semantical markov logic network for distributed...

Research ArticleSemantical Markov Logic Network for Distributed Reasoning inCyber-Physical Systems

Abdul-Wahid Mohammed12 Yang Xu1 Ming Liu1 and Haixiao Hu1

1School of Computer Science and Engineering University of Electronic Science and Technology of China ChengduSichuan 611731 China2School of Engineering University for Development Studies Tamale Ghana

Correspondence should be addressed to Yang Xu xuyanguestceducn

Received 28 May 2016 Accepted 27 November 2016 Published 24 January 2017

Academic Editor Jose A Somolinos

Copyright copy 2017 Abdul-Wahid Mohammed et al This is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

The challenges associated with developing accurate models for cyber-physical systems are attributable to the intrinsic concurrentand heterogeneous computations of these systems Even though reasoning based on interconnected domain specific ontologiesshows promise in enhancing modularity and joint functionality modelling it has become necessary to build interoperable cyber-physical systems due to the growing pervasiveness of these systems In this paper we propose a semantically oriented distributedreasoning architecture for cyber-physical systems This model accomplishes reasoning through a combination of heterogeneousmodels of computation Using the flexibility of semantic agents as a formal representation for heterogeneous computationalplatforms we define autonomous and intelligent agent-based reasoning procedure for distributed cyber-physical systems Sensornetworks underpin the semantic capabilities of this architecture and semantic reasoning based onMarkov logic networks is adoptedto address uncertainty inmodelling To illustrate feasibility of this approach we present aMarkov logic based semantic eventmodelfor cyber-physical systems and discuss a case study of event handling and processing in a smart home

1 Introduction

Cyber-physical systems (CPSs) [1ndash3] represent a novelresearch field with high prospects for providing synergybetween the digital and physical worlds These systemsconsist of interconnected components which collaborativelyexecute tasks in order to bridge the gap between the digitaland physical worlds [4] With the growing complexity oftasks due to the combination of CPSs and Internet of Things(IoT) [5 6] however CPSs have become distributed and it isnecessary developing interoperable CPSs capable of enablingtimely delivery of services In this way CPSs become real-time distributed with multiscale dynamics and networkingfor efficient dependable safe and secure management ofmonitoring and control of objects in the physical domain [7]

Following the integration of CPSs and IoT innovativeconcepts and approaches such as service-oriented archi-tecture (SOA) [8 9] collaborative systems [10] and cloudcomputing [11] have become apparent in the development

of CPSs As devices in CPSs are required to interoperate atboth cyber and physical scales service-oriented CPSs [1213] so far look promising By coupling the need for CPSsto interact with real world objects in real-time with opti-mal resource consumption however using only the service-oriented approach is not enough to realise comprehensivedistributed real-time CPSs that take into account the stronginterdependencies between the cyber and physical com-ponents Therefore context-aware agent-based approach ismore appealing since diverse attributes can be encapsulatedwithin agents and distributed pervasive computing withbetter coordination and interoperability among autonomousand heterogeneous agents is achievable Essentially context-awareness is indispensable in CPSs since sensing resourcediscovery adaptation and augmentation are the key driversof this novel technology [14 15] Additionally with thechanging dynamics of distributed CPSs domains it is nat-ural that partial observability is inherent in CPSs [16 17]Using ontology as underlying semantic technology therefore

HindawiJournal of SensorsVolume 2017 Article ID 4259652 15 pageshttpsdoiorg10115520174259652

2 Journal of Sensors

requires uncertainty modelling techniques towards goodmodel performance

This paper proposes a context-based multiagent archi-tecture for distributed reasoning in CPSs Sensor networksin distributed physical environments provide the semanticcapabilities of this model and semantically annotated low-level contextual information merges with domain knowl-edge in a reasoning engine Derived implicit knowledgethrough semantic reasoning and together with the annotateddata enables distributed software agents operating in thecyberspace to provide decision support for actuation infor-mation Each agent is capable of interacting with the physicalenvironment and sharing some principal commonalities withthe other agents As such the agents are capable of expos-ing consuming and even at times processing collaborativeservices targeting laid down system goals To incorporateuncertainty in modelling semantic reasoning on this modelincorporates the inferential power of Markov logic networks(MLN) [18] into event recognition to reduce inferential andcomputational loads Finally we discuss a case study of asmart home as a CPS and results of our experiments showfeasibility of this approach in modelling concurrent events inCPSs

In summary we describe in this paper our initial workin CPSs which overcomes some of the major limitations inthis research fieldWe provide four main contributions Firstwe propose a multiagent architecture that can bridge the gapbetween the operations of the cyber and physical componentsof CPSs Second we describe a procedure that can be used todynamically compose high-level system goals and underlyingcriteria from low-level contextual information Third weintroduce a smart home ontology which incorporates humanactors in the physical space of CPSs as computing entitiesFinally we present a methodology based on MLN for eventrecognition in CPSs

The rest of the paper is organised as follows we presentthe state of the art and the study background and otherpreliminary information in Sections 2 and 3 respectivelySection 4 gives an agent model for CPSs followed by ourframework for distributed reasoning in CPSs in Section 5uncertainty-based event recognition in CPSs using MLN ispresented in Section 6 experiments and discussions of resultsare presented in Section 7 and finally we conclude andpropose future research in Section 8

2 Related Work

A coherent distributed reasoning architecture is one thatachieves an interoperable CPS to cope with the requirementsof the physical and cyber components Because of the lackof sound theoretical foundation for CPSs currently [19] mostapproaches successfullymodel either the physical component[20] or the cyber component [21] but not both Salientstudies towards comprehensive models for CPSs that takeinto account both the cyber and physical components include[8 22 23] which use service-oriented computing to achieveinteroperable CPSs Service-oriented approach alone how-ever is not suitable for modelling real-time distributed CPSs

with multiscale dynamics and networking In this regard anagent-based modelling is more appropriate for distributedcomplex systems such as CPSs [24 25]

Because of the underlying sensor network in CPSssemantic agent technologies are closely associated with ourapproach As provided in [26] a semantic agent technologyhas been used to describe a battlefield information systemwhich uses information fusion processes to dynamicallyintegrate sensor networks towards real-time context-basedreasoning To enable scalable sensor information access anarchitecture and programming model for service-orientedsensor information platform has also been proposed [27]This approach leverages an ontological abstraction of infor-mation to optimise use of resources in collecting storingand processing data Timeliness and concurrency in dis-tributed processing environments can also be enhanced usingautonomous software agents As such use of autonomoussemantic agents as a new software paradigm for distributedcomputing environments has been proposed [28]

Obviously the complex dynamics of CPSs coupled withthe need to properly represent embedded computing andcommunication capabilities motivate the use of semanticsand distributed agents towards interoperable CPSs As canbe found in [29] a multiagent model for CPSs in which adistributed semantic agent model augments data acquisitionprocess with ontological intelligence has been proposedThismodel however provides no procedure for reasoning locallyabout individual components and globally about system-wide properties But such semantics for instance eventrecognition is very critical for distributed real-time CPSsand can essentially specify components of systems in termsof interfaces and observations [7 30] Additionally ontologyforming the underlying layer in the semantic agent-basedmodel primarily supports certainty-based reasoning and assuch requires techniques to address uncertainty inmodelling

In our approach therefore we augment the semanticmultiagent architecture with a robust reasoning mechanismwhich can support both certainty and uncertainty-basedreasoning To promote concurrency and timeliness in theoperations of CPSs MLN is adopted as an uncertainty mod-elling framework and can compactly represent heterogeneouscomputations using a common set of rules Essentially thisachieves a reasoning procedure forCPSs by leveraging advan-tages of both ontology and probabilistic graphical modelsto model both complexity and uncertainty which reduceslimitations of both the ontology and probabilistic graphicalmodels

3 Background and Preliminaries

We provide in this section aspects of the problem of real-timedistributed reasoning in CPSsMarkov logic and smart homeas a case study of a CPS

31 Problem Description The growing pervasiveness of CPSsfurther keeps the cyber and physical components apart andseparately managing these components would not allow usto realise fully the benefits of these systems Appropriate

Journal of Sensors 3

techniques that would allow interoperability between thesecomponents are essential so that monitoring and actuationcan be invoked remotely In this regard standardised inter-faces that can achieve interoperable CPSs are desirable andshould be guided by the following

(1) Defining an architectural framework that supportsinteroperability and distributed reasoning in CPSs

(2) Applying semantics to explicitly represent contextualinformation and providing an efficient data storagemechanism

(3) Providing distributed reasoning procedure that cre-ates more autonomy and intelligence in operations ofCPSs

(4) Incorporating uncertainty into modelling

Following above challenges and the ability of semanticagents to be discoverable and autonomous we pursue seman-tically oriented techniques in which ontological intelligenceis used to address the problem of uncertainty-based real-timedistributed computing in CPSs

32 Markov Logic Network MLN [18] is an interface layer inartificial intelligence which defines a first-order knowledgebase in terms of first-order logic formulae and associatedweights Given a set of constants depicting objects of adomain MLN defines a ground Markov network whichrepresents a probability distribution over possible worldsEachworld basically represents assignment of truth values toall ground atoms and this distribution is a log-linear modelgiven by

119875 (119909) = 1119885 expsum119894

119908119894119899119894 (119909) = 1119885prod119894

120601119894 (119909119894)119899119894(119909) (1)

where 119899119894(119909) is the amount of true grounding of first-orderformula 119865119894 in 119909 119909119894 is the 119894th state of the predicates appearingin each formula 119908119894 is the weight of 119865119894 120601119894(119909119894) = 119890119908119894 isthe potential function of each clique in the ground Markovnetwork and 119885 is a normalizing constant called the partitionfunction Each weight indicates the strength of a constraintthat a formula represents and is directly proportional to thedifference in the log probability between a world that satisfiesthe formula and one that does not

Due to the varying number of constants that can representthe same knowledge base either in part or full MLN allowsthe same formulae to be applicable under all circumstancesand can be viewed as a template for constructing Markovnetworks In this way different sets of constants can producedifferent groundMarkov networks using a common underly-ing MLNThis ideally is suitable for domains such as CPSswhere the task of reasoning requires combining separatereasoning chunks which need to be processed independently

33 Case Study As intelligence in the home gets moresophisticated intelligent interconnection of distributed con-sumer hardware such as consoles smart home servers andsmart phones running diverse functionalities like assistive

health care and home automation constitutes a CPS Todemonstrate the heterogeneity concurrency and sensitiv-ity to timing of CPSs a case study of temperature eventrecognition is considered This presents a scenario of usingan ontology-based model to achieve an interoperable CPSthat leverages a common event recognition model acrossdifferent layers Specifically using a single computation ofa car system events pertaining to temperature conditionsof the car userrsquos home and that of the car engine canbe achieved Typically this computational platform mustsupport concurrent processing of events since temperatureevents for the home and the car can concur

The fact that CPSs need to be sensitive is a challengingtask in respect of false positives that can arise if uncertaintyis not well managed Apart from noisy sensor informationand incomplete domain knowledge being primary sources ofuncertainty environmental factors are also potential sourcesof uncertainty that cannot be ignored in CPSs For instance atemperature event that considers optimal resource consump-tion in a smart home may trigger opening of windows on acool sunny day for comfort of the homeThis strategy thoughall things being equal sounds ideal but environmentalfactors such as air population and external noise can presenta trade-off between comfortability and minimising resourceconsumption in the home As such the effectiveness of CPSswill be much appreciated if uncertainty which is unavoidablein nature is well managed in these systems

4 Agent Network for CPSs

Building on the foundation of mobile agent network [31] wedefine amultiagent system residing in the cyberspace ofCPSsThis model considers all aspects of agentsrsquo communicationand operation including issues relating to performance ofmultiagent systems and CPSs

41 Cyber Agent Model A cyber agent model is definedby a triple CAM = ⟨119860119863119873⟩ where 119860 represents acommunity of agents depicting distributed computationalenvironments in a CPS119863 denotes specific domains of agentsrsquoservices and 119873 defines networks of agents operating inthe cyberspace Essentially agents in this definition can bestationary and mobile so as to suit the changing dynamicsof CPSs domains In this way an agent 119860 119894 in a multiagentsystem which is defined by 119860 = (1198601 119860119899) representsa specific computational platform and can perform tasksallowed by its domain This specifies somewhat autonomy inthe operations of these agents and tasks can be encapsulatedas agentsrsquo capability from the viewpoint of functionality andperformance As such interactions between agents providethe needed communication and cooperation to bridge the gapbetween the operations of the cyber and physical componentsof CPSs

Contextual reasoning paradigm [32] in which each agentdepends on a domain specific ontology and can link withother agents through semantic mapping makes this designdistributed Since each computational platform provides afunctionality for service execution combining these multiple


Request1 Request2

Agent1 Agent2 Agent3

S11 S12 S21 S31 S32 S33 S34Sc

Figure 1 Service assignment graph for agentsrsquo domains

ontologies through semantic mapping allows us to definejoint functionalities that can be used in complex task oper-ations For a nonempty set 119868 of indices used to identify agentsassociated with domain specific ontologies 119863119894119894isin119868 we definethe joint functionality of our multiagent system as a set ofcross-layer services 119869119878 = 1198781 119878119899 Thus a set of servicesindexed by each domain is defined as S119894119894isin119868 Intuitively S119894represents a service formalisation of the 119894th ontology

Wemust recognise that each service 119878119894 can be provided byone or more agents To avoid conflicts in accessing servicesan agent definition that explicitly specifies computationalplatforms and services provided is critical In this way wemake services distinct by encapsulating each agent definitionas a property of a computational platform in a CPS using thetriple 119886119892119890119899119905119894 = ⟨119899119886119898119890119894 119889119890119901119897119900119910119898119890119899119905119894 119904119890119903V119894119888119890119894⟩ As we can seethis definition of an agent apart from a property describinga given computation also provides information about agentsrsquoservices and domains for those services A service domain isspecified by a deployment property which can be a physicaladdress of a distributed environment in CPSs For instancegiven a set of agentsrsquo services domains 119863 = 1198631 119863119899 theset of services provided by agent 119886119892119890119899119905119894 can be described by119878119894 = 1198781198941 119878119894119899 and each service within this domain can beinvoked using the pair 119886119889119889119903119890119904119904119894 = ⟨119878119894 119886119892119890119899119905119894⟩ This meanswithin the cyberspace the interactions between these agentsdefine an undirected graph 119873 = (119863 119864) where 119864 denotes anedge between any two domains of agents with overlappingfunctionalities

42 Agent Cooperation in CPSs For the multiagent systemto ensure high fidelity between the physical and cybercomponents each agent domainmust trigger requests whichrepresent implicit knowledge inferred from the domainrsquoslow-level contextual information Typically these requestsmay involve complex tasks that can exceed capacity of asingle operational domain and may require cross-domainservices aggregation In this case agents operating in thecyberspace through their interactions can communicate andcooperatively assume specific roles to execute tasks

The idea that agentsrsquo domains are distributed and canform a graph with overlapping functionality presents acomplex network in which fast information sharing becomesnecessary for large agent teams [33] Relying on informationimportance as a determinant for service assignment basedon performance requirements of agentsrsquo domains is ideal forfast information sharing and parallel execution of services In

this regard service requests to agents can be either domainspecific or across multiple domains As shown in Figure 1a special case in the assignment of services to agents is forexample Request1 when a request is domain specific Inthis case all services can be deemed mutually exclusive andcommunication and coordination among agents become lessimportant However when services overlap or a given requestrequires a combination of cross-domain services we face aproblem of a multiagent autonomy denoted by Request2 Butinterestingly instead of employing a planning agent for theassignment of services in this case agents can combine theirinherent intelligence with domain knowledge to negotiate forservices in our design

5 Distributed Reasoning in CPSs

Distributed computing systems inCPSs can employ pervasivecomputing techniques to provide autonomous interoperablecomputational elements that can be described discoveredand orchestrated within and across different layers [15] Inthis regard ontology-oriented modelling of contexts offersa lot of advantages [34] We present next a semantic agent-based architecture for CPSs Additionally semantics forformulation of high-level complex tasks with underlyingcriteria from low-level contextual information is discussed

51 Semantic Multiagent Architecture for CPSs This is abroker-centric multiagent architecture which can supportcross-layer service collaboration in CPSs As shown in Fig-ure 2 this architecture captures into perspective the mod-elling concerns of CPSs raised in [35] Key components of thisarchitecture are data management module context ontologymodule semantic reasoning engine and a confederation ofsemantic agents Detailed descriptions of these componentsare provided in the following subsections

(1) Data Management Module The key functionalities of thismodule are collection and transmission of data to storageareas Raw context data are acquired from distributed sensornetworks in the physical domain and the heterogeneityof these data requires semantic markups that applicationscan easily understand Through semantic annotation thecontext data are transformed into semantic markups thatcan link to external definitions through unique URIs ofontology instances For example the semantic annotation ofa temperature sensor data can be as in Listing 1


ltDevice rdf ID=ldquoampobs TempSensor2rdquogtltowl sameAs rdf resource =ldquoampsmh TempSensorR2rdquogtltsmh observedPhenomenom rdf resource =ldquoampsmh Temperaturerdquogtltsmh qtyValue rdf datatype =ldquoampxsd floatrdquogt245ltsmh qtyValuegtltsmh qtyUnit rdf resource =ldquosmh Celsiusrdquogtltsmh timeStampgt2015ndash07ndash28 165230lttimeStampgtltDevicegtListing 1

Networks

Data management module

Semantic annotation

Domain ontology module

Ontology modelling and processing

Semantic reasoning engineReasoning rules

Markov logic networksHierarchical task networks

Bayesian networksAgents

A

CB

Data

Semantic repository Context ontology repository

Cyberspace

Monitoring information

Actuation information

Sensors amp devices domain

Physical space

repository

Figure 2 Multiagent context architecture for CPSs

As can be seen clearly each annotation contains a uniqueURI such as 119904119898ℎ 1198791198901198981199011198781198901198991199041199001199032 of an ontological instancedescription of the observed phenomenon measured valueof the phenomenon unit of measurement used and thetimestamp for the observation Apart from the timestampand measurement value which are XML Schema datatypesthe other attributes are resources which exist in an externaldefinition

Data storage in this architecture is achieved at two levelsusing different databases that must interoperate with eachother In the first instance the raw sensor data is stored insuch a way that it can be efficiently maintained and exportedby disparate sources Secondly after the raw sensor data issemantically annotated a storage mechanism that supportssemantics is required for storing this annotated dataThus inour architecture the annotated data are stored in a repository


as ontology instances and associated properties thatmachinescan easily interpret This repository is updated whenever anew context event occurs and can be augmented with LinkedData techniques to support semantic data integration at theinstance level

In Linked Data research [36 37] dereferencing theURIs of resources through HTTP protocol can be exploitedto incrementally obtain the description of resources Tofurther support integration of semantic data distributivelyOWL axiom owlsameAs has been used successfully at theinstance level in Linked Data research With such successin a distributed setting this same technique can be adoptedinto CPSs for semantic integration Thus semantic markupsin this paper are linked to external definitions using theowlsameAs axiom As can be seen in the example above theuse of this axiom shows that the two URIs refer to the sameinstance thereby providing a mapping between the semanticrepository and the context ontology repository

(2) Context Ontology Module Ontology modelling and pro-cessing occur in this module In CPSs computing entitiesand services of distributed intelligent environments canbe grouped together forming service-oriented ecosystemsContexts in these domains can share some concepts incommon even though their detailed properties can differ sig-nificantly Instead of completely modelling all contexts acrossdifferent domains the objective here is to model contextsusing a base ontology and a domain specific ontology Entitiesof the base ontology are extensible basic common conceptsacross different environmentsThe domain specific ontologyhowever represents only those concepts that uniquely existin each domain

Specifically in a smart home domain the most funda-mental concepts we have identified as extensible nodes of thebase ontology are user deployment service and computingentity When these entities are linked together a skeletonof contextual entity is formed which allows context-baseddata acquisition Figure 3 shows the context ontology modelwe propose for a smart home domain in this paper Thebase ontology which is extended by both a smart home anda smart institution illustrates the advantage of knowledgereuse using ontology In both cases base entities such asRoom and AdhocService are extended to meet specificationsof the application domain For instance whilst we canspecify bedroom and living room in a home an institu-tion can have rooms such as lecture room and conferenceroom

It is important to note that human as a computing entityin this model is a novel contribution that is ideal for design ofCPSs This essentially extends the service-oriented paradigmto incorporate human services in CPSs towards transmutingsystem components and behavioural practices [38] Speciallywith our design the role of humans as both actors andsources of contextual information in the physical domaincan be explicitly represented and allows social awarenessto be incorporated into CPSs In this view CPSs are wellpositioned to provide emotional intelligence [39] that willrespond appropriately to people and situations

ltReasonedData rdf ID=ldquoampobs TempSensor2rdquogtltsmh domainURI rdf resource =ldquoampsmh room104rdquogtltsmh criticalLevel rdf resource =ldquosmh Highrdquogtltsmh alertgtFire in buildingltsmh qtyValuegtltReasonedDatagtListing 2

(3) Semantic Reasoning EngineThis is the central componentof our design in which high-level implicit knowledge can beinferred from sensed contextual information Semanticallyannotated data and the context ontology are aggregated intoa coherent model that semantic agents and physical objectscan share Both certainty-based and uncertainty models canbe supported by this design But the focus of this research isuncertain decision support in CPSs Specifically uncertainty-based reasoning about resources and events and dynamicformation of collaborative cross-layer services given high-level system goals with underlying criteria are the focus of thisdesign

It is worth noting that putting the reasoned data touse by the semantic agents requires a data structure thatcan easily integrate with the underlying semantic repositoryof this representation Specifically the semantics of thereasoned data when used to feed these agents must specifyamong other needs the referenced domain of the inferredknowledgeThe argument here is that since CPSs domains arehighly distributed in nature agents can cooperate effectivelyacross different domains to achieve better computationalintelligence if we semantically specify domains of inferredknowledge in the reasoned data In line with this paperrsquosobjective such a data structure can allow easy mashup ofresources to solve complex problems Thus Listing 2 is anexample of a semantic markup of a reasoned data

As we can see this example basically demonstratesthat aside the domain of interest referenced using domain-URI other elements fitting a given scenario are allowedAmong these additions that is also unavoidable is the high-level knowledge obtained through semantic reasoning Thisknowledge in this example is specified using the element alertand mostly what users get as prompts Obviously such a datastructure combinedwith the domain knowledge can allow thesemantic agents gain enhanced computing capabilities

(4) Semantic Agents The semantic agents are distributedalgorithms executing on multiple distributed computingentities in the physical space To provide decision supportfor actuation information these agents merge semanticallyannotated data with the reasoned data By describing theseagents as a community they are ostensibly the control pointof this architecture and can advertise their services in thereasoner through interactions and semantic reasoningThuseach agentrsquos behaviour is well suited for its environmentand such behaviours are well suited for resource discoverythrough semantic reasoning


Base ontology

Home ontology

Institution ontology

Extends

Device

Agent

Human

Network

Thing

ContextEntity

ComputingEntity

Deployment

Address

Building

GeoCoordinates

Location

Passageway

Room

Specification

Service

AdhocService

RoutineService

User

WaterService

LibraryService

Thing

Institution

AdhocService

RegistrationService

RoutineService

Room

BathRoom

LectureRoom

ConferenceRoom

Laboratory

Room

BathRoom

Thing

Home

AdhocService

AssistiveHealth

RoutineService

Bedroom

ChildrenBedroom

GuestBedroom

MasterBedroom

Dining

Kitchen

LaundryRoom

LivingRoom

Extends

Figure 3 Smart home context model for CPSs

52 Dynamic Composition of High-Level Complex TasksFacilitation of dynamic formation of collaborative servicestowards execution of complex tasks requires elicitation ofhigh-level complex services from low-level informationThisprocess can guarantee better quality CPSs and overcomescommon engineering design flaws to provide right actuationinformation for the needs of physical objects For instancethrough low-level information we can compose a task suchas put off the fire as an event for handling fire outbreak ina CPS environment However this particular task unlikesome tasks requires complex functionality and needs to bedecomposed into primitive level tasks which can then beserviced by specific resources This is a challenging process

and therefore requires a dedicated framework on how tofigure out complex functionality from low-level contexts

As shown in Figure 4 this approach is motivated by theestablished actuation relationship between the cyberspaceand distributed CPS environments For clarity of represen-tation the physical environment is categorised into usageand context environments The usage environment describesprocesses performed by semantic agents and how systemscan achieve tasks in the environment Because each agentrsquosbehaviour best suits its environment the usage environmentostensibly contains specific objectives which describe theactivities of agents towards useful output The context envi-ronment as specified by the context acquisition process is the


Physical environment

Context environment

Usage environment

CyberspaceM

onito

ring

Actu

atio

n

Figure 4The relationships between a distributed CPS environmentand the cyberspace

source of domain knowledge which underpins the semanticcapabilities of this approach

In the distributed setting of CPSs it is required thatthe cyberspace provides specifications that can addressrequirements of the physical environment We can see fromFigure 2 that these requirements in the cyberspace aremodels and processes that control objects in the physicalenvironment and vice versa This brings to the fore issuesabout domain-driven and user-driven requirements Sincethe domain-driven requirements which fundamentally holdthe underlying semantics of this approach have been dis-cussed in the previous section our focus now is the user-driven requirements These requirements are enshrined inthe actuation information and form the underlying ideatowards the elicitation of high-level complex services fromthe domain-driven requirements Therefore understandingthe relationship that the actuation information establishesbetween the cyberspace and the physical environment isessential towards mapping contextual information to high-level complex services

High-level composite context can be derived from low-level contextual information through the semantic reasonerof Figure 2 As shown in Figure 5 the logic flow of thereasoning engine of this architecture consists of three mainfunctional blocks models filter and composer Reasoneddata from reasoning models are passed through a filteringprocess in which statements are categorised based on apredefined set of rules All statements through this filterare either categorised as executable or nonexecutable but notboth An executable statement represents a single servicephenomenon such as high temperature which can directlybe executed either remotely or centrally by reducing thehomersquos temperature through an air-conditioner But when astatement requires aggregation of services in order to achieveits objective it is filtered to be nonexecutable An example of

Models Filter

Composer

Nonexecutable

Executable

Semantic reasoning engine

Figure 5 Logic flow of the semantic reasoner of Figure 2

Input inOutput out(1) if executable then(2) 119900119906119905 larr 119904119905119886119905119890119898119890119899119905(3) else(4) 119894119899 larr 119875119897119886119899119899119894119899119892119875119903119900119887119897119890119898(119904119905119886119905119890119898119890119899119905)(5) goto step 1(6) end if(7) return out

Algorithm 1 Algorithm for composition and processing of com-posite context

a nonexecutable statement is when sensors detect smoke andhigh temperature in a building and the reasoned informationisfire outbreak in building Obviously an appropriate action inthis case is to put off the fire which requires different serviceswrapped as capabilities of different physical objects in thisapproach In this regard this reasoned information requiresfurther processing in the form of composition of appropriateservices and constraints so that tasks can be appropriatelyscheduled among computing entities

Whilst the executable statements directly feed the seman-tic agents the nonexecutable statements are transformedinto high-level composite contexts for further processingthrough the composer Following the objective of this paperthese statements are composed into high-level tasks withunderlying criteria Specifically this stage generates an HTNplanning problem which is passed as an input to the reasonerfor automatic composition of collaborative services based onAlgorithm 1 As we can see the filtering process creates acycle of execution of information whenever a nonexecutablestatement is encountered From lines 4 and 5 a nonexecutablestatement results in a planning problem consisting of tasksand constraints Generation of the planning problem isan instance of Algorithm 2 which takes a nonexecutablestatement as an inputWith this transformation collaborativeservices can then be formed using a specified model and candirectly feed the semantic agents To support the needs oftime-criticality in CPSs this design promotes efficient use ofcomputational resources by ensuring that all nonexecutablestatements experience one-off processing In essence the


Input inOutput tasks preconditions(1) generate output(2) return out

Algorithm 2 Planning problem

set of rules provided in the model block ensures that allcomposite tasks are reduced to primitive tasks which can bedirectly executed

6 Event Recognition Using Markov Logic

Context-based events are central in initiating activities inCPSs and naturally specify real-time demand responsivenessof systemsrsquo components in terms of interfaces and observa-tions [39] From the viewpoint of ourmultiagent architectureevents can stimulate services of one or more agents in thenetwork and it is therefore important to detect events ofpredefined operations that are desirable to both systems andusers of CPSs

An event ontology is required to augment the proposedcontext ontology in the previous section towards event-based reasoning However ontology in its classical formcurrently cannot represent and reason under uncertainty Inview of good modelling practice towards best performanceof CPSs we adopt MLN based event recognition to addressuncertainty whilst keeping the structure of the underlyingontology intact This exploits the view of MLN as a templatefor Markov networks so that only a part of OWL rulesapplicable to events are considered in themodel constructionOne obvious advantage is in the compact representation ofmodel complexity which can guarantee incorporation of richdomain knowledge for high sensitivity and good concurrentprocessing of events

MLN allows existing knowledge bases in first-order logicto incorporate uncertainty in knowledge representation byadding weights to logic formulae Since first-order logicunderlines the fundamental theory of ontology OWL rulesfor knowledge discovery can therefore be transformed intoMLN weighted formulae towards event recognition

61 Rules for Event Recognition OWL rules form the under-lying logical framework of our event recognition process andthe semantics we provide holistically capture the domainrsquosinterest phenomena in the rules As shown in Figure 6an event is described by a tuple ⟨119888119900119898119901119900119899119890119899119905 119891119906119899119888119905119894119900119899⟩which denotes event components and event semantic functions[40] We use event components to compose heterogeneoussensor data to form contextual information that matcheventsrsquo requirements Thus 119888119900119898119901119900119899119890119899119905119894 = 1199001198941 119900119894119899where 119900119894119895 represents observations driving the occurrenceof an event To be able to propagate logical constraintsof events in rules we use semantics of event functions⟨119878119905119900119901 119862ℎ119886119899119892119890 119862119900119898119901119886119903119894119904119900119899⟩ to distinguish between cate-gories of events With this specification Stop is a predicate

Device

Room Car

Deployment Event

Component Function

Consists

Value

Output

PropertyPropert

y

Sub property

Trig

ers

Figure 6 Context-based event model in CPSs

we use to express action of an event that changes spontaneousstates of natural phenomena Thus the predicate 119878119905119900119901(119909 119910)defined by

119860119888119905119894V119890 (119910) and (119860119888119905119894119900119899 (119909 119910) or 119860119891119891119890119888119905 (119909 119910)) 997904rArr not119860119888119905119894V119890 (119910) (2)

indicates that the state 119910 changes when acted upon by anaction of event 119909 In some cases of CPSs the event mayhave effects on the state as 119860119891119891119890119888119905(119909 119910) indicates In a firescenario for example this predicate ensures the right eventinvocation that will put off the fire We must note that thesame predicate cannot be applied to routine events such asputting off the air-conditioner Unlike in the fire case wherebythe new state of affairs after putting off the fire may beperpetual devices are only eligible for temporal state changesAs such we use the unary predicate 119862ℎ119886119899119892119890(119909) as the eventfunction for achieving temporal state changes for events inCPSs 119862ℎ119886119899119892119890(119909) is defined as

exist119910 (119909 = 119910 997904rArr 119909 = 119910) (3)

where state119909 changes to new state119910 using three subfunctionsSpecifically a state change can be rise in degree of some-thing using the predicate 119868119899119888119903119890119886119904119890(119909) reduction in degreeof something using predicate 119863119890119888119903119890119886119904119890(119909) and togglingbetween on and off modes of devices using the predicate119878119908119894119905119888ℎ(119909) All these subfunctions operate by conditioning thecurrent state against the new state For instance to increasethe temperature of a room using an air-conditioner thesemantics of the predicate 119868119899119888119903119890119886119904119890(119909) is defined as

exist119909 119878119905119886119905119906119904 (119909) and exist (119909 119910)119881119886119897119906119890 (119910) 997904rArrexist (gt 119909 119910)119881119886119897119906119890 (119910) (4)

to indicate a change in status value of the air-conditionerwhen 119910 is greater than 119909 Obviously this predicate like theother two requires semantics that compares the current andnew states in the change process

The Comparison predicate is used to define con-ditions that describe state changes We consider Less-Than(119909 119910) LessThanEqual(119909 119910) GreaterThan(119909 119910) Grea-terThanEqual(119909 119910) and Equal(119909 119910) as the predicates forconditions for state changes For example the semantics for119871119890119904119904119879ℎ119886119899(119909 119910) is defined as

exist119888 119862119900119899119889 (119888) and 119881119886119897119906119890 (119909) = 119888 and 119881119886119897119906119890 (119910) lt 119888 (5)


hasDeployment(TSLR)

Room(LR) Device(TS) hasValue(TS34)

Output(HT) greaterThanEq(3422)

Figure 7 A section of ground Markov network constructed from the MLN for event recognition

to express the condition that the value of 119909 is 119888 and thevalue of 119909 is less than the value of 119910 Therefore eventrecognition based on component structures and semanticfunctions provide logical operations in formulae that formgood basis for Markov logic based event recognition

62 Translation of Rules into MLN The first step towardsconversion of OWL rules into MLN requires transformationof OWL rules into first-order logic formulae As providedin [41] OWL classes and properties respectively representunary and binary predicates in first-order logic and can becombined using logical connectives to form atomic formulaeFor example the first-order logic translation of a class Roomis 119877119900119900119898(119909) where 119909 denotes instances of the given class Fora property hasDeployment the equivalent first-order logicformula is

ℎ119886119904119863119890119901119897119900119910119898119890119899119905 (119909 119910) 997904rArr 119863119890V119894119888119890 (119909) and 119877119900119900119898 (119910) forall119909 119910 (6)

where 119909 and 119910 respectively denote the domain and rangeclasses of this property Axiomatization of classes and prop-erty restrictions can also be translated into first-order logicFor instance rdfssubClassOf axiom can be translated intofirst-order logic as

119877119900119900119898 (119909) 997904rArr 119861119890119889119903119900119900119898 (119909) forall119909 (7)

to indicate that bedRoom is a subclass of RoomOn this basiswe can use logical connectives to compose the first-orderlogic formula of the concept class Device and its property as

119863119890V119894119888119890 (119909) and ℎ119886119904119863119890119901119897119900119910119898119890119899119905 (119909 119910) 997904rArr 119877119900119900119898 (119910)forall119909 119910 (8)

Interestingly the Alchemy [41] tool for MLN providesbuilt-in functions that simplify the translation of logical con-ditions into MLN For instance the predicate LessThan(119909 119910)in MLN is simply represented using the internal predicate ofAlchemy lessThan(119894119899119905 119894119899119905) Basically this predicate tests if thefirst argument is less than the second argument

Once we obtain the first-order logic translation of rulesthe MLN is achieved by adding weights to each formula The

MLN together with a set of constants define a groundMarkovnetwork on which probabilistic reasoning can be performedFigure 7 shows a section of the groundMarkov network basedon theMLNof our case study for event recognition Aswe cansee in this figure links exist between any two ground termsappearing together in the same formula in the MLN Hencegiven a MLN and a set of constants arbitrary queries such asthe conditional probability that a formula holds given anotherformula in the MLN can be addressed

63 Fuzzy Markov Logic Network We recognise that theaxiomatic notion of probability as presented in the lastsubsection is incapable of dealing with vague information inknowledge This becomes apparent in MLN when multival-ued clauses are encountered and this presents a challengebeyond the classical notion of MLN In this view we providea fuzzy notion of Markov logic called fuzzy MLN in whichinference to queries requires the inferencemachinery of fuzzylogic

The basic idea serving as a point of departure in fuzzyMLN lies in the fact that a formula 119865 in first-order logic canbe viewed as a collection of elastic constraints which restrictthe weights119882 associatedwith each grounding of its terms Toachieve this we define a fuzzy membership function in termsof weights and ground terms of MLN clauses and obtain anextension of MLN to fuzzy MLN as

119865MLN = (120583 119865 997888rarr 119882) (9)

This represents a mapping of a set 119865 of grounded first-orderlogic formula into a set of MLN weights 119882 The idea thatdifferent constants refer to different objects in MLN and aformula can contain more than one ground clause allowsfor separate assignments of weights to each ground clause inMLN Essentially this achieves fuzzy membership functionsmapping ground MLN clauses into an ordered set of fuzzypairs from whichMLN inferences can be performed Clearlythis fuzzy set is completely determined by the set of tuples

119865119888 = (119888 120583 (119888)) | 119888 isin 119862 (10)

denoting the assignment of weights to each grounding of 119865for a set of constants 119862

As shown in Figure 8 the membership function in fuzzyMLN assuming without loss of generality that all weights are


Very cold Cold Hot

Very hot

24 26 320

w1

w2w3

120583

Figure 8 Membership function of fuzzy MLN

positive is a representation of the magnitude of participationof the weight of each ground term as an inputThis associatesdifferent weights with the same formula for different groundterms and defines functional overlaps between these groundterms which determines outcomes of rules As we can seein this figure a typical case of an overlap is the temperaturevalue 26∘C which presents a case of a multivalued clause fordifferent grounding of the same formula This value exists inthe interval of the minimum criterion of the two fuzzy setsdefined by cold and hot

120583119888119900119897119889capℎ119900119905 = min (120583119888119900119897119889 120583ℎ119900119905) (11)

where 120583119888119900119897119889 and 120583ℎ119900119905 respectively define the membershipfunctions of the states cold and hot Intuitively this can bedescribed as to what degree a cold temperature 26∘C is hotObviously knowledge about true state of this temperaturevalue smacks of vagueness and can be efficiently interpretedas a fuzzy constraint on a collection of ground terms

In fuzzyMLN unlike in classical fuzzy logic the situationdescribed above is easily handled by specifying groundclauses involving both cases in a training set As Figure 9depicts this defines a ground Markov network in which thetwo outcomes are conditionally independent given the inputEach dotted circle in this figure indicates a weighted groundclause of the same formula and either clause is a complementof the other which in the absence of one clause defines theclassical notion of MLN This means learning the weight 1199081as shown in Figure 8 in this particular case produces twovalues that define the fuzzy set for reasoning Thus fuzzyMLN leveraging these weights and together with a set ofconstants defines a ground Markov network which can bereasoned upon using the inference machinery of fuzzy logicwithout employing any formal fuzzy logic semantics

7 Results and Discussion

In this section we present and discuss results of eventrecognition under uncertainty in a smart home as a CPSTheconditions for all experiments were designed to test perfor-mance of this approach using key intrinsic requirements ofCPSs such as sensitivity to timing and concurrency In thisregard thrust of our analyses bothers mainly about precision

as a measure of sensitivity to timing of occurrence of singleand concurrent events in CPSs

To incorporate uncertainty into modelling for our exper-iments we considered an MLN based event model rooted inthe OWL ontology of this paperrsquos case study This ontologycaptures into perspective semantics of our event model andthe key properties include

(i) hasDeployment which relates the concept Device toconcepts Room and Engine

(ii) hasValue which relates the concept Device to a datavalue

(iii) hasOutput which relates a data value of a device to itssemantic interpretation using the concept Output

(iv) hasEvent which relates the concept Output to theconcept event

Essentially Room and Engine which are subclasses ofLocation define heterogeneous computational platforms fordevices using the deployment property Consequently eventsassociated with different platforms denote effects of interpre-tations of devicesrsquo values on those platforms and are stored inthe ontology as type event

For compact modelling towards expedited processing ofevents OWL rules provided partial specification of domainconcepts relevant for the construction of the MLN eventmodel With the ability of OWL to support heterogeneousprocesses we used the same set of rules to represent differentcomputational platforms in the MLN Specifically compu-tations in relation to a homersquos indoor comfort index andoperational safety condition of a carrsquos engine were consideredto be two likely synchronous events that a single computationcan represent The ostensible need for a single computationfor distributed environments in CPSs can be seen in a caseof driving towards home whereby the computation of thecarrsquos console monitors both the home condition and theengine temperature of the car In this way a distributedsensor network of the home and the vehicle engine providescontextual information for the computational intelligenceWith the deployment property of devices in the underlyingontology contextual information can be accurately filteredand applied according to domain specifications

As shown in Table 1 five events were defined to rep-resent the two distributed environments considered in ourexperiments Essentially the heterogeneity in these eventsis enshrined in the different conditions pertaining to tem-perature measurements in these environments For instancewhilst we can specify a normal room temperature to be inthe range 21∘Cndash27∘C a normal operating temperature rangeof an engine is 180∘Cndash205∘C which is high temperature in acase of the home and way beyond limits of human survivaltemperature Clearly these two cases represent vagueness inknowledge as both denote a normal temperature and it istherefore important disambiguating between heterogeneoussensed information in modelling In this regard our eventmodel can be described as a composite model designedto precede any recognition process with precise knowledgediscovery Aside event recognition this same model canbe used to perform semantic reasoning towards mashup of


hasDeployment(TSLR)


Output(HT)

Output(N)

Figure 9 An example of ground Markov network of a fuzzy MLN

Table 1 Event specification for home and engine domains

Domain Value range Inferred knowledge Event

HomeLess than 21∘C Low temperature Increase temperature21∘Cndash27∘C Normal temperature No event

Greater than 27∘C High temperature Reduce temperature

Engine 180∘Cndash205∘C Normal temperature Normal operationGreater than 205∘C Overheat Switch off engine

Prec

ision

07

06

05

04

Number of constants in evidence1 10 20 30 40 50

TS = 100

TS = 500

TS = 1000TS = 1500

Figure 10 Precision of a single event recognition with increasingevidence

resources For instance temperature sensors deployed at anyof the two distributed environments in our case study can beinferred with this model

We evaluated the performance of this model by consid-ering both single event and multiple event recognition tasksIn either case we varied the training set of the MLN from100 constants to 1500 constants As we can see in Figure 10the precision of a single event recognition improves as moreground terms are introduced into the training set Lookingdown the column from left to right we will notice that the

effect of the number of constants in the training set stabilisesat some point We found this development interesting inour preliminary analysis because if we consider the columnsrepresenting the number of constants 100 and 500 or columnsrepresenting the number of constants 1000 and 1500 for theevent of a single constant as evidence one may be tempted toconclude that any two training sets differing by 500 constantsgive approximately the same results This notion howeveris different when we look at the columns representing thenumber of constants 500 and 1000 Consequently we triedvarying the number of constants in the evidence set as well tobetter understand this trend

By increasing the constants of the evidence set to 50we observe that the precision of the recognition increasesfrom below 60 to about 70 Whilst much improvementis achieved between 1 and 10 constants in evidence theperformance stabilises from 20 constants This gives theintuition that a caveat can be established for the density ofsensory information at a given location since more evidenceafter some point can be inadmissible in the recognitionprocess From this observation we hypothesize that a coher-ent representation of events combining modalities can allowCPSs to efficiently monitor and control environments withmultiple sensors

Concurrent event recognition was also investigated usingmultiple events representing the two distributed domainsunder consideration Similarly wemeasured the precision forrecognition of multiple events using training sets containingconstants 100 500 1000 and 1500 As shown in Figure 11the precision for recognition of multiple events follows thetrend of the single event recognition This observation isattributable to the fact that even though different constants


Number of constants in training set = 100 Number of constants in training set = 500


Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00


Home eventEngine event







Figure 11 Precision of a multiple event recognition with increasing evidence

applicable to the single event and multiple event casesgenerate different ground Markov networks with varyingstructures the model performance is still guided by thecommon underlying MLN However we recognise that themonotonicity in the precision with increasing constants inevidence is not upheld fully in themultiple event recognitionWhereas the precision increases monotonically in the engineevent the home event does not follow this trend completelyBut this new phenomenon is yet another indication thatdetermining a caveat on the size of contextual data towardsoptimal precision in event recognition is paramount Obvi-ously all training data sets hold that the precision of the homeevent ismodal at 20 constants in evidence which is consistentwith the precision of single event recognition Because themultiple event recognition process contains more combinedconstants than the single event recognition the multipleevent process performs better In essence this approach cansupport concurrency in operations of CPSs for collaborativeprocesses

Finally we investigated a multivalued logic in whichMLN clauses are somewhat fuzzified to represent partialknowledge In mimicking rectangular fuzzy functions weused the built-in Alchemy predicate greaterThanEq(119909 119910) todefine a single MLN clause that represents both normaland high temperature event inputs Weights of MLN clausescharacterise subregions of high and normal temperaturemeasurements Overlapping regions as enshrined in fuzzylogic [42] can be treated as a straightforward constantsdeclaration in the training data set For instance temperaturevalue 26∘Cwas considered ambiguous in our experiments Soin the training data set this temperature value was declared afew times to indicate that this same value could be describedas normal and slightly warmAswe can see fromFigure 12 weobtained impressive results with even least training data setUsing a training data set of 100 constants the precision for asingle fuzzy event recognition also improves with increasingnumber of constants in the evidence set Overall usingMLN event recognition in CPSs can be modelled to handle

14 Journal of SensorsPr

ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398

Figure 12 Precision of a fuzzy event recognition with increasingevidence

both uncertainty and vagueness whilst modelling domainuncertainty

8 Conclusion

In this paper we proposed a context-aware multiagent archi-tecture for distributed reasoning in cyber-physical systemsThis architecture is rooted in service-oriented computingand with the incorporation of services of semantic agents theseamless integration of the cyber and physical componentscan be achieved Ontological intelligence provides the under-lying semantics of this approach and together with Markovlogic networks we defined an uncertainty-based reasoningprocedure for event recognition in cyber-physical systemsWith the results of our experiments it is convincing thatthis framework can be relied upon for concurrent processingof events in cyber-physical systems Because these semanticagents are thought to be autonomous and intelligent in theiroperations future work of this research shall consider agentcommunication techniques that can ensure good level ofcooperation among these agents

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This research was sponsored by NSFC 61370151 and 61202211National Science and Technology Major Project of China2015ZX03003012 Central University Basic Research FundsFoundation of China ZYGX2014J055 and Huawei Technol-ogy Foundation YB2013120141

References

[1] R Baheti and H Gill ldquoCyber-physical systemsrdquo The Impact ofControl Technology vol 12 pp 161ndash166 2011

[2] E A Lee ldquoCyber physical systems design challengesrdquo in Pro-ceedings of the 11th IEEE International Symposium on ObjectOriented Real-Time Distributed Computing (ISORC rsquo08) pp363ndash369 Orlando Fla USA May 2008

[3] L Sha SGopalakrishnanX Liu andQWang ldquoCyber-physicalsystems a new frontierrdquo inMachine Learning in Cyber Trust pp3ndash13 Springer Berlin Germany 2009

[4] Y Tan M C Vuran and S Goddard ldquoSpatio-temporal eventmodel for cyber-physical systemsrdquo in Proceedings of the 29thIEEE International Conference on Distributed Computing Sys-tems Workshops (ICDCS rsquo09) pp 44ndash50 Montreal CanadaJune 2009

[5] L Atzori A Iera and G Morabito ldquoThe internet of things asurveyrdquoComputer Networks vol 54 no 15 pp 2787ndash2805 2010

[6] G Kortuem F Kawsar V Sundramoorthy and D FittonldquoSmart objects as building blocks for the internet of thingsrdquoIEEE Internet Computing vol 14 no 1 pp 44ndash51 2010

[7] J Singh O Hussain E Chang and T Dillon ldquoEventhandling for distributed real-time cyber-physical systemsrdquoin Proceedings of the 15th IEEE International Symposiumon ObjectComponentService-Oriented Real-Time DistributedComputing (ISORC rsquo12) pp 23ndash30 Shenzhen China April 2012

[8] S Karnouskos A W Colombo T Bangemann et al et al ldquoTheimcaesop architecture for cloud-based industrial cyber-physicalsystemsrdquo in Industrial Cloud-Based Cyber-Physical Systems pp49ndash88 Springer Berlin Germany 2014

[9] A Boyd D Noller P Peters et al Soa in Manufacturing GuideBook MESA International IBM Corporation and CapgeminiCo-Branded White Paper 2008

[10] R Harrison and A W Colombo ldquoCollaborative automationfrom rigid coupling towards dynamic reconfigurable produc-tion systemsrdquo IFAC Proceedings Volumes vol 38 no 1 pp 184ndash192 2005

[11] M Armbrust A Fox R Griffith et al ldquoA view of cloudcomputingrdquo Communications of the ACM vol 53 no 4 pp 50ndash58 2010

[12] K-J Lin and M Panahi ldquoA real-time service-oriented frame-work to support sustainable cyber-physical systemsrdquo in Pro-ceedings of the 8th IEEE International Conference on IndustrialInformatics (INDIN rsquo10) pp 15ndash21 IEEE Osaka Japan July2010

[13] D DHoang H-Y Paik andC-K Kim ldquoService-orientedmid-dleware architectures for cyber-physical systemsrdquo InternationalJournal of Computer Science and Network Security vol 12 no 1pp 79ndash87 2012

[14] J Pascoe ldquoAdding generic contextual capabilities to wearablecomputersrdquo in Proceedings of the Digest of Papers SecondInternational Symposium on Wearable Computers pp 92ndash99Pittsburgh PA USA Oct 1998

[15] S Noor and H N Minhas ldquoContext-aware perception forcyberphysical systemsrdquo in Computational Intelligence for Deci-sion Support in Cyber-Physical Systems pp 149ndash167 Springer2014

[16] Z Song Y Chen C R Sastry and N C Tas Optimal Observa-tion for Cyber-Physical Systems Springer London UK 2009

[17] L-A Tang X Yu S Kim et al ldquoTrustworthiness analysis ofsensor data in cyber-physical systemsrdquo Journal of Computer andSystem Sciences vol 79 no 3 pp 383ndash401 2013


[18] M Richardson and P Domingos ldquoMarkov logic networksrdquoMachine Learning vol 62 no 1-2 pp 107ndash136 2006

[19] J Wan H Yan H Suo and F Li ldquoAdvances in cyber-physical systems researchrdquo KSII Transactions on Internet andInformation Systems vol 5 no 11 pp 1891ndash1908 2011

[20] G Quan ldquoAn integrated simulation environment for cyber-physical system co-simulationrdquo in Proceedings of the NationalWorkshop on High-Confidence Automotive Cyber-Physical Sys-tems Troy Mich USA April 2008

[21] T W Hnat T I Sookoor P Hooimeijer W Weimer and KWhitehouse ldquoMacroLab a vector-based macroprogrammingframework for cyber-physical systemsrdquo in Proceedings of the6th ACM Conference on Embedded Networked Sensor Systems(SenSys rsquo08) pp 225ndash238 ACM Raleigh NC USA November2008

[22] H J La and S D Kim ldquoA service-based approach to designingcyber physical systemsrdquo in Proceedings of the 9th IEEEACISInternational Conference on Computer and Information Science(ICIS rsquo10) pp 895ndash900 Potsdam Germany August 2010

[23] C-F Lai Y-W Ma S-Y Chang H-C Chao and Y-M HuangldquoOSGi-based services architecture for cyber-physical homecontrol systemsrdquo Computer Communications vol 34 no 2 pp184ndash191 2011

[24] P Pederson D Dudenhoeffer S Hartley and M PermannCritical Infrastructure InterdependencyModeling A Survey of Usand International Research Idaho National Laboratory 2006

[25] P P Datta M Christopher and P Allen ldquoAgent-based mod-elling of complex productiondistribution systems to improveresiliencerdquo International Journal of Logistics Research and Appli-cations vol 10 no 3 pp 187ndash203 2007

[26] G Jiang W W Chung and G Cybenko ldquoSemantic agenttechnologies for tactical sensor networksrdquo in Proceedings ofthe in SPIE Conference on AeroSense pp 311ndash320 InternationalSociety for Optics and Photonics 2003

[27] J Liu and F Zhao ldquoTowards semantic services for sensor-richinformation systemsrdquo in Proceedings of the 2nd InternationalConference on Broadband Networks (BROADNETS rsquo05) pp 44ndash51 Boston Mass USA October 2005

[28] A Elci and B Rahnama ldquoConsiderations on a new softwarearchitecture for distributed environments using autonomoussemantic agentsrdquo inProceedings of the 29thAnnual InternationalComputer Software and Applications Conference (COMPSACrsquo05) vol 2 pp 133ndash138 IEEE July 2005

[29] J Lin S Sedigh and A Miller ldquoModeling cyber-physical sys-tems with semantic agentsrdquo in Proceedings of the 34th AnnualIEEE International Computer Software and Applications Con-ference Workshops (COMPSACW rsquo10) pp 13ndash18 Seoul Korea(South) July 2010

[30] C Talcott ldquoCyber-physical systems and eventsrdquo in Software-Intensive Systems and New Computing Paradigms pp 101ndash115Springer Berlin Germany 2008

[31] I Lovrek ldquoContext awareness in mobile software agent net-workrdquo Tehnicke Znanosti vol 513 no 15 pp 7ndash28 2012

[32] L Serafini and A Tamilin ldquoDrago distributed reasoning archi-tecture for the semantic webrdquo in The Semantic Web Researchand Applications pp 361ndash376 Springer 2005

[33] Y Zhang Y Xu H Hu and X Liu ldquoSemantical informationgraph model toward fast information valuation in large teamworkrdquo in Proceedings of the 21st European Conference onArtificial Intelligence (ECAI rsquo14) August 2014

[34] X H Wang D Q Zhang T Gu and H K Pung ldquoOntologybased context modeling and reasoning usingOWLrdquo in Proceed-ings of the 2nd IEEE Annual Conference on Pervasive Computingand Communications (PerCom rsquo04) pp 18ndash22 Orlando FlaUSA March 2004

[35] P Derler E A Lee and A Sangiovanni Vincentelli ldquoModelingcyberndashphysical systemsrdquo Proceedings of the IEEE vol 100 no 1pp 13ndash28 2012

[36] L Ding J Shinavier T Finin and D L McGuinness OwlSameas and Linked Data An Empirical Study 2010

[37] H Halpin P J Hayes J P McCusker D L McGuinness andH S Thompson ldquoWhen owl sameas isnrsquot the same an analysisof identity in linked datardquo in The Semantic WebmdashISWC 2010vol 6496 of Lecture Notes in Computer Science pp 305ndash320Springer Berlin Germany 2010

[38] N Bieberstein S Bose L Walker and A Lynch ldquoImpact ofservice-oriented architecture on enterprise systems organiza-tional structures and individualsrdquo IBM Systems Journal vol 44no 4 pp 691ndash708 2005

[39] N S Schutte J M Malouff L E Hall et al ldquoDevelopment andvalidation of a measure of emotional intelligencerdquo Personalityand Individual Differences vol 25 no 2 pp 167ndash177 1998

[40] K Kaneiwa M Iwazume and K Fukuda ldquoAn upper ontologyfor event classifications and relationsrdquo in AI 2007 Advancesin Artificial Intelligence 20th Australian Joint Conference onArtificial Intelligence GoldCoast Australia December 2ndash6 2007Proceedings Lecture Notes in Computer Science pp 394ndash403Springer Berlin Germany 2007

[41] S K P S M Richardson P Domingos and M S H PoonTheAlchemy System for Statistical Relational AI User Manual 2007

[42] L A Zadeh G J Klir and B Yuan Fuzzy Sets Fuzzy Logic andFuzzy Systems Selected Papers vol 6 World Scientific 1996

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpswwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



requires uncertainty modelling techniques towards goodmodel performance

This paper proposes a context-based multiagent archi-tecture for distributed reasoning in CPSs Sensor networksin distributed physical environments provide the semanticcapabilities of this model and semantically annotated low-level contextual information merges with domain knowl-edge in a reasoning engine Derived implicit knowledgethrough semantic reasoning and together with the annotateddata enables distributed software agents operating in thecyberspace to provide decision support for actuation infor-mation Each agent is capable of interacting with the physicalenvironment and sharing some principal commonalities withthe other agents As such the agents are capable of expos-ing consuming and even at times processing collaborativeservices targeting laid down system goals To incorporateuncertainty in modelling semantic reasoning on this modelincorporates the inferential power of Markov logic networks(MLN) [18] into event recognition to reduce inferential andcomputational loads Finally we discuss a case study of asmart home as a CPS and results of our experiments showfeasibility of this approach in modelling concurrent events inCPSs

In summary we describe in this paper our initial workin CPSs which overcomes some of the major limitations inthis research fieldWe provide four main contributions Firstwe propose a multiagent architecture that can bridge the gapbetween the operations of the cyber and physical componentsof CPSs Second we describe a procedure that can be used todynamically compose high-level system goals and underlyingcriteria from low-level contextual information Third weintroduce a smart home ontology which incorporates humanactors in the physical space of CPSs as computing entitiesFinally we present a methodology based on MLN for eventrecognition in CPSs

The rest of the paper is organised as follows we presentthe state of the art and the study background and otherpreliminary information in Sections 2 and 3 respectivelySection 4 gives an agent model for CPSs followed by ourframework for distributed reasoning in CPSs in Section 5uncertainty-based event recognition in CPSs using MLN ispresented in Section 6 experiments and discussions of resultsare presented in Section 7 and finally we conclude andpropose future research in Section 8

2 Related Work

A coherent distributed reasoning architecture is one thatachieves an interoperable CPS to cope with the requirementsof the physical and cyber components Because of the lackof sound theoretical foundation for CPSs currently [19] mostapproaches successfullymodel either the physical component[20] or the cyber component [21] but not both Salientstudies towards comprehensive models for CPSs that takeinto account both the cyber and physical components include[8 22 23] which use service-oriented computing to achieveinteroperable CPSs Service-oriented approach alone how-ever is not suitable for modelling real-time distributed CPSs

with multiscale dynamics and networking In this regard anagent-based modelling is more appropriate for distributedcomplex systems such as CPSs [24 25]

Because of the underlying sensor network in CPSssemantic agent technologies are closely associated with ourapproach As provided in [26] a semantic agent technologyhas been used to describe a battlefield information systemwhich uses information fusion processes to dynamicallyintegrate sensor networks towards real-time context-basedreasoning To enable scalable sensor information access anarchitecture and programming model for service-orientedsensor information platform has also been proposed [27]This approach leverages an ontological abstraction of infor-mation to optimise use of resources in collecting storingand processing data Timeliness and concurrency in dis-tributed processing environments can also be enhanced usingautonomous software agents As such use of autonomoussemantic agents as a new software paradigm for distributedcomputing environments has been proposed [28]

Obviously the complex dynamics of CPSs coupled withthe need to properly represent embedded computing andcommunication capabilities motivate the use of semanticsand distributed agents towards interoperable CPSs As canbe found in [29] a multiagent model for CPSs in which adistributed semantic agent model augments data acquisitionprocess with ontological intelligence has been proposedThismodel however provides no procedure for reasoning locallyabout individual components and globally about system-wide properties But such semantics for instance eventrecognition is very critical for distributed real-time CPSsand can essentially specify components of systems in termsof interfaces and observations [7 30] Additionally ontologyforming the underlying layer in the semantic agent-basedmodel primarily supports certainty-based reasoning and assuch requires techniques to address uncertainty inmodelling

In our approach therefore we augment the semanticmultiagent architecture with a robust reasoning mechanismwhich can support both certainty and uncertainty-basedreasoning To promote concurrency and timeliness in theoperations of CPSs MLN is adopted as an uncertainty mod-elling framework and can compactly represent heterogeneouscomputations using a common set of rules Essentially thisachieves a reasoning procedure forCPSs by leveraging advan-tages of both ontology and probabilistic graphical modelsto model both complexity and uncertainty which reduceslimitations of both the ontology and probabilistic graphicalmodels

3 Background and Preliminaries

We provide in this section aspects of the problem of real-timedistributed reasoning in CPSsMarkov logic and smart homeas a case study of a CPS

31 Problem Description The growing pervasiveness of CPSsfurther keeps the cyber and physical components apart andseparately managing these components would not allow usto realise fully the benefits of these systems Appropriate









119875 (119909) = 1119885 expsum119894

119908119894119899119894 (119909) = 1119885prod119894

120601119894 (119909119894)119899119894(119909) (1)











Request1 Request2


S11 S12 S21 S31 S32 S33 S34Sc













Networks


Semantic annotation






A

CB

Data


Cyberspace




Physical space

repository
















Base ontology

Home ontology


Extends

Device

Agent

Human

Network

Thing

ContextEntity

ComputingEntity

Deployment

Address

Building

GeoCoordinates

Location

Passageway

Room

Specification

Service

AdhocService

RoutineService

User

WaterService

LibraryService

Thing

Institution

AdhocService

RegistrationService

RoutineService

Room

BathRoom

LectureRoom

ConferenceRoom

Laboratory

Room

BathRoom

Thing

Home

AdhocService

AssistiveHealth

RoutineService

Bedroom

ChildrenBedroom

GuestBedroom

MasterBedroom

Dining

Kitchen

LaundryRoom

LivingRoom

Extends







Context environment

Usage environment

CyberspaceM

onito

ring

Actu

atio

n





Models Filter

Composer

Nonexecutable

Executable
















Device

Room Car

Deployment Event

Component Function

Consists

Value

Output

PropertyPropert

y

Sub property

Trig

ers



119860119888119905119894V119890 (119910) and (119860119888119905119894119900119899 (119909 119910) or 119860119891119891119890119888119905 (119909 119910)) 997904rArr not119860119888119905119894V119890 (119910) (2)


exist119910 (119909 = 119910 997904rArr 119909 = 119910) (3)





exist119888 119862119900119899119889 (119888) and 119881119886119897119906119890 (119909) = 119888 and 119881119886119897119906119890 (119910) lt 119888 (5)


hasDeployment(TSLR)








119877119900119900119898 (119909) 997904rArr 119861119890119889119903119900119900119898 (119909) forall119909 (7)








119865MLN = (120583 119865 997888rarr 119882) (9)


119865119888 = (119888 120583 (119888)) | 119888 isin 119862 (10)




Very cold Cold Hot

Very hot

24 26 320

w1

w2w3

120583



120583119888119900119897119889capℎ119900119905 = min (120583119888119900119897119889 120583ℎ119900119905) (11)















hasDeployment(TSLR)


Output(HT)

Output(N)







Prec

ision

07

06

05

04


TS = 100

TS = 500

TS = 1000TS = 1500










Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation

















119875 (119909) = 1119885 expsum119894

119908119894119899119894 (119909) = 1119885prod119894

120601119894 (119909119894)119899119894(119909) (1)











Request1 Request2


S11 S12 S21 S31 S32 S33 S34Sc













Networks


Semantic annotation






A

CB

Data


Cyberspace




Physical space

repository
















Base ontology

Home ontology


Extends

Device

Agent

Human

Network

Thing

ContextEntity

ComputingEntity

Deployment

Address

Building

GeoCoordinates

Location

Passageway

Room

Specification

Service

AdhocService

RoutineService

User

WaterService

LibraryService

Thing

Institution

AdhocService

RegistrationService

RoutineService

Room

BathRoom

LectureRoom

ConferenceRoom

Laboratory

Room

BathRoom

Thing

Home

AdhocService

AssistiveHealth

RoutineService

Bedroom

ChildrenBedroom

GuestBedroom

MasterBedroom

Dining

Kitchen

LaundryRoom

LivingRoom

Extends







Context environment

Usage environment

CyberspaceM

onito

ring

Actu

atio

n





Models Filter

Composer

Nonexecutable

Executable
















Device

Room Car

Deployment Event

Component Function

Consists

Value

Output

PropertyPropert

y

Sub property

Trig

ers



119860119888119905119894V119890 (119910) and (119860119888119905119894119900119899 (119909 119910) or 119860119891119891119890119888119905 (119909 119910)) 997904rArr not119860119888119905119894V119890 (119910) (2)


exist119910 (119909 = 119910 997904rArr 119909 = 119910) (3)





exist119888 119862119900119899119889 (119888) and 119881119886119897119906119890 (119909) = 119888 and 119881119886119897119906119890 (119910) lt 119888 (5)


hasDeployment(TSLR)








119877119900119900119898 (119909) 997904rArr 119861119890119889119903119900119900119898 (119909) forall119909 (7)








119865MLN = (120583 119865 997888rarr 119882) (9)


119865119888 = (119888 120583 (119888)) | 119888 isin 119862 (10)




Very cold Cold Hot

Very hot

24 26 320

w1

w2w3

120583



120583119888119900119897119889capℎ119900119905 = min (120583119888119900119897119889 120583ℎ119900119905) (11)















hasDeployment(TSLR)


Output(HT)

Output(N)







Prec

ision

07

06

05

04


TS = 100

TS = 500

TS = 1000TS = 1500










Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Request1 Request2


S11 S12 S21 S31 S32 S33 S34Sc













Networks


Semantic annotation






A

CB

Data


Cyberspace




Physical space

repository
















Base ontology

Home ontology


Extends

Device

Agent

Human

Network

Thing

ContextEntity

ComputingEntity

Deployment

Address

Building

GeoCoordinates

Location

Passageway

Room

Specification

Service

AdhocService

RoutineService

User

WaterService

LibraryService

Thing

Institution

AdhocService

RegistrationService

RoutineService

Room

BathRoom

LectureRoom

ConferenceRoom

Laboratory

Room

BathRoom

Thing

Home

AdhocService

AssistiveHealth

RoutineService

Bedroom

ChildrenBedroom

GuestBedroom

MasterBedroom

Dining

Kitchen

LaundryRoom

LivingRoom

Extends







Context environment

Usage environment

CyberspaceM

onito

ring

Actu

atio

n





Models Filter

Composer

Nonexecutable

Executable
















Device

Room Car

Deployment Event

Component Function

Consists

Value

Output

PropertyPropert

y

Sub property

Trig

ers



119860119888119905119894V119890 (119910) and (119860119888119905119894119900119899 (119909 119910) or 119860119891119891119890119888119905 (119909 119910)) 997904rArr not119860119888119905119894V119890 (119910) (2)


exist119910 (119909 = 119910 997904rArr 119909 = 119910) (3)





exist119888 119862119900119899119889 (119888) and 119881119886119897119906119890 (119909) = 119888 and 119881119886119897119906119890 (119910) lt 119888 (5)


hasDeployment(TSLR)








119877119900119900119898 (119909) 997904rArr 119861119890119889119903119900119900119898 (119909) forall119909 (7)








119865MLN = (120583 119865 997888rarr 119882) (9)


119865119888 = (119888 120583 (119888)) | 119888 isin 119862 (10)




Very cold Cold Hot

Very hot

24 26 320

w1

w2w3

120583



120583119888119900119897119889capℎ119900119905 = min (120583119888119900119897119889 120583ℎ119900119905) (11)















hasDeployment(TSLR)


Output(HT)

Output(N)







Prec

ision

07

06

05

04


TS = 100

TS = 500

TS = 1000TS = 1500










Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











Networks


Semantic annotation






A

CB

Data


Cyberspace




Physical space

repository
















Base ontology

Home ontology


Extends

Device

Agent

Human

Network

Thing

ContextEntity

ComputingEntity

Deployment

Address

Building

GeoCoordinates

Location

Passageway

Room

Specification

Service

AdhocService

RoutineService

User

WaterService

LibraryService

Thing

Institution

AdhocService

RegistrationService

RoutineService

Room

BathRoom

LectureRoom

ConferenceRoom

Laboratory

Room

BathRoom

Thing

Home

AdhocService

AssistiveHealth

RoutineService

Bedroom

ChildrenBedroom

GuestBedroom

MasterBedroom

Dining

Kitchen

LaundryRoom

LivingRoom

Extends







Context environment

Usage environment

CyberspaceM

onito

ring

Actu

atio

n





Models Filter

Composer

Nonexecutable

Executable
















Device

Room Car

Deployment Event

Component Function

Consists

Value

Output

PropertyPropert

y

Sub property

Trig

ers



119860119888119905119894V119890 (119910) and (119860119888119905119894119900119899 (119909 119910) or 119860119891119891119890119888119905 (119909 119910)) 997904rArr not119860119888119905119894V119890 (119910) (2)


exist119910 (119909 = 119910 997904rArr 119909 = 119910) (3)





exist119888 119862119900119899119889 (119888) and 119881119886119897119906119890 (119909) = 119888 and 119881119886119897119906119890 (119910) lt 119888 (5)


hasDeployment(TSLR)








119877119900119900119898 (119909) 997904rArr 119861119890119889119903119900119900119898 (119909) forall119909 (7)








119865MLN = (120583 119865 997888rarr 119882) (9)


119865119888 = (119888 120583 (119888)) | 119888 isin 119862 (10)




Very cold Cold Hot

Very hot

24 26 320

w1

w2w3

120583



120583119888119900119897119889capℎ119900119905 = min (120583119888119900119897119889 120583ℎ119900119905) (11)















hasDeployment(TSLR)


Output(HT)

Output(N)







Prec

ision

07

06

05

04


TS = 100

TS = 500

TS = 1000TS = 1500










Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation





















Base ontology

Home ontology


Extends

Device

Agent

Human

Network

Thing

ContextEntity

ComputingEntity

Deployment

Address

Building

GeoCoordinates

Location

Passageway

Room

Specification

Service

AdhocService

RoutineService

User

WaterService

LibraryService

Thing

Institution

AdhocService

RegistrationService

RoutineService

Room

BathRoom

LectureRoom

ConferenceRoom

Laboratory

Room

BathRoom

Thing

Home

AdhocService

AssistiveHealth

RoutineService

Bedroom

ChildrenBedroom

GuestBedroom

MasterBedroom

Dining

Kitchen

LaundryRoom

LivingRoom

Extends







Context environment

Usage environment

CyberspaceM

onito

ring

Actu

atio

n





Models Filter

Composer

Nonexecutable

Executable
















Device

Room Car

Deployment Event

Component Function

Consists

Value

Output

PropertyPropert

y

Sub property

Trig

ers



119860119888119905119894V119890 (119910) and (119860119888119905119894119900119899 (119909 119910) or 119860119891119891119890119888119905 (119909 119910)) 997904rArr not119860119888119905119894V119890 (119910) (2)


exist119910 (119909 = 119910 997904rArr 119909 = 119910) (3)





exist119888 119862119900119899119889 (119888) and 119881119886119897119906119890 (119909) = 119888 and 119881119886119897119906119890 (119910) lt 119888 (5)


hasDeployment(TSLR)








119877119900119900119898 (119909) 997904rArr 119861119890119889119903119900119900119898 (119909) forall119909 (7)








119865MLN = (120583 119865 997888rarr 119882) (9)


119865119888 = (119888 120583 (119888)) | 119888 isin 119862 (10)




Very cold Cold Hot

Very hot

24 26 320

w1

w2w3

120583



120583119888119900119897119889capℎ119900119905 = min (120583119888119900119897119889 120583ℎ119900119905) (11)















hasDeployment(TSLR)


Output(HT)

Output(N)







Prec

ision

07

06

05

04


TS = 100

TS = 500

TS = 1000TS = 1500










Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











Context environment

Usage environment

CyberspaceM

onito

ring

Actu

atio

n





Models Filter

Composer

Nonexecutable

Executable
















Device

Room Car

Deployment Event

Component Function

Consists

Value

Output

PropertyPropert

y

Sub property

Trig

ers



119860119888119905119894V119890 (119910) and (119860119888119905119894119900119899 (119909 119910) or 119860119891119891119890119888119905 (119909 119910)) 997904rArr not119860119888119905119894V119890 (119910) (2)


exist119910 (119909 = 119910 997904rArr 119909 = 119910) (3)





exist119888 119862119900119899119889 (119888) and 119881119886119897119906119890 (119909) = 119888 and 119881119886119897119906119890 (119910) lt 119888 (5)


hasDeployment(TSLR)








119877119900119900119898 (119909) 997904rArr 119861119890119889119903119900119900119898 (119909) forall119909 (7)








119865MLN = (120583 119865 997888rarr 119882) (9)


119865119888 = (119888 120583 (119888)) | 119888 isin 119862 (10)




Very cold Cold Hot

Very hot

24 26 320

w1

w2w3

120583



120583119888119900119897119889capℎ119900119905 = min (120583119888119900119897119889 120583ℎ119900119905) (11)















hasDeployment(TSLR)


Output(HT)

Output(N)







Prec

ision

07

06

05

04


TS = 100

TS = 500

TS = 1000TS = 1500










Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation


















Device

Room Car

Deployment Event

Component Function

Consists

Value

Output

PropertyPropert

y

Sub property

Trig

ers



119860119888119905119894V119890 (119910) and (119860119888119905119894119900119899 (119909 119910) or 119860119891119891119890119888119905 (119909 119910)) 997904rArr not119860119888119905119894V119890 (119910) (2)


exist119910 (119909 = 119910 997904rArr 119909 = 119910) (3)





exist119888 119862119900119899119889 (119888) and 119881119886119897119906119890 (119909) = 119888 and 119881119886119897119906119890 (119910) lt 119888 (5)


hasDeployment(TSLR)








119877119900119900119898 (119909) 997904rArr 119861119890119889119903119900119900119898 (119909) forall119909 (7)








119865MLN = (120583 119865 997888rarr 119882) (9)


119865119888 = (119888 120583 (119888)) | 119888 isin 119862 (10)




Very cold Cold Hot

Very hot

24 26 320

w1

w2w3

120583



120583119888119900119897119889capℎ119900119905 = min (120583119888119900119897119889 120583ℎ119900119905) (11)















hasDeployment(TSLR)


Output(HT)

Output(N)







Prec

ision

07

06

05

04


TS = 100

TS = 500

TS = 1000TS = 1500










Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










hasDeployment(TSLR)








119877119900119900119898 (119909) 997904rArr 119861119890119889119903119900119900119898 (119909) forall119909 (7)








119865MLN = (120583 119865 997888rarr 119882) (9)


119865119888 = (119888 120583 (119888)) | 119888 isin 119862 (10)




Very cold Cold Hot

Very hot

24 26 320

w1

w2w3

120583



120583119888119900119897119889capℎ119900119905 = min (120583119888119900119897119889 120583ℎ119900119905) (11)















hasDeployment(TSLR)


Output(HT)

Output(N)







Prec

ision

07

06

05

04


TS = 100

TS = 500

TS = 1000TS = 1500










Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Very cold Cold Hot

Very hot

24 26 320

w1

w2w3

120583



120583119888119900119897119889capℎ119900119905 = min (120583119888119900119897119889 120583ℎ119900119905) (11)















hasDeployment(TSLR)


Output(HT)

Output(N)







Prec

ision

07

06

05

04


TS = 100

TS = 500

TS = 1000TS = 1500










Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










hasDeployment(TSLR)


Output(HT)

Output(N)







Prec

ision

07

06

05

04


TS = 100

TS = 500

TS = 1000TS = 1500










Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00

Prec

ision

08

07

06

05

04

03

02

01

00













ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










ecisi

on

08

07

06

05

04

03

02

01

00


056899

065698

070598073398 073398 073398



8 Conclusion


Competing Interests


Acknowledgments


References














































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation





































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









semantical markov logic network for distributed...

Documents