How to connect design thinking and cyber-physical systems:the s*IoT conceptual modelling approach

Michael WalchUniversity of Vienna

michael.walch@univie.ac.at

Dimitris KaragiannisUniversity of Viennadk@dke.univie.ac.at

Abstract

The alignment of enterprise models and informationsystems is a factor that influences the efficiencyof enterprise practices. Considering the changinglandscape in the age of the fourth industrial revolution,it is imperative that alignment methodologies areevolved with the progression of enterprise modelsand the transformation from information systems tocyber-physical systems (CPSs). This issue was dissectedin three layers – scenario layer, modelling layer, andrun-time environment. In this structure, design thinkingand CPSs were extended from the scenario layer and therun-time environment to the modelling layer. Focusingon the modelling layer, progress was made towardscomposing ”smart” models that innovate enterprisemodels according to novel influences from designthinking while abstracting from run-time environmentsthat CPS provide. The hypothesis was to considerthe automated transformation of knowledge as anaxle around which artifacts on the modelling layerrevolve. Based on this hypothesis, the modellinglayer was structured in a modelling hierarchy, inwhich a metamodel was defined using a metamodellingplatform. The metamodel is the direct model ofmodelling methods which were used to build ”smart”models that connect design thinking and CPSs.

1. Introduction

The digital transformation age is revolutionizingthe co-dependence of society and technology, as acomposition of new conceptual designs, modellingartifacts, and socio-technical systems emergesby embracing digital innovation for just abouteverything [1, 2]. In a business context, enterprisemodels1 and the means to build them lie at theheart of this revolution. They are the artifacts thatrealize conceptual designs while being put to use bysocio-technical systems. As the digital transformationage brings about a revolution of conceptual designs

and socio-technical systems, enterprise models have toadopt accordingly.

The transformation of innovation into businessvalue is not a straight forward task but rather awicked problem [4]. Design thinking is one approachto tackle this problem by conjuring and capturingconceptual designs in tangible artifacts, some of which– as illustrated in this paper – can be refined intoenterprise models by means of conceptual modelling.Standardized enterprise models are often not sufficientfor this task. Rather, agile modelling methodengineering can be applied to progress enterprise modelsin an innovative direction [5].

The development of enterprise models is one task ina business context. Another one is the operationalizationof enterprise models, which implies that all theknowledge that is recorded in enterprise models forhuman interpretation has to be put to use. For quitesome time now, information systems have supportedhumans in the task of putting enterprise models to use.To guarantee smooth enterprise operations, enterprisemodels have to be appropriately aligned with suitableinformation systems.

In the digital transformation age, the task of puttingenterprise models to use is shifted from humanstowards automated tasks performed by machines and inparticular by cyber-physical systems (CPSs) [6, 7]. Thecomplexity of this progress stems from two challengesof CPSs: their high variability at design-time andtheir complex dynamics at run-time. High variabilityat design-time is a result of applying design thinking& conceptual modelling to hypothetical applicationscenarios of CPSs, while complex dynamics at run-timeresults from the reality of executed CPS behaviour.

While design thinking, conceptual modelling, andagile modelling method engineering can provide meansto record conceptual designs in progressive enterprisemodels for human interpretation, the question is howto put these models to use by executing them in

1The term enterprise models is used in a broad sense that includesbusiness models on different organizational layers [3].

Proceedings of the 52nd Hawaii International Conference on System Sciences | 2019

URI: https://hdl.handle.net/10125/60161ISBN: 978-0-9981331-2-6(CC BY-NC-ND 4.0)

Page 7242

an automated manner. That is because currentlyfew enterprise models can be executed and of thosefar fewer are rewarding to use [8]. Consequently,research is required on how to reduce the human effortwhen enterprise models are put into operation. Inparticular, this paper asks the question how progressiveenterprise models can be aligned with CPSs, consideringthat such an alignment should enable the automatedexecution of said models. The problem is to connectthe decomposition of conceptual designs for businessinnovation and for hypothetical application scenarios ofCPSs with the abstraction of executed CPS behaviour.Furthermore, eliminating the dependence on humaninterpretation and action when enterprise models areput to use requires that the intelligence required toturn humanized knowledge into machine interpretableform has to be reproduced either by implementing itinherent to CPSs or by extending progressive enterprisemodels even further. The second option is the onefavoured in this paper which composes the necessarymethodological framework in Section 2. With respectto this framework, the state of the art is discussedin Section 3. Section 4 applies the methodologicalframework in a case study which involves the use ofa metamodelling platform to build the metamodel thatis necessary for connectivity between design thinkingand CPSs. The metamodel from Section 4 is the directmodel of modelling methods which are applied as toolsin laboratory experiments for validation purposes inSection 5. Section 6 concludes the topics.

2. The s*IoT Methodology

This section provides a methodology to conductapplied research on the introduced topic. Based on ascientific approach [9], the methodology is presentedby addressing its goals, conceptual framework,contributions, methods, and validation strategies. Aninstance of the methodology is applied to conduct thecase study that is covered in Section 4 and 5.

2.1. Goals

The overall goal of the s*IoT methodology is tofacilitate connectivity between design thinking andCPSs to enable synergistic effects in the digitaltransformation age. By considering enterprise modelsas first-class citizens, the goal can be broken downinto two subgoals. The first is to innovate andrecord conceptual designs from the business domainin progressive enterprise models, which are a subsetof conceptual models. The second is to align theresulting models and CPSs. Together, the subgoalsrequire the search for mutual artifacts (as seen in Fig. 1).

Run-Time Environment

Modelling Layer

Mutual Artefacts

Conceptual Modelling

Operatio-nalization

Requires

Methodology"Smart" Model

Modelling Method

2

1

Decomposition of Conceptual Designs

Knowledge-drivenEnrichment

Abstraction of Fun-ctional Capabilities

Scenario Layer

Meta-Model

ModelValue

CPSs@Run-Time

DesignThinking

3

Figure 1. Connectivity between design thinking and

CPSs relies on mutual artifacts on the modelling layer.

The immediate artifacts under scrutiny are progressiveenterprise models that contain the intelligence andabilities required for their operationalization. The term”smart” models has been coined for these kinds ofmodels. Modelling methods frame the search for modelsin general, which is true for ”smart” models as well.As the modelling methods for ”smart” models do notyet exist, this methodology aims to search for them.Metamodels are the direct models of modelling methodsand frame the search. Consequently, the goal of thes*IoT methodology is to search for (1) ”smart” modelsthat are co-constructed by design thinking and CPSs, (2)modelling methods for building ”smart” models, and (3)a model of the modelling method that covers invariantsof ”smart” models on the meta-level, i.e., a metamodel.The desired effect of these artifacts is to increase theefficiency of enterprise practices.

On a methodological level, the first subgoal iscovered by the practiced reality of design thinking,conceptual modelling, and agile modelling methodengineering, which is why it can be excluded forthe most part from a discussion in this section. Thesecond subgoal, however, requires further examination.The relation between models and CPSs is thatdifferent domain-specific modelling methods areapplied to understand, plan, and operate CPSs,2 whichresults in heterogeneous models [11, 12, 13]. Theresulting models separate into two general typesbased on established community practices: modelsthat decompose conceptual designs to tackle thehigh variability of CPS at design-time and modelsthat abstract functional capabilities to tackle thecomplex dynamics of CPS at run-time. For theclarity of presentation, these two types of modelsare addressed as conceptual models and execution

2Additional benefits of a model-based approach can be foundin [10].

Page 7243

environments respectively. Conceptual models aresymbolic representations of a problem domain, madeat design-time by humans for human use [14]; andexecution environments are practical reflections ofcomputation, networking, and physical processes,that provide a context for operationalization atrun-time [15]. Correspondingly, conceptual models arebuilt in conceptual modelling to represent design-timeaspects of CPSs [16]; and execution environments arebuilt using knowledge-driven enrichment to reflectrun-time aspects of CPS [17]. A qualitative distinctionbetween the two types of models results from theirdifferent purpose, which is reinforced by establishedcommunity practices. The problem is that visionaryopportunities could emerge from synergies betweenconceptual models and execution environments inthe digital transformation age. For example, orderpicking in the factory of the future could benefit fromconnectivity between an innovative process model thatdecomposes progressive concepts for this scenario fromdesign thinking and an execution environment like amicroservice portal that abstracts functional capabilitiesfrom an available pick-and-place machine, resultingin a potentially more agile, transparent, and effectiveautomation. Consequently, it is no longer justified toisolate conceptual designs and functional capabilitiesin different types of models. Rather, establishedcommunities have to bridge existing gaps to worktowards the hypothesis that connectivity between designthinking and CPSs is required in terms of design-timeand run-time aspects; and in terms of business conceptsand the behaviour, structure, and function of CPSs.

2.2. Conceptual Framework

Integrated with computer science, design scienceis researching designs of information systems inform of purposeful artifacts for information systemsengineering [19, 20]. Likewise, a specialization of thedesign science paradigm with a model-based approachis searching for artifacts on a modelling layer (as seen inFig. 2). Applied to the introduced issue of integratingdigital technologies into just about everything, new

Environment DSR in IS Knowledge Base

Evaluation

"Smart"

Models

Design Cycle

Rig

or C

ycle

Rele

van

ce C

ycle

• Application

Domain

• People

• Organizations

• Systems

• Problems &

Opportunities

• Foundations

• Artefacts

• Meta-Artefacts

• Expertise

• Technology

• References

Figure 2. Agility within s*IoT. Adopted from [18].

Do

main

Design Thinking

Cyber-Physical Systems

Scenario Layer

Modelling Layer

Run-Time Environment

Decomposition of

Conceptual Designs

Abstractio of Func-

tional Capabilities

So

ftware

In

frastr

uctu

re

Figure 3. The three layer s*IoT architecture.

and evolved model-based artifacts are required forconnectivity between design thinking and CPSs. Theinstantiation of the design science paradigm for theintroduced issue results in a three layer architecture(as seen in Fig. 3) which contains the scenario layer,the modelling layer, and the run-time environment.The conceptual framework for building model-basedartifacts is based on this three layer architecture.

The scenario layer covers conceptual designs thatare – by design thinking – transformed from mentalmodels into tangible artifacts. When design thinkingis applied in combination with conceptual modelling ina business context, a transformation from the scenariolayer to the modelling layer decomposes conceptualdesigns in enterprise models, which provides modelvalue for human stakeholders [21]. The necessaryenterprise models can be evolved by agile modellingmethod engineering. To put enterprise models touse, a transformation into an operation environment isrequired. When CPSs are considered, the operationenvironment structures itself into two categories:run-time environment and execution environment. Theformer frames the realization of CPSs and theirbehaviour into the non-deterministic physics of reality,of which the latter is a formal abstraction for machineinterpretation that can be thought of conceptually. Whenrelating enterprise models and operation environments,two methodological directions from systems modellingare relevant. The first is model-driven engineering,which continues the decomposition of model-basedconceptualizations to deploy CPSs. Thereby, CPSdesigns are derived in relation to enterprise modelsat design-time, which results in conceptual models ofCPS that are infused with (semi)-formal semantics toenable the (semi)-automatic deployment of run-timeenvironments. The second direction, which abstractsmodel-based conceptualization from CPS ecosystems,is lacking a decisive term. Nevertheless, the idea isthat execution environments, which provide means foroperationalization on the modelling layer, are linked toCPSs at run-time. The nature of this link is that of a

Page 7244

feedback loop between abstract categories in executionenvironments and their concrete physical realization inrun-time environments. One example for executionenvironments is found in service oriented architecturesthat define aggregated services for applications, basedon computation, networking, and physical processes thatsome run-time environment offers. The second directionis the one this methodology focuses on, as the feedbackloop enables a more agile composition of enterprisemodels, conceptual models of CPSs, and their operationenvironment.

In ”smart” models, the function and structureof the composition and its resulting behaviour isenabled by artificial intelligence technologies. Todesign & engineer ”smart” models, a meta-level viewconsiders the purposeful artifacts that result from designscience as systems under study by design science.Consequently, the modelling layer is structured ina modelling hierarchy. A transformation betweenartifacts on different levels of the modelling hierarchy ispossible using metamodel-based implementation, e.g.,metamodels are direct models of modelling methods,which in turn are used to build models [22].

2.3. Contributions

Possible contributions provide a point of departurefor research that plans to apply the s*IoT methodology.Thereby, connectivity between design thinking andCPSs is broken down into smaller issues worth solving.This subsection structures possible research questionsin three overall categories: (1) which research topicsbenefit from the methodology, (2) what is the natureof connectivity, and (3) how to facilitate research,education, and practice in a community.

The first category examines topics that can betackled by applying the s*IoT methodology. Thecommon theme of these topics is that a combinationof business innovation and technological innovationprovides immediate benefits but relies on knowledgeengineering for connectivity on the modelling layer.

• Knowledge Creation by Artificial Recognition– possible scenarios revolve around CPSs thatrecognize the real world by coordinating computerboards, IoT protocols, and sensors, which isenabled through knowledge engineering, e.g., byartificial neural networks, hidden Markov models,and support vector machines.

• Knowledge Execution through Reasoning &Planning – possible scenarios revolve around CPSsthat provide innovative services by controllingkinematics of actuators in difficult environments,which is enabled through knowledge engineering,

e.g., by ontologies, rule engines, and fuzzy logic.• Knowledge Transfer with (Eco-)System

Architectures – possible scenarios revolve aroundcommunication, negotiation, and collaborationbetween humans and heterogeneous CPSs, whichis enabled through knowledge engineering, e.g., bysemantic alignment, microservice frameworks, andmulti-agent systems.

• Knowledge Validity & Traceability – possiblescenarios revolve around humanized white-boxmodels that explain the behaviour or CPSs, whichis enabled through knowledge engineering, e.g., bydecision trees, evolutionary computation to invertneural networks, and data visualization.

The second category tries to understand thenature of connectivity between design thinkingand CPSs. By applying the s*IoT methodology toco-construct collective artifacts, the issue boils downto a composition of (1) design thinking & conceptualmodelling, which transforms conceptual designsfor innovative business scenarios and design-timeaspects of CPSs into conceptual models, and (2)conceptual modelling & knowledge-driven enrichment,which transforms atomic and also abstract, intricatefunctional capabilities of CPS run-time environmentsinto execution environments. Fig. 4 condenses theresulting problem of aligning conceptual models andexecution environments in ”smart” models. A closerlook unveils research questions regarding the nature ofconnectivity. In detail, connectivity is required betweeninnovative business scenarios, CPS designs from whichrequirements for CPS capabilities can be inferred,functional capabilities of CPSs, and their realization.However, alignment is facilitated on the modelling layerin a modelling hierarchy. The expected outcome ofthis are metamodels, modelling methods, and ”smart”models. To design & engineer artifacts in the modellinghierarchy, the following questions have to be answered.

• What is the common ontological structure formodels of conceptual designs and functional

Conceptual

Model

Execution

Environment

Line of

Alignment

Business DomainDesign Thinking in the

CPS Run-TimeEnvironment

Figure 4. The four quadrants for alignment.

Page 7245

capabilities? How is it realized in concretemetamodels and modelling methods?

• How are models that use concepts and capabilitiesrealized by ”smart” models? How efficient are themodelling methods at building ”smart” models?

• What tooling has to be provided to constructmetamodels, modelling methods, and ”smart”models? How can it be improved and distributed?

• Which aspects of conceptual models and executionenvironments should be connected in detail? Howis connectivity realized, e.g., is it done on-line oroff-line, synchronous or asynchronous, exogenousor endogenous?

• How can connectivity benefit from artificialintelligence technologies, knowledge representationschemes, and semantic services?

The third category is concerned with theinterdisciplinary competence and mentality betweenscientific and professional communities. The separationbetween conceptual models and executionenvironments is not only found in artifacts, but alsoin established community practices. The s*IoTmethodology is interested in enabling differentcommunities from different domains to cooperate.Cooperation is required to contribute in a businesscontext to a common understanding of design thinkingand CPSs. The importance of this pragmatic aspectcannot be underrated if the existing communities are toprevail and thrive in the digital transformation age.

2.4. The Decomposition and AbstractionMethods

To build ”smart” models in the three layerarchitecture that has been introduced as part of theconceptual framework, it is necessary to extend thescenario layer to the modelling layer, to extend run-timeenvironments to the modelling layer, and to connectboth in mutual artifacts. Similarly, a modellingmethod for combining the decomposition of conceptualdesigns and the abstraction of functional capabilities isnecessary. The ingredients of this combined modellingmethod are design thinking, conceptual modelling, andknowledge-driven enrichment (as seen in Fig. 5).

Design thinking & conceptual modelling starts as aprocess that is assigned to the scenario layer: creativeand educated minds conceive mental models of problemdomains and collaborate on them. These mental modelsare then decomposed at design time by the formalizationof conceptual designs in conceptual models, whichextends the scenario layer to the modelling layer.Conceptual models are structured by models of concepts

and models that use concepts [23]. The former representinvariants of problem domains, which is a prerequisitefor finding more specific artifacts of the latter type.

Conceptual modelling & knowledge-drivenenrichment starts as a process that is assigned tothe run-time environment: the concrete physicalcharacteristics and configurations of CPSs provide acontext for operationalization. The means necessary foroperationalization are abstracted by the compositionof functional capabilities in execution environments,which extends the run-time environment to themodelling layer. Execution environments determinemodels of functional capabilities which provide arun-time environment for models that use functionalcapabilities. As a consequence, specific functionalcapabilities make up more intricate, abstract ones. Thespecific end of the spectrum more directly relates to therun-time environment that CPSs provide and groundsthe model hierarchy in reality. Likewise, executionenvironments abstract the automated processing of data,information, and formal knowledge, by processors,networks, operating systems, and so on. Thereby, a lossof detail is desired to gain usability when modellingfunctional capabilities.

However, the results of decomposition andabstractions would be two types of models – conceptualmodels and execution environments – which wouldcontain knowledge in different form (humanized andmachine-executable). Instead, ”smart” models requirea transformation of knowledge on the modellinglayer to bridge the gap. To transform betweeninformal knowledge (which is found predominantly inconceptual models) and formal knowledge, information,and data (which is predominantly found in executionenvironments), artificial intelligence technologies haveto be embedded in modelling methods that combine thedecomposition and abstractions methods, in addition toknowledge engineering methods.

To enable the combination of decomposition,abstraction, and knowledge engineering in a singlemodelling method, metamodelling has to be considered,as it provides the means to engineer (hybrid) modellingmethods [24]. As mentioned, metamodels are directmodels of modelling methods, which in turn aredeployed as tools to build models [22]. The processthat leads to these artifacts is practiced in fivephases [25]: ”Create”, in which modelling requirementsare conceived; ”Design”, in which the ontological scopeof modelling requirements is constructed; ”Formalize”,in which the results from the previous phase arerepresented in a non-ambiguous and formal way bymetamodels; ”Develop”, in which the ontological scopeis combined with an operational scope in modelling

Page 7246

Conceptual Modelling &

Knowledge-Based Enrichment

Informal Knowledge

Figure 5. Decomposition and abstraction in a design

context

methods; and ”Deploy”, in which a modelling tool isevaluated for upcoming iterations. For increased agility,the design & engineering lifecycle can be extended withadditional feedback channels [25].

2.5. Validation Strategies

To validate artifacts that result from applying thes*IoT methodology, naturalistic ex post evaluation [26]is proposed. Thereby, connectivity between designthinking and CPSs – through the composition ofconceptual models and execution environments – isconducted in experiments. To conduct the experiments,a laboratory has to be set up that realizes a spacefor the scenario layer, modelling layer, and run-timeenvironment – to facilitate the conceptualization ofscenarios from the digital transformation age, theimplementation of mutual artifacts that fuse & evolveconceptual models and execution environments, andthe deployment of CPSs in run-time environments thatprovide means for operationalization. Focusing on themodelling layer, the experiments should revolve around”smart” models. Thereby, experimental evaluation isobtained for the applicability of a metamodel whenconstructing modelling methods (that are deployed astools) in the task of modelling method engineering;and for the applicability of modelling methods andtools when building ”smart” models in the task ofdomain-specific modelling.

3. Related Work

A body of work exists on standardized, butalso on domain-specific enterprise models that recordinnovative scenarios, mental models, and conceptualdesigns [16]. A body of work also exists on executionenvironments that control CPSs and give an account oftheir behaviour. Such work and a combination thereof is

situated on a model level in the modelling hierarchy andcannot be presented adequately by this paper. Rather,this section provides an overview of work done on themeta-level.

Two tracks can be identified on the meta-level. Inthe first track, metamodels are discussed for run-timeenvironments that contain models. In the second,metamodels are discussed for models that are executed.For the first track, prominent examples can be foundfor the topic of digital service discovery, where WSDLprovides a web service description [27], Hypercat andits RDF-based alternatives provide semantic servicecatalogs [28], and XMPP provides means for agents todetect their presence [29]. These examples are standardsfor models that are provided by run-time environmentsfor consumes that want to use services. These standardsdefine languages, procedures, and are foundations formechanisms & algorithms. Therefore, they operate onthe meta-level. In the second track, models are extendedby or transformed into execution environments. Thesemodels can be standard model types like UML whichare extended by fUML [30] for an executable subsetof UML or executed via code generation in modeldriven engineering [31]. These models can also bedomain-specific model types like SysML which requirededicated execution environments [32].

The open issue is to find a holistic structurein which the existing metamodels can be integratedas metamodel building blocks, while tackling theintroduced topic of connecting design thinking andCPSs. The issue is relevant as it is not feasible toconnect execution environments and conceptual modelson the meta-level, yet alone on the model-level, foreach type of progressive enterprise model and executionenvironment. Rather, a holistic metamodel for the issueat hand can be used to structure, integrate, and reusedifferent overlapping approaches on the matter.

4. Case Study Setup

The s*IoT methodology could be used in a widevariety of research projects. This claim has beenverified by a concrete collaboration between previouslyunrelated research projects. Using the introducedmethodology, a correlation between the projects couldbe found, which enabled deeper collaboration [33]. Inthis section, an instance of the methodology is appliedby focusing on a single research question, which isto find a metamodel for connectivity between designthinking and CPSs.

In particular, the idea is to model and bridge thecapability gap between (1) capability requirements forCPSs that can be inferred from enterprise models at

Page 7247

arrangesaccording to

de�nes grammar

Semantics

de�nes meaning

SemanticSchema

Syntax

SemanticMapping

connectsconsiders

Notation

ModellingLanguage

semantics

semanticdomain

syntaxnotation

modellinglanguage

de�nes visualization

visualizes

semanticmapping

describesmeaning of

de�nes way of language application

modelling

modellingmethod

mechanisms& algorithms

used for

used in

genericmechanisms& algorithms

hybridmechanisms& algorithms

speci�cmechanisms& algorithms

delivers

results

modelingprocedure

steps(design logic)

technique

Figure 6. Framework for modelling method

engineering [22].

design-time and (2) functional capabilities that reflectthe operation of CPSs at run-time. As a consequence,capability requirements and functional capabilitiesco-construct the axle around which enterprise modelsand execution environments revolve, which involves atransformation of knowledge using artificial intelligencetechnologies. The metaphorical axle has to be realizedon the modelling layer in the modelling hierarchy.

To build ”smart” models that feature the propertiesof the previous paragraph, a dedicated modellingmethod is required. To build modelling methods,a modelling method engineering framework has beenproposed (as seen in Fig. 6). The framework relies ona metamodel that is realized by a modelling language,modelling procedures, and mechanisms & algorithms.

The direct model of a modelling method is ametamodel. Metamodelling platforms implementa meta2model and enable the construction ofmetamodels. Some freely available choices formetamodelling platforms are ADOxx (www.adoxx.org),MetaEdit+ (www.metacase.com/mep), and EMF(www.eclipse.org/emf). However, when selectinga specific platform, one has to take into accounthow components of the platform support design &engineering [34]. Alternatively, a domain-specificmetamodelling language (MM-DSL) could be usedfor cross-platform support [18]. After the metamodelis defined in a metamodelling platform, it can beturned into a modelling method that includes toolsupport. Consequently, modelling methods for theproblem at hand can be derived from the metamodelthat is presented in this section. Furthermore, theresulting modelling methods can be applied as a tool indomain-specific modelling to build ”smart” models.

In Fig. 7-9, the metamodel of ”smart” models isdescribed in different levels of detail. The starting pointis the three layer architecture as discussed. Detailsare added on each level until a full fledged metamodelemerges that makes use of the elements in the discussedmodelling method engineering framework – modellinglanguage, modelling procedures, and mechanisms &algorithms.

Figure 7. This level (0) of the metamodel shows the

abstract concepts that are necessary for connectivity

between design thinking and CPSs.

By adding details to the metamodel, the natureof connectivity becomes more clear. Likewise, theextension of design thinking and CPSs to the modellinglayer becomes more clear, up to the point where itcan be supported by automation. The same is true forconnectivity between conceptual models and executionenvironments. For example, conceptual models canbe further decomposed into hybrid models that are acomposition of domain models, enterprise models, andscenario models; while execution environments furtherabstract CPSs into configuration models, capabilitymodels, and CPS class models. Connectivity then boilsdown to multi-agent models that instantiate CPS classmodels and negotiate between the functional capabilitiesthey offer and the requirements for capabilities fromhybrid models, which can be done using artificialintelligence technologies.

A more detailed version of the metamodel isnot suitable for presentation within this paper dueto the size of the illustration (it can be accessedat www.omilab.org/sIoT). At some point, details thatare specific to a metamodelling platform have to beadded. This process is shown in the validation section,where the ADOxx metamodelling platform is used tospecialize the metamodel. The results are modellingmethods and tools that enable experiments that allow usto show how design thinking and CPSs can be connectedconceptually and in practice.

5. Case Study Validation within OMiLAB

The previous section set up a metamodel of ”smart”models, which is the foundation for the case study that isdiscussed in this section. In the case study, the validationstrategy of the s*IoT methodology is instantiated andapplied to validate the metamodel, which requires toconducting experiments in a laboratory. Thereby,naturalistic ex post validation is looking at how

Page 7248

Figure 8. This level (1) of the metamodel shows the concepts that connect conceptual models and execution

environments.

artifacts of a research progress perform. Likewise,the metamodel of ”smart” models and implicitly alsothe s*IoT methodology are validated by the perceivedsuccess when corresponding modelling methods andtools are used to conduct experiments with ”smart”models. Both, laboratory and experiments, are describedin this section.

The laboratory for this study is provided as part ofthe OMiLAB non-profit organization. The organizationis a global network of local research, application, andeducation spaces. The individual spaces are realizedby different collaborators like research groups andindustry partners, each focusing on a topic surroundingmetamodelling. OMiLAB Vienna hosts different

Figure 9. This level (2) and more detailed levels of the metamodel can be enriched with details specific to

metamodelling platforms.

Page 7249

Table 1. Summary from 40 experiments.

Moveable CPSs Stationary CPSs

Drive Fly Walk Arms Sense ARS

Industry 4.0 2 2 6 1EMO EGI EMO EIO

Recreation 3 1 1 4 3 1ESR ENO ES ECGO EI ITV

Auton. Driving 11 1 1EIFN E IT

Smart Assistant 1 1 5IN ES EGO

Constraint Satisfaction (C), Fuzzy Logic (F), Genetic Algorithms (G), Image Recognition (I),Multi-Agent System (M), Neural Network (N), Object Tracking (T), Ontology (O), Rule

Engine (E), Speech Recognition (R), Speech Synthesis (S), Virtual Reality (V)

spaces, one of them being the Digital Product Space.One part of this space is the OMiRob laboratory, whichis a physical and virtual realization of the introducedthree layer architecture. The scenarios layer is realizedby a socio-technical system that decomposes conceptualdesigns semi-automatically into conceptual models. Themodelling layer is built around the ADOxx communityand embraces the use of the ADOxx metamodellingplatform for building modelling methods. The run-timeenvironment consists of CPSs in form of different typesof robots of varying complexity.

Within the OMiRob laboratory, experimentswere conducted (some of which are detailed onhttp://austria.omilab.org/psm/omirob). The invariantin these experiments was to build a modellingmethod that covers a building block of the introducedmetamodel. Thereby, an important aspect was tomove the source of intelligent behaviour away fromthe run-time environment, where minimal, atomic,and ”assembler-like” capabilities were sufficient,towards ”smart” models. However, the task wasnot to implement instances of intelligent behaviouron the model level, but to provide it as part of amodelling method that a metamodel building block isthe direct model of. The idea is to collect a repositoryof metamodelling building blocks that modellingmethod engineers can use as they seem fit for buildingdomain-specific modelling methods for ”smart” modelsthat can be put to use in an automated manner.

The experiments themselves are based on innovativescenarios in the digital transformation age and therun-time environment that is provided by OMiRob. Anoverview is provided in Table 1, which groups theexperiments in categories for scenarios and CPSs. Italso list artificial intelligence technologies that wereused. Examples are the use of robotic arms and selfdriving cars for a delivery on demand scenario, the useof humanoid robots as smart assistants in a health carescenario, or the use of quadcopters in a recreationaltourism scenario. The modelling methods that were

developed to conduct the experiments decomposedconceptual designs, abstracted functional capabilities,and employ artificial intelligence technologies in”smart” models. To do so, they refined a part ofthe introduced metamodel with additional detail interms of modelling language, modelling procedures,and mechanisms & algorithms until it was possibleto deploy a modelling tool using ADOxx. Thesuccessful application of this modelling tool verifies thecorresponding metamodel building block. In the samemanner, the s*IoT methodology is verified as well.

6. Conclusion

Overall, this study provides general and systematictheories by discussing the s*IoT methodology, ametamodel of ”smart” models, and proof-of-conceptmodelling methods that compose and refine metamodelbuilding blocks in tools that individuals can use as theyseem fit to understand and influence the co-dependenceof society and technology by building ”smart” modelsthat connect design thinking and CPSs. Connectivitywas achieved on the modelling layer in laboratoryexperiments by utilizing decomposition, abstraction,and artificial intelligence technologies to composeprogressive enterprise models, which extend designthinking, and execution environments, which extendrun-time environments of CPSs. The insights thatwere gained from the experiments suggest that thereemerging potential of machine learning and formalmathematical methods could provide yet to be exploredbenefits on the modelling layer for the issue at hand.Future work aims to develop a modelling method notonly for specific metamodel building blocks, but forthe integrated metamodel of ”smart” models. Theresult will be called the s*IoT modelling method. Theanticipated benefit is that this modelling method couldbe used in a wide variety of experiments and realworld applications by people unfamiliar with modellingmethod engineering.

References

[1] J. Gray and B. Rumpe, “Models for the digitaltransformation,” Software & Systems Modeling, vol. 16,no. 2, pp. 307–308, 2017.

[2] K. Smolander, M. Rossi, and S. Pekkola,“Infrastructures, integration and architecting duringand after digital transformation,” in Proceedings ofJSOS’17, (Buenos Aires, Argentina), pp. 23–26, 2017.

[3] H. Osterle and D. Blessing, “Ansatze des BusinessEngineering,” HMD, vol. 42, no. 241, pp. 7–17, 2005.

[4] T. Kelly, The Art of Innovation: Lessons in Creativityfrom IDEO, America’s Leading Design Firm. New York:Broadway Business, 2001.

Page 7250

[5] D. Karagiannis, “Agile modeling method engineering,”in Proceedings of PCI’15, (New York, NY, USA),pp. 5–10, 2015.

[6] J. Lee, B. Bagheri, and H.-A. Kao, “A cyber-physicalsystems architecture for industry 4.0-basedmanufacturing systems,” Manufacturing Letters,vol. 3, pp. 18 – 23, 2015.

[7] A. Al-Fuqaha, M. Guizani, M. Mohammadi,M. Aledhari, and M. Ayyash, “Internet of Things:A survey on enabling technologies, protocols, andapplications,” IEEE Communications Surveys Tutorials,vol. 17, no. 4, pp. 2347–2376, 2015.

[8] B. Thalheim, “Conceptual model notions - a matter ofcontroversy: Conceptual modelling and its lacunas,” inSpecial Issue on Conceptual Modelling in Honour ofHeinrich C. Mayr (J. Michael, C. Steinberger, V. A.Shekhovtsov, S. Ranasinghe, and F. A. Machot, eds.),pp. 9–27, Hagen: EMISA, 2018.

[9] J. A. Maxwell, Qualitative Research Design: AnInteractive Approach, vol. 41 of Applied Social ResearchMethods. London: Sage Publications, 2012.

[10] B. Selic, “The pragmatics of model-drivendevelopment,” IEEE Softw., vol. 20, no. 5, pp. 19–25,2003.

[11] A. R. da Silva, “Model-driven engineering: A surveysupported by the unified conceptual model,” ComputerLanguages, Systems & Structures, vol. 43, pp. 139 – 155,2015.

[12] C. Schlegel, T. Haßler, A. Lotz, and A. Steck,“Robotic software systems: From code-driven tomodel-driven designs,” in Proceedings of ICAR’09,(Munich, Germany), pp. 1–8, 2009.

[13] B.-H. Schlingloff, “Cyber-physical systemsengineering,” in Engineering Trustworthy SoftwareSystems: First International School, SETSS 2014,Tutorial Lectures (Z. Liu and Z. Zhang, eds.),pp. 256–289, Cham: Springer, 2016.

[14] J. Mylopoulos, “Conceptual modelling and Telos,”Conceptual Modelling, Databases, and CASE: anIntegrated View of Information System Development,pp. 49–68, 1992.

[15] A. Burns and A. Wellings, Real-Time Systems andProgramming Languages: Ada, Real-Time Java andC/Real-Time POSIX (4th Edition). Longmain: AddisonWesley, 2005.

[16] D. Karagiannis, H. C. Mayr, and J. Mylopoulos,Domain-Specific Conceptual Modeling: Concepts,Methods and Tools. Cham: Springer, 2016.

[17] M. Walch, “Knowledge-driven enrichment ofcyber-physical systems for industrial applicationsusing the KbR modelling approach,” in Proceedings ofICA’17, (Beijing, China), pp. 84–89, 2017.

[18] N. Visic and D. Karagiannis, “Developing conceptualmodeling tools using a DSL,” in Proceedings ofKSEM’14, (Sibiu, Romania), pp. 162–173, 2014.

[19] H. Osterle, J. Becker, U. Frank, T. Hess, D. Karagiannis,H. Krcmar, P. Loos, P. Mertens, A. Oberweis, and E. J.Sinz, “Memorandum on design-oriented informationsystems research,” European Journal of InformationSystems, vol. 20, no. 1, pp. 7–10, 2011.

[20] A. Hevner and S. Chatterjee, Design Research inInformation Systems: Theory and Practice, vol. 22 ofIntegrated Series in Information Systems. New York:Springer, 2010.

[21] E. Miron, C. Muck, D. Karagiannis, and D. Gotzinger,“Transforming storyboards into diagrammatic models,”in Proceedings of Diagrams’18, (Edinburgh, UK),pp. 770–773, 2018.

[22] D. Karagiannis and H. Kuhn, “Metamodellingplatforms,” in Proceedings of EC-Web’02,(Aix-en-Provence, France), pp. 182–182, 2002.

[23] D. Gotzinger, E.-T. Miron, and F. Staffel, “OMiLAB:An open collaborative environment for modeling methodengineering,” in Domain-Specific Conceptual Modeling:Concepts, Methods and Tools (D. Karagiannis, H. C.Mayr, and J. Mylopoulos, eds.), pp. 55–76, Cham:Springer, 2016.

[24] D. Karagiannis and M. Schwab, “An engineeringapproach for the design of hybrid modelling methods,”in Enterprise Information Systems: 14th InternationalConference, ICEIS 2012, Wroclaw, Poland, June 28- July 1, 2012, Revised Selected Papers (J. Cordeiro,L. A. Maciaszek, and J. Filipe, eds.), pp. 3–17, Berlin,Heidelberg: Springer, 2013.

[25] N. Efendioglu, R. Woitsch, and W. Utz, “A toolboxsupporting agile modelling method engineering:Adoxx.org modelling method conceptualizationenvironment,” in Proceedings of PoEM’16, (Skovde,Sweden), pp. 317–325, 2016.

[26] J. Venable, J. Pries-Heje, and R. Baskerville, “Acomprehensive framework for evaluation in designscience research,” in Proceedings of DESRIST’12, (LasVegas, NV, USA), pp. 423–438, 2012.

[27] F. Curbera, M. Duftler, R. Khalaf, W. Nagy, N. Mukhi,and S. Weerawarana, “Unraveling the web services web:an introduction to SOAP, WSDL, and UDDI,” IEEEInternet Computing, vol. 6, no. 2, pp. 86–93, 2002.

[28] I. Tachmazidis, S. Batsakis, J. Davies, A. Duke,M. Vallati, G. Antoniou, and S. S. Clarke, “Ahypercat-enabled semantic internet of things data hub:Technical report,” CoRR, vol. abs/1703.00391, 2017.

[29] M. E. Gregori, J. P. Camara, and G. A. Bada,“A jabber-based multi-agent system platform,”in Proceedings of AAMAS’06, (Hakodate, Japan),pp. 1282–1284, 2006.

[30] T. Mayerhofer, P. Langer, M. Wimmer, and G. Kappel,“xMOF: Executable DSMLs based on fUML,” inProceedings of SLE’13, (Indianapolis, IN, USA),pp. 56–75, 2013.

[31] G. Devai, M. Karacsony, B. Nemeth, R. Kitlei, andT. Kozsik, “UML model execution via code generation,”in Proceedings of EXE@MoDELS’15, (Ottawa, Canada),pp. 9–15, 2015.

[32] S. Wolny, “A runtime model for SysML,” DoctoralCollege Cyber-Physical Production Systems, pp. 43–48,2017.

[33] T. Morita, N. Kashiwagi, A. Yorozu, M. Walch,H. Suzuki, D. Karagiannis, and T. Yamaguchi, “Practiceof multi-robot teahouse based on printeps and evaluationof service quality,” in Proceedings of COMPSAC’18,(Tokyo, Japan), pp. 147–152, 2018.

[34] H.-G. Fill and D. Karagiannis, “On the conceptualisationof modelling methods using the ADOxx meta modellingplatform,” Enterprise Modelling and InformationSystems Architectures - An International Journal, vol. 8,no. 1, pp. 4–25, 2013.

Page 7251