PIM has to be done manually, i.e. with the
Knowledge Link Configurator by knowledge link cre-
ation.
5.1.2 Platform Independent Model (PIM)
Knowledge links themselves are models of the PIM.
Their description is platform independent and can
be graphically modelled with the Knowledge Link
Configurator. Furthermore, they are translated into
other PIM models, the event automata and business
processes (PIM to PIM transformation), after storing
them on the platform. The transformation result is in-
dependent from enterprise applications dthat are im-
plemented by business partners. Business processes
and event descriptions are also platform independent
and could be used by various engines or middleware
components in different integrated systems.
By way of example, the quality manager in the
scenario described above uses the Knowledge Link
Configurator and creates a knowledge link between
the concept of deviation’ in the quality managers’
knowledge domain and the ’measurement’ and set
point’ concepts of a business partners’ knowledge do-
main. After uploading this knowledge link, the trans-
lation algorithm generates an event automata that lis-
tens to changes of ’measurement’ and ’set point’ val-
ues and triggers business processes that are generated
from building blocks if changes to these values hap-
pen or knowledge about the ’deviation’ is requested.
5.1.3 Platform Specific Model (PSM)
During knowledge link translation, the PIM models
are transformed to PSM models taking semantic an-
notations into account. Semantic annotations map the
knowledge concepts of ontologies to real data like
documents in different formats. Thus, knowledge
gathering depends on the underlying platform and en-
terprise application that business partners are using.
Additionally, code fragments like event listeners and
event triggers are generated in order to control the
knowledge and event flow of the knowledge link at
the middleware level. Business processes are trans-
lated and stored into specific process files that can be
executed by the business process execution engine.
In case of the quality manager that created and up-
loaded the ’deviation’ knowledge link to the middle-
ware, the business processes of the knowledge link is
provided with input from the semantically annotated
files that are maintained by the other business part-
ners’ enterprise applications. the event listeners and
triggers are created according to the middleware tech-
nology that is implemented.
Business Partner
Requirements
Business Partner
Requirements
Business Partner
Requirements
Business Partner
Requirements
Business Partner
Requirements
Business Partner
Requirements
Business Partner
Requirements
Business Partner
Requirements
Business Partner
Requirements
Business Partner
Requirements
Business Partner
Requirements
Business
Network
Requirement
Knowledge Link
Event Automata
Business
Processes
Platform Specific
Events/Listeners
Process Files
Real Data
CIM
PIM
PSM
Figure 5: Model Transformations of Knowledge Links.
6 SUMMARY AND CONCLUSION
We combined semantic web technologies, event-
drivenand model-drivenarchitecture approaches with
knowledge links to propose a solution for sustainable
interoperability in collaborative environments. Our
approach is based on three pillars: (1) a hierarchi-
cal and modular knowledge base concept utilizing
upper and modular ontologies, (2) an event-driven
and message-message oriented middleware based on
enterprise service busses that can be connected and
(3) knowledge links, a model-driven approach with
which domain ontologies can be linked in order to
achieve enterprise interoperability. Sustainable inter-
operability is enabled due to the possibility to create
and maintain knowledge links during runtime allow-
ing to react to changes in collaborative environments
by end users that are affected by these changes as they
appear. Middleware technologies and a tool to create
knowledge links have been presented. Consistency
between knowledge bases of business partners is war-
ranted automatically in the background. Obstacles are
reduced due to the ability of business partners to de-
fine their own knowledge structures.
We exemplifiedour approach with use cases of the
OSMOSE Project and will further evaluate our so-
lution in this Project. The software components of
the architecture presented in this paper are currently
prototypical and will be evaluated after integration
with the proof-of-concept scenarios of the OSMOSE
Project.
EnablingSustainableInteroperabilityforEnterpriseApplicationswithKnowledgeLinks
605
ACKNOWLEDGEMENTS
The research leading to these results has re-
ceived funding from the European Union Seventh
Framework Programme (FP7/2007-2013) under grant
agreement n
610905, the OSMOSE project.
REFERENCES
Agostinho, C., Sarraipa, J., Goncalves, D., and Jardim-
Goncalves, R. (2011). Tuple-based semantic and
structural mapping for a sustainable interoperability.
In Camarinha-Matos, L. M., editor, Technological In-
novation for Sustainability, volume 349 of IFIP Ad-
vances in Information and Communication Technol-
ogy, pages 45–56. Springer Berlin Heidelberg, Berlin,
Heidelberg.
ATHENA Project (2007). Da8.2: Guidelines and
best practices for applying the athena inter-
operability framework to support sme partici-
pation in digital ecosystems. http://www.asd-
ssg.org/html/ATHENA/Deliverables/Deliverables,
Accessed 17.11.2014.
Ben Abbs, S., Scheuermann, A., Meilender, T., and
D’Aquin, M. (2012). Characterizing modular ontolo-
gies. In 7th International Conference on Formal On-
tologies in Information Systems - FOIS 2012, pages
13–25, Graz, Austria.
Chen, D. and Daclin, N. (2006). Framework for enter-
prise interoperability. Interoperability for Enterprise
Software and Applications: Proceedings of the Work-
shops and the Doctorial Symposium of the Second
IFAC/IFIP I-ESA International Conference: EI2N,
WSI, IS-TSPQ 2006, pages 77–88.
Chen, D., Doumeingts, G., and Vernadat, F. (2008). Archi-
tectures for enterprise integration and interoperabil-
ity: Past, present and future. Computers in Industry,
59(7):647–659.
Chungoora, N., Young, R. I., Gunendran, G., Palmer, C.,
Usman, Z., Anjum, N. A., Cutting-Decelle, A.-F.,
Harding, J. A., and Case, K. (2013). A model-driven
ontology approach for manufacturing system interop-
erability and knowledge sharing. Computers in Indus-
try, 64(4):392–401.
Dassisti, M., Jardim-Goncalves, R., Molina, A., Noran, O.,
Panetto, H., and Zdravkovi´c, M. M. (2013). Sustain-
ability and interoperability: Two facets of the same
gold medal. In On the Move to Meaningful Inter-
net Systems: OTM 2013 Workshops, volume 8186 of
Lecture Notes in Computer Science, pages 250–261.
Springer Berlin Heidelberg, Berlin, Heidelberg.
EsperTech (2014). Esper - complex event processing.
http://esper.codehaus.org/, Accessed 30.11.2014.
Etzion, O. and Niblett, P. (2010). Event Processing in Ac-
tion. Manning Publications Co, Greenwich, CT, USA,
1st edition.
Felic, A., K¨onig-Ries, B., and Klein, M. (2014). Process-
oriented semantic knowledge management in product
lifecycle management. Procedia CIRP, 25(0):361
368. 8th International Conference on Digital Enter-
prise Technology - DET 2014 Disruptive Innovation
in Manufacturing Engineering towards the 4th Indus-
trial Revolution.
Foundation, A. S. (2014). Apache jena - home.
https://jena.apache.org/, Accessed 30.11.2014.
Fowler, M. (2003). Patterns of enterprise application
architecture. The Addison-Wesley signature series.
Addison-Wesley, Boston.
Grabs, T. and Lu, M. (2012). Measuring performance of
complex event processing systems. In Topics in Per-
formance Evaluation, Measurement and Characteri-
zation, pages 83–96. Springer.
Gruber, T. R. (1993). A translation approach to portable
ontology specifications. Knowledge Acquisition,
5(2):199–220.
Healy, M., Poli, R., and Kameas, A. (2010). Theory and
Applications of Ontology: Computer Applications.
Springer.
IDABC, Echipa, and D. G. Industry (2004). European inter-
operability framework for pan-european egovernment
services: Draft document as basis for eif 2.0. Euro-
pean Communities.
IEEE (1991). Ieee standard computer dictionary: A compi-
lation of ieee standard computer glossaries. IEEE Std
610, pages 1–217.
Jardim-Goncalves, R., Popplewell, K., and Grilo, A. (2012).
Sustainable interoperability: The future of internet
based industrial enterprises. Computers in Industry,
63(8):731–738.
Kadar, M., Muntean, M., Cretan, A., and Jardim-
Goncalves, R. (2013). A multi-agent based negoti-
ation system for re-establishing enterprise interoper-
ability in collaborative networked environments. In
2013 UKSim 15th International Conference on Com-
puter Modelling and Simulation (UKSim), pages 190–
195.
Lampathaki, F., Koussouris, S., Agostinho, C., Jardim-
Goncalves, R., Charalabidis, Y., and Psarras, J.
(2012). Infusing scientific foundations into enterprise
interoperability. Computers in Industry, 63(8):858–
866.
Luckham, D. (2002). The power of events, volume 204.
Addison-Wesley Reading.
Luckham, D. (2008). The power of events: An introduction
to complex event processing in distributed enterprise
systems. Springer.
McGovern, J., Sims, O., Jain, A., and Little, M. (2006).
Event-driven architecture. Enterprise Service Ori-
ented Architectures: Concepts, Challenges, Recom-
mendations, pages 317–355.
Michelson, B. M. (2006). Event-driven architecture
overview. Patricia Seybold Group, 2.
Ni, Y. and Fan, Y. (2008). Ontology based cross-domain
enterprises integration and interoperability. In IEEE
Congress on Services Part II (SERVICES-2), pages
133–140.
MODELSWARD2015-3rdInternationalConferenceonModel-DrivenEngineeringandSoftwareDevelopment
606
Niles, I. and Pease, A. (2001). Towards a standard upper
ontology. In Guarino, N., Smith, B., and Welty, C.,
editors, the international conference, pages 2–9.
Object Management Group (2014a). Model driven archi-
tecture (mda). http://www.omg.org/mda, Accessed
17.11.2014.
Object Management Group (2014b). Object management
group. http://www.omg.org/, Accessed 17.11.2014.
Oren, E., M¨oller, K. H., Scerri, S., Handschuh,
S., and Sintek, M. (2006). What are se-
mantic annotations? http://www.siegfried-
handschuh.net/pub/2006/whatissemannot2006.pdf,
Accessed 29.11.2014.
OSMOSE (2014). Osmose project - homepage.
http://www.osmose-project.eu/, Accessed 29.11.2014.
Peristeras, V. and Tarabanis, K. A. (2006). The connection,
communication, consolidation, collaboration interop-
erability framework (c4if) for information systems in-
teroperability. Ibis, 1(1):61–72.
Perrochon, L., Kasriel, S., and Luckham, D. C. (2001).
Managing event processing networks. Stanford Uni-
versity.
Pivotal Software, I. (2014). Rabbitmq - messaging that
just works. http://www.rabbitmq.com/, Accessed
30.11.2014.
Robins, D. (2010). Complex event processing. In Second
International Workshop on Education Technology and
Computer Science. Wuhan.
Schaaf, M., Grivas, S. G., Ackermann, D., Diekmann, A.,
Koschel, A., and Astrova, I. (2012). Semantic com-
plex event processing. Recent Researches in Applied
Information Science, pages 38–43.
Singh, Y. and Sood, M. (2009). Model driven architecture:
A perspective. In 2009 IEEE International Advance
Computing Conference (IACC 2009), pages 1644–
1652.
Stojanovic, N., Stojanovic, L., Anicic, D., Ma, J., Sen,
S., and St¨uhmer, R. (2011). Semantic complex
event reasoning—beyond complex event processing.
Springer.
Truyen, F. (2006). The fast guide to model driven architec-
ture. Cephas Consulting Corp.
Vernadat, F. B. (2007). Interoperable enterprise systems:
Principles, concepts, and methods. Annual Reviews in
Control, 31(1):137–145.
W3C (2004). Owl-s: Semantic markup for web ser-
vices. http://www.w3.org/Submission/OWL-S/, Ac-
cessed 13.12.2014.
W3C (2006). Ontology driven architectures
and potential uses of the semantic web
in systems and software engineering.
http://www.w3.org/2001/sw/BestPractices/SE/ODA/,
Accessed 17.11.2014.
W3C (2008). Sparql query language for rdf.
http://www.w3.org/TR/rdf-sparql-query/, Accessed
3.12.2014.
W3C (2012). Owl 2 web ontology lan-
guage document overview (second edition).
http://www.w3.org/TR/owl2-overview/, Accessed
3.12.2014.
W3C (2014). Rdf - semantic web standards.
http://www.w3.org/RDF/, Accessed 3.12.2014.
WSO2 (2014). Wso2 enterprise service bus.
http://wso2.com/products/enterprise-service-bus/,
Accessed 30.11.2014.
EnablingSustainableInteroperabilityforEnterpriseApplicationswithKnowledgeLinks
607