2 1094-7167/04/$20.00 © 2004 IEEE IEEE INTELLIGENT SYSTEMSPublished by the IEEE Computer Society

E - S c i e n c e

Realizing the HydrogenEconomy throughSemantic WebTechnologiesJane Hunter, Distributed Systems Technology CentreJohn Drennan, Centre for Microscopy and Microanalysis, University of QueenslandSuzanne Little, University of Queensland

D iminishing reserves of cheap, recoverable crude oil, together with increasing ten-

sions between the Middle East and the West, will likely threaten our access to

affordable oil in the future. Alternative fossil fuels such as coal, tar sand, and heavy oil

will only worsen global warming.1 Hydrogen, however, is plentiful and clean, stores

energy more effectively than batteries, burns twice asefficiently in a fuel cell as gasoline does in an inter-nal-combustion engine, leaves only water behind,and can power cars. So, many observers view asinevitable the transition from an economy poweredby fossil fuels to one based on hydrogen.

But many challenges remain before fuel cells willreplace oil and other fossil fuels. Materials scientistsmust determine how to increase the cells’efficiency,reduce their production costs, determine how theydegrade over time, extend their life, and recycle theircomponents. Overcoming these challenges requiresa deep understanding of the components’ micro-scopic structure, detailed analytical informationabout how the various components interact, and theability to predict how the components will performunder load conditions.

To solve these problems, materials scientists areturning to knowledge management techniques, par-ticularly Semantic Web technologies, to make senseof and assimilate the vast amounts of microstruc-tural, performance, and manufacturing data that theyacquire during their research.

At the University of Queensland, the DistributedSystems Technology Centre and the Centre forMicroscopy and Microanalysis are collaborating ona project that’s applying Semantic Web technologiesto optimize fuel cell design. (For the DSTC URLand other URLs pertinent to this article, see therelated sidebar.) We call it FUSION (Fuel Cell Under-standing through Semantic Inferencing, Ontologiesand Nanotechnology).

Applying Semantic Web technologiesto fuel cell understanding

Because fuel cell efficiency depends on the fuelcell layers’ internal structure and the interfacesbetween them, an analysis of electron-microscopyimages of cross-sectional samples through fuel cellscan reveal valuable information. (For more on howfuel cells work, see the “Hydrogen Power” sidebar.)Simple macrolevel information such as the thick-ness, surface area, roughness, and densities of thecell layers can help us determine gas permeation ofthe electrode materials. Nanolevel information aboutthe electrode’s internal interface structure providesdata on exchange reaction efficiency. Figure 1 illus-trates the image data obtainable at different magni-fications, which must be analyzed and semanticallyindexed to fully mine the potential knowledge in theimages.

Besides the complex multilayered microstructuralinformation that the images reveal, another consid-eration is the manufacturing conditions and pro-cessing parameters used to produce the cell config-urations. Fuel cell production involves a highlycomplex set of procedures that introduce a furtherset of variables (temperature, time, pressure, and soon) that influence the cell’s performance.

Additionally, for each cell configuration, perfor-mance data is available in the form of complexgraphs that plot voltage against current density (seeFigure 2).

The problem’s crux is to enable the assimilation ofthe three highly complex data sets—microstructural,

The FUSION (Fuel

Cell Understanding

through Semantic

Inferencing,

Ontologies and

Nanotechnology)

project applies,

extends, and combines

Semantic Web

technologies and

image analysis

techniques to

develop a knowledge

management system

to optimize the

design of fuel cells.

processing, and cell performance data—sothat we can interrogate the information toreveal trends that could help improve fuel celldesign and efficiency. However, the amountof information is such that human processingis impossible; the task requires more sophis-ticated means of data mining. Making senseof the vast amounts of heterogeneous, com-plex, and multidimensional information thatfuel cells present to materials scientists pro-vides a challenging and ideal application fortesting Semantic Web technologies’ capabil-ities and the Semantic Web community’sclaims.2

The FUSION project has two main goals.First, we aim to apply and evaluate existingSemantic Web technologies. Second, we aim

JANUARY/FEBRUARY 2004 www.computer.org/intelligent 3

• Algaewww.w3.org/1999/02/26-modules/User/Algae-HOWTO.html

• Centre for Microscopy and Microanalysiswww.uq.edu.au/nanoworld/about_cmm.html

• Distributed Systems Technology Centrewww.dstc.edu.au

• FUSION

http://metadata.net/sunago/fusion.htm• FUSION ontology

http://metadata.net/sunago/fusion/fusion.owl• Harmony project

http://metadata.net/harmony• JESS (Java Expert System Shell)

http://herzberg.ca.sandia.gov/jess• Mandarax

www.mandarax.org• MathML

www.w3.org/TR/MathML2• MATLAB

www.mathworks.com/products• MPEG-7

www.chiariglione.org/mpeg/standards/mpeg-7/mpeg-7.htm• OME (Open Microscopy Environment)

www.openmicroscopy.org

• OME ontologyhttp://metadata.net/sunago/fusion/ome.owl

• OME CoreSemanticTypeshttp://openmicroscopy.org/XMLschemas/STD/RC2/OMECoreTypes.ome

• OWL-Swww.daml.org/services/owl-s/1.0

• Perllibwww.w3.org/1999/02/26-modules/User/Annotations-HOWTO

• RuleMLwww.dfki.uni-kl.de/ruleml

• The Semantic Webwww.w3.org/2001/sw

• SMIL (Synchronized Multimedia IntegrationLanguage)www.w3.org/TR/smil20

• SOAPwww.w3.org/TR/2003/REC-soap12-part0-20030624

• WSDLwww.w3.org/TR/wsdl

• Xpointerwww.w3.org/XML/Linking

• Zopewww.zope.org/Members/Crouton/ZAnnot

Related URLs

0

0.7

0.6

0.5

0.4

0.3

0.2

0.1

00.1 0.2 0.3

Current density (A/cm2)

No. 1 (5.8 mΩ)No. 2 (11.6 mΩ)No. 3 (7.5 mΩ)

0.4 0.5

Volta

ge

Figure 2. Typical performance data for a fuel cell.

Figure 1. Microscopic images of a fuel cell at three different magnifications.

to extend and refine these technologies todevelop a knowledge management systemfor the fuel cell research and industry com-munities. The project has eight key phases:

1. Analyze fuel cells’ workflow or lifecycle from the manufacturing stage tothe sectioning, analysis, and knowledgeextraction stages (see Figure 3).

2. Determine the metadata capture require-ments at each step, relevant metadatastandards, and the availability of auto-matic metadata extraction tools.

3. Develop metadata schemas and con-trolled vocabularies to describe andrecord the fuel cell data (manufacturingparameters, performance results, andcross-sectional images at different lev-els of magnification) to document datasources and control metadata quality.

4. Develop metadata input and capturetools based on the manufacturing andanalysis workflow (see Figure 3). Then,integrate these tools with automaticimage-processing technologies andpopulate the underlying, networkeddatabases.

5. Develop a fuel cell ontology. Then,merge it with ontologies for multimediadescription and microscopy, using acommon top-level ontology.

6. Enable the definition and application ofinferencing rules that generate high-level semantic descriptions of the fuelcells (defined in the fuel cell ontology)

from automatically extracted low-levelfeatures.

7. Develop a semantic querying, browsing,and presentation and visualization inter-face based on the fuel cell ontology. Theinterface will use an interactive presen-tation generator system3 to assimilatesemantically related images and datainto coherent multimedia presentations.

8. Integrate collaborative image and videoannotation tools. These tools will letdomain experts label a small domain-specific image database to improveimage querying and to attach newknowledge generated from using thesystem.

We describe some of these phases in moredetail throughout this article.

Applying ontologies and semantic inferencing

Figure 4 illustrates the system componentsand architecture for the knowledge manage-ment system we’ve developed to manage themanufacturing, performance, and image datacaptured from fuel cell components.

We use a subset of the ISO/IEC MPEG-7(Multimedia Content Description)4 standardto describe the low-level image features. Weuse the OME (Open Microscopy Environ-ment) standard to capture associated micro-scope details and settings—essential sourceinformation within any scientific context. Wedeveloped FUSION metadata schemas to sat-isfy fuel cell analysts’ descriptive require-ments. In addition to the XML Schemas weuse to define and validate the metadatadescriptions, we developed ontologies todefine the semantics and relationshipsbetween the concepts used in each of theseschemas. We describe these ontologies indetail later.

RuleML rules, which domain expertsdefine through a GUI, are used to relate com-binations of low-level automatically extractedMPEG-7 features to high-level FUSION con-cepts or terms. When the system appliesthese rules to images in the database, it cangenerate high-level semantic descriptions.The system stores all the generated and val-idated metadata in a central knowledgerepository. A query engine, visualizationengine, and knowledge capture (annotation)tools sit on top of the knowledge repository.Together, these components let fuel cell ana-lysts access, interpret, assimilate, and minethe stored data.

4 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS

E - S c i e n c e

Manufacturing

Performanceanalysis

Sectioning

Imageanalysis and annotation

Semantic inferencing

Semantic querying, browsing,visualization, presentation tools

Knowledge extraction

Figure 3 Fuel cell workflow.

FUSION knowledgerepository

Manufacturing.xsd manufacture.xml

Manufacturing.xsd manufacture.xml

Performance.xsd performance.xml

Microscopy.xsd microscopy.xml

Visual engineQuery engineKnowledge capture

Rules-by-Example

Collaborative annotation

Image analysis.xsd image.xml

rules.xml(RuleML/MATHML)

ABContology

FUSIONschema

OME.owl

MPEG-7.owl

FUSION.owl

Inference engine

image.jp2

Figure 4. The FUSION system architecture.

The MPEG-7 ontologyFigure 5 illustrates a subset of the

MPEG-7 OWL ontology.5 We developedthis subset to define the semantics of termsused in the MPEG-7 standard for describ-ing low-level visual features. Figure 5shows only those classes for which “color”is a visual descriptor and color’s five sub-properties. A complete description of theMPEG-7 ontology is available at http://metadata.net/mpeg7.

The OME ontologyThe OME6 is an open-source collabora-

tive software project that aims to develop acommon environment for the analysis, man-agement, and exchange of biological micro-scopic images. A key component of the OMEis an XML-encoded file standard for the stor-age, analysis, and exchange of image dataoutput from microscopy procedures. A com-mon standard for recording microscope-related metadata is essential for capturingmicroscopy images’ source. This standardincludes information such as the micro-scope’s manufacturer, serial number andmodel, instrument settings, detectors, filters,and light sources. We developed an OMEontology from the OME CoreSemantic-Types. It defines the semantics associatedwith microscope output data and settings andlets us relate them to the image descriptorsdefined by the MPEG-7 and FUSION ontolo-gies. Figure 6 illustrates a subset of the OMEOWL ontology.

JANUARY/FEBRUARY 2004 www.computer.org/intelligent 5

Hydrogen-powered fuel cells could provide us with aflexible, pollution-free method of producing electricity.Figure A illustrates how hydrogen produces electricity ina fuel cell. Hydrogen passes over an active electrode,which catalyzes the production of hydrogen ions. Theseions travel through a membrane to an electrode on thesystem’s air side, reacting with the oxygen to producewater. The process involves the transport of electronsthat are effectively harvested to produce energy. Thesesystems work at elevated temperatures and have a pre-dicted efficiency of over 60 percent.1

Reference1. J. Larminie and A. Dicks, Fuel Cell Systems Explained, 2nd ed.,

John Wiley & Sons, 2003.

Oxygen

Water

Electronflow

Load

Hydrogen

Anode Cathode

Hydrogenions

Figure A. How Proton Exchange Membrane (PEM) fuel cells operate. (This figure is based on an illustration at the Smithsonian Institution Web page http://fuelcells.si.edu/basics.htm.)

Hydrogen Power

abc:containsabc:containsabc:contains

otfotf

abc:contains

string

string

string

string

rdfs:subClassOf manufacturerserialNumber

model

string

string

string

manufacturer

modelserialNumber

otftype

abc:Actuality Instrument

OpticalTransferFunction

OpticalTransferFunction

LightSource

rdsf:subClassOf

Dichroic

EmissionFilter

FilterSet

ExcitationFilter

FilterObjectiveDetector

manufacturermodel

serialNumbertypegain

voltageoffset

stringstringstringstringfloatfloatfloat

rdsf:subClassOf

Laser

Filament

Arc

Figure 6. A subset of the OME ontology showing an instrument and its components.

rdfs:subClassOf

rdfs:subClassOfrdfs:subClassOf

rdfs:subClassOf

rdfs:sub

ClassO

f

MovingRegion

StillRegion

VideoSegment

visualdescriptor

color

rdfs:subClassOf rdfs:subClassOf

domain

domain

domain

ColorStructure

ColorLayout

GoFGoPColor

ScalableColor

DominantColor

range

rdfs:Resource

Color

Figure 5. The MPEG-7 visual descriptor “color.”

The FUSION ontologyUsing the approach recommended by

Natalya Noy and Deborah McGuiness,7 we

developed the FUSION fuel cell ontology incollaboration with domain experts. We use itto generate the high-level semantic descrip-tions that these experts use when searching,retrieving, and annotating images. Figure 7illustrates the key classes or concepts andtheir relationships, as defined in FUSION.

To enable semantic interoperabilitybetween the MPEG-7, OME, and FUSION

metadata vocabularies and to define thesemantic and inferencing relationshipsbetween the terms across these differentontologies, we harmonize or relate themusing the top-level or core ABC (A BoringCore) ontology.8 This ontology, developedwithin the Harmony project, provides a

global, extensible model that expresses thebasic concepts that are common across avariety of domains. It also provides the basisfor specialization into domain-specific con-cepts and vocabularies. Other possible upper-level ontologies that we could use for thispurpose include SUMO9 (Suggested UpperMerged Ontology) and DOLCE10 (DescriptiveOntology for Linguistic and Cognitive Engi-neering). Figure 8 illustrates the harmoniza-tion process.

Inferring semantic descriptions of images

In recent years, significant progress hasoccurred in the automatic recognition of low-level features in images. However, compar-atively little progress has occurred inmachine generation of high-level semanticdescriptions of images. In the FUSION project,we’ve developed a unique, user-assistedapproach to generating ontology-basedsemantic descriptions of images from low-level automatically extracted features. Ourapproach lets domain experts define rules(specific to their domain) that map particularcombinations of low-level visual features(color, texture, shape, and size) to high-levelsemantic terms defined in their domainontology.

Such semantic descriptions enable moresophisticated semantic querying of theimages in terms familiar to the user’s domain(as Figure 9 shows). These descriptions alsoensure that services, agents, and applicationson the Web have a greater chance of discov-ering and exploiting the information andknowledge in the images. The use of ontolo-gies reduces the semantic descriptions’potential subjectivity, and the inference rules’documentation provides a mechanism forcapturing and sharing human domain knowl-edge, in the form of the domain experts’analysis methods.

The Rules-by-Example interfaceTo define semantic-inferencing rules for

images, users must be able to specify val-ues for low-level visual features. Consider,for example, a fuel cell expert labeling elec-tron-microscopy images, to enable thesearch and retrieval of particular types offuel cell components:

IF [(color is like this) AND (texture islike this) AND (shape is like this) ]THEN (the object is a platinumsubstrate)

6 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS

E - S c i e n c e

Catalyst

Support

1

1

n

abc:contains

abc:contains

1

1

1

1

n

abc:contains

abc:containsElectrode

n

1

1

abc:Actuality

rdsf:subClassOf

rdfs:subClassOf

FuelCell

MEA

1

1 2

Electrolyte

abc:containsabc:co

ntains

abc:contains

Node

rdfs:subClassOf

Cathode

CatalystSupport

ActiveMetal

Figure 7. The key classes and their relationships in the FUSION fuel cell ontology.

ABC

MPEG-7

OpenMicroscopyEnvironment

Fusion

Figure 8. Using ABC to merge ontologies.

Bibliographic:Creator, publisher,date published

Semantics:Objects, events,people, places

Structural:Regions, segments

Features:Colors, textures,shapes, motion,pitch, tempo, volume

Human interaction

Automatic extraction

VideoImagesGraphs

SMIL filesWeb pages

Heterogeneous complexmultimedia data

Semantic searchand retrieval—for example,“Give me high-porosity fuelcell images.”

Inferencingrules

Figure 9. Inferring semantic descriptions from combinations of low-level features.

The simplest, most intuitive way to spec-ify such rules is to provide a query-by-example (QBE)-type interface by whichusers can specify color, texture, shape, andsize through visual examples, drawing fromimages in their collection of interest. So, wedeveloped our Rules-by-Example interface—a GUI that provides users with color palettes,texture palettes, predefined shapes, or draw-ing tools, by which they can define their ownsemantic-inferencing rules. The RBE is gen-erated from and dependent on the chosenback-end (OWL) ontology (specified duringsystem configuration), which defines andconstrains the semantic descriptions that canbe generated. We can migrate the system toa different domain simply by choosing a dif-ferent back-end ontology.

Figure 10 illustrates the initial prototypeinterface we developed to investigate theRBE’s user requirements and usability. Themenu options are populated dynamically fromthe back-end ontologies. For example, a prop-erty’s possible operators and values are gen-erated from its range definition; for instance,if the type is a string, then relationships suchas “less than” are irrelevant. Such dynamicpopulation of the GUI provides user supportwhile maintaining the domain independencethat’s a strength of this system. Standard log-ical operators (AND/OR) enable the specifi-cation of more advanced logic-based rules.

To specify the values required for atoms inthe rule body, the user can select an existingimage (or set of images) and select imageregions or segments as examples. For instance,in the screen shot in Figure 10, the color spec-ification in this rule states that for the rule tobe true (that is, for StillRegion to depict ActiveMetal),the color must be one of the set on the palette.This palette displays all the colors used in theselected segment, and the user can adjust thisas he or she sees fit. An additional complexitywhen searching for visual features is thefuzzy-matching problem, which usuallyrequires similarity rather than exact matches.To support this, the interface provides the abil-ity to set a threshold value for individual imageattributes or features.

The RBE interface lets the user quicklydevelop, apply, and refine highly complexrules without understanding complex low-level MPEG-7 terms or values. Manual cre-ation of such rules would be extremely diffi-cult and time consuming. In addition, theinterface lets the users focus the system onthe recognition of objects, regions, or fea-tures of highest priority or interest. Initial

user feedback has been promising, and we’remaking further refinements before imple-menting the final interface design in Java.

Once the user is satisfied with a rule defi-nition, he or she can save it in the RuleMLformat (possibly augmented with MathML)for processing by an inferencing engine. Pos-sible candidates for the inferencing-enginecore include JESS (Java Expert SystemShell) and Mandarax, a Java RuleML engine.(Such engines will require extensions if therule definitions include MathML expres-sions.) Input to the inferencing engine con-sists of a set of semantic-inferencing rules, acollection of images, and their automaticallyextracted low-level MPEG-7 features.

To automatically analyze the fuel cellimages, we use MATLAB, a popular, powerfultool widely used by scientists and micro-scopists. MATLAB can produce a large amountof low-level data about the features andobjects in an image—for example, color, tex-ture, shape, area, mean density, standard devi-ation density, perimeter and length, centercoordinates, and integrated density. Althoughwe’re currently calling MATLAB analysis meth-ods directly, we plan to employ SOAP,WSDL, and OWL-S to make MATLAB’S auto-matic feature extraction tools available as WebServices. This would allow greater choiceand the flexibility to plug in more advancedmultimedia extraction and analysis tools andservices as they become available.

The data that’s output from MATLAB analy-sis methods is transformed to MPEG-7descriptions using Perl scripts that we’vedeveloped. Given the automatically extracted

MPEG-7 descriptions of images togetherwith the user-specified rules, the inferenceengine can generate semantic descriptions,determine new semantic relationships (notexplicitly defined), and populate and expandthe knowledge repository. We can evaluatethe results by comparing the automaticallygenerated semantic descriptions with manu-ally attached annotations for the same images.

Given the semantic descriptions generatedby applying the rules defined in the RBEGUI, domain experts can perform muchmore complex and sophisticated queries overthe image data. As we mentioned before, theycan use terms that are useful and familiar totheir domain—for example, “give me all theimages depicting high-porosity platinumsubstrates.”

Querying, presentation, and visualization tools

Existing Web-based search and retrievalinterfaces, which present the results of a topicor keyword search as a list of URIs, aretotally inadequate for e-science applications.New search, browse, and navigation inter-faces are needed that can present complexmultidimensional, multimedia data in novelways that let scientists detect trends or pat-terns that wouldn’t be revealed through tra-ditional search interfaces.

An earlier research prototype, developedthrough a collaboration between the DSTCand the CWI (the Netherlands’ NationalResearch Institute for Mathematics and Com-puter Science),11 is undergoing refinement for this project. The system dynamically

JANUARY/FEBRUARY 2004 www.computer.org/intelligent 7

Figure 10. The Rules-by-Example interface prototype.

generates multimedia presentations in theSMIL 2.0 (Synchronized Multimedia Inte-gration Language) format by aggregatingretrieved results sets on the basis of thesemantic relationships between them. Thesystem infers the semantic relationships byapplying predefined inferencing rules to theassociated metadata. CWI’s Cuypers systemapplies constraints such as page size, band-width, and user preferences to adapt the pre-sentation components to the user’s device

capabilities and platform.3 In the earlier OpenArchives Initiative-based prototype11 (seeFigure 11), we hardwired the mappings fromsemantic relationships between retrieved dig-ital objects to spatiotemporal relationshipsin the presentations.

In the FUSION project, we’re developing aGUI that lets fuel cell experts interactivelymap semantic relationships to their preferredspatiotemporal presentation modes. Forexample, the GUI can display sequences of

images, in increasing order of magnification,as a slide show. Or, it can tile them horizon-tally or vertically and to the left or right ofthe corresponding performance data. Thedomain experts can interactively define andmodify their preferred modes of layout andpresentation, and their next search’s resultswill reflect their preferences.

For instance, the SMIL visualization tool(the OAI-based prototype we mentioned ear-lier) will help enable a better understanding ofhow fuel cells degrade over time. Sequencesof micrographic images retrieved from iden-tically manufactured fuel cells after increas-ing duration of use could be displayed as ani-mations, synchronized in parallel with thedisplay of corresponding performance data.

When SMIL presentations reveal newinformation or previously unrecognized pat-terns or trends, users will be able to save andannotate them, for later retrieval, as evidenceto support new hypotheses or theories.

Knowledge capture and annotation tools

Assuming that domain experts (in thiscase, materials scientists) elicit new knowl-edge by using our system, we’ll need toolsthat can record the newly extracted informa-tion or ideas and the evidence or contextualinformation to support them.

In the DSTC’s FilmEd project (http://metadata.net/filmed), we’ve been developingthe prototype Vannotea system (see Figure12).12 Using the GrangeNet broadbandresearch network and access grid nodes thatsupport large-scale group-to-group collabo-ration and high-quality audio-video,Vannoteausers can open an MPEG-2 video file orJPEG 2000 image and share tools to collab-oratively segment, browse, describe, anno-tate, and discuss the particular video or imageof interest.

Vannotea is an extension of the W3C’sAnnotea prototype.13 Whereas Annoteaenables the stand-alone annotation of Webpages, Vannotea supports the annotation ofaudiovisual documents (or segments thereof)in a high-bandwidth collaborative environ-ment. We’ve extended Annotea’s RDF-basedannotation schema and Xpointer to supportthe annotation of audiovisual documents.The user specifies an annotation’s contextthrough spatiotemporal extensions toXPointer that enable the location of specificsegments, keyframes, or regions in key-frames. This approach also lets us use exist-ing annotation server implementations such

8 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS

E - S c i e n c e

Interactivesearch

Semanticinferencing

Spatial/temporalmapping

ArchiveLoC

ArchiveUIUO

Archive

Oprn Archives Initiative

RDF model

SMIL presentation

User

Figure 11. The automatic Synchronized Multimedia Integration Language presentationgenerator.

Figure 12. The Vannotea video annotation tool.

as Zope or the W3C Perllib server. These areRDF databases that sit on top of MySQL andprovide their own query language, Algae.

We initially designed Vannotea for anno-tating high-quality MPEG-2 video and JPEG2000 images. However, enabling the collab-orative annotation and discussion of docu-ments of other media types—for example,text, Web pages, audio, and SMIL files—willbe relatively trivial. We envision a demandfor tools such as Vannotea in both FUSION andother e-science projects.14 Such tools willenable the collaborative annotation of tele-microscopic video and images as well as thestorage and annotation of SMIL presenta-tions generated dynamically as a result ofsemantic searches.

This article describes how we’ve beenextending, refining, and applying sev-

eral key Semantic Web technologies to areal-world scientific problem—fuel celloptimization. Although this research is stillin progress, we believe that this uniqueapplication of Semantic Web technologiesto knowledge mining in the materials-engi-neering domain will prove extremely valu-able. Initially, it will help improve the designand performance of solid polymer elec-trolyte fuel cells. In the longer term, it willhelp set new paradigms for knowledge man-agement and sharing across a range of sci-entific and microstructural-engineeringapplications.

AcknowledgmentsThe Distributed Systems Technology Centre

(DSTC), through the Australian Federal Govern-ment’s Cooperative Research Centre Programme(Department of Education, Science and Training),partly funded this research. Thanks to the Univer-sity of Queensland’s Centre for Microscopy andMicroanalysis for the microscopy images used inthe FUSION project. The FilmEd project is part ofthe GrangeNet program, funded by the AustralianFederal Government’s BITS (Building on ITStrengths) Advanced Network Initiative (Depart-ment of Communications, Information Technol-ogy and the Arts).

References

1. J. Rivkin, The Hydrogen Economy: The Cre-ation of the World Wide Energy Web and theRedistribution of Power of Earth, Tarcher Put-nam, 2002.

2. J. Hendler, “Science and the Semantic Web,”Science, vol. 299, no. 5606, 24 Jan. 2003, pp.520–521.

3. J. van Ossenbruggen et al., “Towards Secondand Third Generation Web-Based Multi-media,” Proc. 10th Int’l World Wide WebConf., 2001.

4. ISO/IEC 15938-5 FDIS Information Tech-nology, Multimedia Content DescriptionInterface—Part 5: Multimedia DescriptionSchemes, Fraunhofer Inst. für Integrierte Pub-likations und Informationssysteme, July2001.

5. J. Hunter, “Adding Multimedia to the Seman-tic Web—Building an MPEG-7 Ontology,”Proc. Int’l Semantic Web Working Symp.(SWWS), 2001.

6. J.R. Swedlow et al., “Informatics and Quan-titative Analysis in Biological Imaging,” Sci-ence, vol. 300, no. 5616, 4 Apr. 2003, pp.100–102.

7. N. Noy and D. McGuinness, Ontology Devel-opment 101: A Guide to Creating Your FirstOntology, tech. report SMI-2001-0880, Stan-ford Medical Informatics, Stanford Univ.,2001.

8. C. Lagoze and J. Hunter, “The ABC Ontol-ogy and Model (v3.0),” J. Digital Informa-tion, vol. 2, no. 2, 2001.

9. I. Niles and A. Pease, “Towards a StandardUpper Ontology,” Proc. 2nd Int’l Conf. For-

mal Ontology in Information Systems (FOIS2001), 2001.

10. A. Gangemi et al., “Sweetening Ontologieswith DOLCE,” Proc. 13th Int’l Conf. Knowl-edge Eng. and Knowledge Management(EKAW 2002), LNCS 2473, Springer-Verlag,2002, pp. 166–181.

11. S. Little, J. Geurts, and J. Hunter, “TheDynamic Generation of Intelligent Multi-media Presentations through Semantic Infer-encing,” Proc. 6th European Conf. DigitalLibraries (ECDL 2002), LNCS 2458,Springer-Verlag, 2002, pp. 158–175.

12. R. Schroeter, J. Hunter, and D. Kosovic, “Van-notea—A Collaborative Video Indexing,Annotation and Discussion System for Broad-band Networks,” to be published in Proc. 10thInt’l Multimedia Modelling Conf. (MMM2004), IEEE CS Press, 2004.

13. J. Kahan et al. “Annotea:An Open RDF Infra-structure for Shared Web Annotations,” Proc.10th Int’l World Wide Web Conf. (WWW10),Int’l World Wide Web Conf. Committee,2001; www10.org/cdrom/papers/frame.html.

14. Technology Theme: Collaborative Visualisa-tion, Grid Outreach Programme, 2003, www.gridoutreach.org.uk/themes/cv_intro.htm.

For more information on this or any other com-puting topic, please visit our Digital Library atwww.computer.org/publications/dlib.

JANUARY/FEBRUARY 2004 www.computer.org/intelligent 9

T h e A u t h o r sJane Hunter is a Distinguished Research Fellow at the Distributed SystemsTechnology Cooperative Research Centre. For the past 10 years she has beendeveloping data models, ontologies, metadata standards (Dublin Core, MPEG-7,and MPEG-21), schemas, software tools, and query languages to enable theindexing, archival, analysis, integration, and management of large multimediacollections. She’s also the liaison between MPEG (Moving Pictures ExpertsGroup) and the W3C. Contact her at DSTC, Level 7, GP South, Univ. ofQueensland, St Lucia, Queensland 4072, Australia; jane@dstc.edu.au.

John Drennan is the director of the University of Queensland’s Centre forMicroscopy and Microanalysis. An expert on the microstructural/propertyrelationships in ion-conducting solids, he has published over 90 refereedpapers in the field and has made a significant impact on the microstructuraldesign of solid electrolyte materials for fuel cell applications. Contact him atthe Centre for Microscopy and Microanalysis, Univ. of Queensland, St. Lucia,Queensland 4072, Australia; j.drennan@mailbox.uq.edu.au.

Suzanne Little is a PhD student at the School of Information Technologyand Electrical Engineering at the University of Queensland. Her currentresearch focus is the application of Semantic Web technologies to multimediaindexing and querying. She has an honours degree in information technologyfrom the University of Queensland. Contact her at the School of InformationTechnology and Electrical Eng., Univ. of Queensland, St Lucia, Queensland4072, Australia; s.little@dstc.edu.au.