13 pages

A Knowledge Dashboard for Manufacturing Industries

of 13
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
A Knowledge Dashboard for Manufacturing Industries
  A Knowledge Dashboard for Manufacturing Industries S. Mazumdar 1 , A. Varga 2 , V. Lanfranchi 2 and F. Ciravegna 2   1 Information School 2 OAK Group, Department of Computer ScienceUniversity of Sheffield, Regent Court – 211 Portobello Street,S1 4DP Sheffield,{a.varga, v.lanfranchi, f.ciravegna} Abstract. The manufacturing industry offers a huge range of opportunities andchallenges for exploiting semantic web technologies. Collating heterogeneousdata into semantic knowledge repositories can provide immense benefits tocompanies, however the power of such knowledge can only be realised if endusers are provided visual means to explore and analyse their datasets in aflexible and efficient way. This paper presents a high level approach to unify,structure and visualise document collections using semantic web andinformation extraction technologies. Keywords: Semantic Web, Information Visualisation, User Interaction. 1 Introduction Modern manufacturing is a complex domain where productivity and efficiency arestrongly affected by a broad range of factors such as site locations, cultural values,management decisions and communication capabilities. For example, largemanufacturing organizations are usually globalised, with facilities geographicallydistributed, making use of multiple manufacturing machines, interacting with severalsuppliers and warehouses. Also, a recent trend in large organisations has been thepresence of dynamic, interdisciplinary working groups and communities of practice   who require rapid, flexible customisation of information to their specific needs [1]. Atthe same time, the information they generate needs to be shared with the rest of theorganisation, and hence, must be presented to other communities in ways that can beeasily understood (and correctly interpreted) and reused [2]. The underlying commonality between these phenomena is information availability:if information is captured, stored and shared between different departments andlocations then efficient communication can be reached and stronger support formanagerial decisions can be provided. Unfortunately this information is oftencollected in a wide variety of formats (e.g., text files, images, PDF documents) and 51  dispersed in independent repositories, including shared directories, local andcompany-wide databases, ad hoc information systems, etc. Critical knowledge may behidden in the huge amount of manufacturing data, and the cost of exhaustivelyidentifying, retrieving and reusing information across this fragmentation is very highand often a near impossible task.  This paper presents how Semantic Web and Information Extraction (IE)technologies can be adopted to unify such collections of documents and formalizetheir knowledge content, bringing together information from different domains, whichcan feed into organisational knowledge. Visualisation techniques can then be appliedon top of the semantically structured data to explore, contextualise and aggregate it,  offering multiple perspectives on the information space and provide analytic tools thatcould support users in spotting trends and identifying patterns and relationships. Inorder to achieve this goal two steps are required:- Knowledge Acquisition: acquiring information from different documents andcorpora and semantically structuring it in a semi-supervised manner.- Knowledge Visualisation: creating multiple views over the semantic knowledgespace.Our methodology is innovative compared to previous literature (analysed inSection 2) as it defines the Knowledge Acquisition and Visualisation steps at anabstract level: the use of ontologies to extract, structure and visualise informationmake our approach flexible, reusable and extensible.The Knowledge Acquisition and Visualisation steps will be described in details inSection 3, before providing implementation details (Section 4) and discussingconclusions and future work (Section 5).  The following scenario (taken from SAMULET 1 , an existing research project onadvanced manufacturing in the aerospace industry in which the authors are involved)has been considered as a foundation for the work: in a manufacturing industry a hugenumber of components are produced every day based on design data provided byDesign departments, and are reused in other divisions of the company. When thesecomponents are produced manufacturing data is collected such as manufacturing time,location of the plant and of the manufacturing machine, type of component and details(possibly linked to design data). Additional information includes the person andmachine responsible for the production, manufacturing costs and so on. This data iscollected in a wide variety of formats (e.g. Excel spreadsheets, images, WordDocuments), stored in independent repositories and often distributed using personalchannels (such as e-mails, or shared network drives).  Manufacturing data are essential to resolving any issue that may arise on acomponent, in order to be able to clearly identify the driving factors behind the issueand to discover any significant trends or patterns related to individual manufacturingunits/machines/personnel. Identifying non-obvious patterns in the data is fundamentalto increasing productivity and efficiency: for example, a consistently poorlyperforming machine may be over-shadowed by a well performing manufacturing unit 1 SAMULET (Strategic Affordable Manufacturing in the UK with Leading EnvironmentalTechnology), Last Accessed 14/04/2011 52   – data analysis and visualisation would help in spotting such trends and supportputting corrective measures in place. 2 Related work Our approach aims to provide a consistent and coherent environment for knowledgeexploration in the manufacturing domain, encompassing knowledge acquisition andknowledge visualisation techniques. Related work in both these areas is nowanalysed, with particular emphasis on the adoption in the manufacturing domain. 2.1 Knowledge Acquisition Traditional machine learning (ML) approaches for knowledge acquisition inmanufacturing started to gain much attention only in recent years [3-10], mostlybecause the majority of the ML algorithms and tools require skilled individuals tounderstand the output of ML process [3]. However there has been some work onusing traditional ML techniques for specific areas (such as fault detection, qualitycontrol, maintenance, engineering design, etc.) employing classification [6,7],clustering [8] and association rule mining [9,10] algorithms [3-5]. Classificationalgorithms were used for categorising data into different classes, for exampleclassifying defects in the semi-conductor industry [5]. [6] employed a hybridapproach combining neural networks and decision tree classification algorithms forrecognising false classifications in control chart pattern recognition (CCPR) thusfacilitating quality control. [7] used decision tree algorithms for producingclassification rules which were then saved in the competitive decision selector (CDS)knowledge bases enabling efficient job shop scheduling. Clustering algorithms werealso used to group similar data into clusters, for example clustering the orders intobatches for speeding up the product movement within a warehouse [5]. [8] appliedfuzzy c-means clustering algorithm for identifying changes in traffic states thusimproving the traffic management systems. Association rule mining algorithms wereused to identify relationships among the attributes describing the data. [9] usedassociation rule mining for detecting the source of assembly faults, thus improving thequality of assembly operations. [10] extracted association rules from historicalproduct data to identify the limitations of the manufacturing processes. Thisinformation can then be used to improve the quality of the product and identify therequirements for design change.Despite the increased interest, most of these approaches still lack portability andrequire a large amount of annotated data to achieve high performance, which isusually tedious and costly [13] to obtain. Furthermore recent advances in domainadaptation show that traditional Machine Learning (ML) approaches for IE are nolonger the best choices [11,12]. These algorithms work only well when the format,writing style in which the data (e.g. manufacturing time, location of the plant and themachine) is presented is similar across different corpora [11,12]. In dynamic andheterogeneous corpora, these ML based systems need to be rebuilt for each corpus orformat, making them impractical in many scenarios [11], such as the one presented in 53  this paper. To enable effective knowledge capture in manufacturing our approachemploys an adaptable IE framework based on domain adaptation techniques, aspresented in Section 3. 2.2 Knowledge Visualisation Information visualisation techniques have been extensively adopted in themanufacturing domain to display and illustrate different processes such as simulationof model verification and validation, planning, decision making purposes and so on[14, 20]. Though most simulation results are based on data models, visualisations areessential to efficiently communicate information to end-users [15]. For examplevisualising CAD (Computer Aided Design) models enriched with performance scoresprovides analysts insights into the performances of different manufacturing units;alternative techniques provide ways for manufacturing units to validate their productsagainst software models [14] (to evaluate compliance of manufacturing units todesign).Commercial tools generally focus on 3D visualisations of manufacturing models,factories, machines and so on. Examples of such commercially available tools used inthe manufacturing industry include Rockwell’s FactoryTalk 2 (remote monitoring of manufacturing processes); Autodesk’s 3ds Max 3 and Maya 4 (modelling of productdesigns, animation, virtual environments); VSG’s OpenInventor 5 (3D Graphics toolkitfor developing interactive applications); DeskArtes ViewExpert 6 (viewing, verifying,measuring CAD data); Oracle’s AutoVue 7 (Collaboration tool to annotate 3D or 2Dmodels). These 3D commercial tools are also adopted in other industries like gaming,animation and so on [17]. However the high cost of 3D hardware and software makesthis option unfeasible for smaller companies [16].3D visualisation techniques have also been investigated in academic works, such asCyberbikes, a tool for interaction with and exploration using head-mounted displays.[21] presents another example of 3D visualisation, providing factory floor mapswhich use animations to convey real-time events.Using visualisations to communicate high-quality data in manufacturing scenarioscan greatly reduce the amount of time and effort taken by engineers to resolve anissue: in a study by [18], engineers provided with animated visualisations combiningseveral steps of a simulation could substantially reduce their analysis time. [22]discusses how factory map visualisation based navigation can often provide means tosignificantly reduce the cognitive load on analysts monitoring a typical manufacturingfactory, when compared to list-based navigation of factory machines and theirperformances. Our approach takes inspiration from this latter works in aiming to 2 FactoryTalk, LastAccessed 14/04/2011 3 AutoDesk 3ds Max, Last Accessed 14/04/2011 4 AutoDesk Maya, Last Accessed 14/03/2011 5 VSG OpenInventor, Last Accessed 14/04/2011 6 DeskArtes ViewExpert, Last Accessed 14/04/2011 7 Oracle AutoVue, LastAccessed 14/04/2011 54  provide efficient visualisation techniques that will reduce engineers cognitiveworkload and facilitate knowledge analysis. 3. Adding semantics to the manufacturing domain Given the large scale and the heterogeneity both in data types and data formats,automatic techniques are required to process the data, unifying the documentcollections and formalising their knowledge content. In the following we distinguishbetween data, information and knowledge as proposed in [27]. Namely, data refers tothe basic raw unit without any implicit meaning, information refers to data enhancedwith context and perspective, and knowledge is information connected by patternsand relations. In our case the outcome of our Information Extraction framework isconsidered knowledge as it extracts entities and relations and assigns semanticmeaning to them.Our approach (shown in Figure 1) is therefore based on the use of a commonknowledge representation in the form of ontologies describing the manufacturingdomain. The ontologies are created manually so that the high-level ontology coversthe generic manufacturing scope (common concepts and relationships between them),and the local ontologies (interlinked by the over-arching high-level ontology) capturethe information specific to the different corpora. An adaptable Information Extractionframework considering the high-level ontology then extracts the common conceptsacross the corpora, thus avoiding ontology mapping and integration (see Section 3.1). Figure 1 - The knowledge acquisition and visualisation process 55
Related Documents
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!