Analysis, Diagnosis, Planning and Simulation of Industrial ... · 3 Keynote by Stelios Orphanoudakis, ERCIM President 4 Tribute to Gerard van Oortmerssen by Stelios Orphanoudakis,

European Research Consortium for Informatics and Mathematicswww.ercim.org

Number 56, January 2004

Analysis, Diagnosis,Planning and Simulationof IndustrialSystems

Analysis, Diagnosis,Planning and Simulationof IndustrialSystems

S p e c i a l :S p e c i a l :

Next issue: April 2004Special theme: Games Technology

Cover: Train station in Borlänge, Sweden (photo: Green Cargo). Efficientresource utilisation is a key factor in the railway industry See article by Martin Aronsson onpage 27.

Contents

JOINT ERCIM ACTIONS

3 Keynoteby Stelios Orphanoudakis, ERCIM President

4 Tribute to Gerard van Oortmerssenby Stelios Orphanoudakis, ERCIM President

4 Ronald de Wolf - Winner of the 2003 Cor Baayen Award

5 Possibilities and Limitations of Quantum Computing

7 First ERCIM Workshop on Informatics and MathematicsApplied to Interventional Medicineby Marc Thiriet, INRIA

8 News from W3C

SPECIAL THEME

10 Industrial Diagnosis, Planning and Simulation —Introductionby Per Kreuger, SICS

12 Efficient Algorithms for Inverse Magnetic Field Problemsby Domenico Lahaye, CWI

13 Deviation Detection of Industrial Processes by Anders Holst, Jan Ekman and Daniel Gillblad, SICS

14 Bidding Agents Acquire Cargo in Online Auctionsby Pieter Jan 't Hoen, CWI

15 Decision Support System for Process Analysis and Supervisionby Galia Weidl, University of Stuttgart, Germany

17 Mathematical Challenges in the Electronics Industryby Wil Schilders, Eindhoven University of Technology, The Netherlands

18 Contpack — Maximising the Volume Utilisation of Containersby Onno Garms, Ralf Heckmann and Stefan Rank, Fraunhofer SCAI

20 First Commercial Bayesian Software for IntelligentTroubleshooting and Diagnosticsby Claus Skaanning, Dezide, Denmark

22 Innovating Shoe Manufacturing using AdvancedSimulation Techniquesby Emanuele Carpanzano and Andrea Cataldo, ITIA-CNR

23 Billy Goat Detects Worms and Virusesby James Riordan and Diego Zamboni, IBM Zurich ResearchLab/SARIT

24 Decision Support for Personnel and ResourcesManagement in Multi-Project Environmentsby Antonia Bertolino, Eda Marchetti and Raffaela Mirandola,ISTI-CNR

26 Towards Useful Tools for Estimating RailwayInfrastructure Capacity by Jan Ekman, SICS

27 Efficient Utilisation of Resources in the Railway Industryby Levelling Departures and Arrivals of Trainsby Martin Aronsson, SICS

28 Real-Time Optimisation of Empty Rail Freight Car Distributionby Martin Joborn, Carmen Consulting, Sweden

29 KUORMA: A Collection of APS-Algorithms for ForestIndustry Wood Transportby Juha Savola, VTT

31 Parc Bandwidth on Demandby Helmut Simonis, Parc Technologies Ltd, UK

32 DesParO - A Design Parameter Optimisation Toolboxusing an Iterative Kriging Algorithmby Christof Bäuerle, Clemens-August Thole and Ulrich Trottenberg, Fraunhofer SCAI

R&D AND TECHNOLOGY TRANSFER

45 Solving Complex Problems with Advanced Techniquesby Enrique Alba, University of Malaga/SpaRCIM

47 Nipper – the Neat Internet Protocol Packet EditorSami Pönkänen and Marko Lyijynen, VTT

48 On Describing B2B Processes with Semantic WebTechnologiesby Santtu Toivonen, Tapio Pitkäranta, VTT and Jung Ung Min,Stanford University, USA

50 Tracking Assets and Vulnerabilities in Corporate Networks by Dieter Gantenbein, IBM Zurich Research Lab/SARIT

51 Numerical Simulation through Web Servicesby Stephan Springstubbe, Jürgen Klein and Ottmar Krämer-Fuhrmann, Fraunhofer SCAI

53 Advanced Communication Networksby Michel Mandjes, CWI

54 Information Security Research at NTNU by Stig F. Mjølsnes and Ingvild Ytrehus, NTNU

55 Supervised Term Weighting for Automated TextCategorizationby Franca Debole and Fabrizio Sebastiani, ISTI-CNR

56 Remote Monitoring of Health Conditionsby Dirk Husemann, IBM Zurich Research Lab/SARIT

EVENTS

57 FM 2003 – 12th International Symposium on FormalMethodsby Stefania Gnesi, ISTI-CNR

58 ERCIM Workshop on Soft Computing 2003by Petr Hájek, Institute of Computer Science, Academy ofSciences of the Czech Republic/CRCIM

58 SC 2003 — Industrial Simulation Conferenceby Ana Pajares, Juan C. Guerri and Carlos Palau, UniversidadPolitécnica de Valencia/SpaRCIM

58 Announcements

61 IN BRIEF

33 On-line Testing of the Reactor Protection System in the Paks Nuclear Power Plantby István Varga, Tamás Bartha and Alexandros Soumelidis,SZTAKI

35 Process Systems: Theory and Applications from Different Aspectsby Katalin Hangos and Gábor Szederkényi, SZTAKI

37 Multi-Agent System Technology in a Port ContainerTerminal Automationby Vicent J. Botti, Universidad Politécnica de Valencia/SpaRCIM

39 Contribution to Fault Detection and Diagnosis: Mixture-based Modellingby Tatiana V.Guy, Dani Juriãiã, Miroslav Kárny and AndrejRakar, Institute Information Theory and Automation, Academyof Sciences of the Czech Republic/CRCIM

40 Data Mining Applied in Anaerobic Wastewater Treatmentby Simon Lambert, CCLRC

41 Integrated Multivariate Statistical Process Control and Condition Monitoring for Fed-batch FermentationProcessesby Hongwei Zhang, Northeast Wales Institute of HigherEducation, UK

43 Saving Energyby Claude Lemaréchal, INRIA

ERCIM News No. 56, January 2004 3

KEYNOTE

In 2004, ERCIM will celebrate its fifteenth anniversary, andit is a time in which to take pride in the ERCIM commu-nity’s many achievements. ERCIM brings together

research institutions from eighteen different countries inEurope, and encompasses more than 12 000 researchers.ERCIM is a de facto premier Network of Excellence in theemerging European Research Area, and adistributed think tank in both informationand communications technologies (ICT)and applied mathematics (AM). ERCIMWorking Groups are currently active inmany significant areas of research withinICT and AM. ERCIM is well positioned tobenefit Europe through strategiccooperation with the EuropeanCommission, the European ScienceFoundation, the National ScienceFoundation in the US, and other interna-tional organisations. ERCIM News is animportant publication with a wide distribu-tion in both paper and electronic form.ERCIM’s high-profile fellowshipprogramme provides young researchersaround the world with the opportunity toundertake cutting-edge research at ERCIMmember institutions and to become familiarwith the diversity of European cultures.Furthermore, the hosting of W3C in Europeis an important activity not only for ERCIM, but also forEuropean industry and the European research community atlarge. ERCIM is developing quickly, and as it matures, it hasthe potential to become much more than the sum of its parts.However, ERCIM is currently at a crossroads, and many chal-lenges and opportunities lie ahead. We must meet these chal-lenges and exploit every opportunity in order to enhanceEurope’s innovation capacity and competitiveness in the globalresearch arena.

ERCIM is currently an organisation of rather loosely coupledmembers. This coupling is primarily achieved through interac-tions at Working Group level, joint participation in EU projectsand Networks of Excellence, and the dissemination throughERCIM News of R&D results and other developments frommember institutions. Nevertheless, cohesion at the overallERCIM level could be further strengthened. ERCIM mustbetter define its own identity and deliver a clear message to itsmembers, to the broader European research community and toEuropean industry on what may be expected as ‘ERCIM addedvalue’. While ERCIM has this issue under constant considera-tion, it is under some pressure and perhaps feels the need toexpand with more members and associate members. However,unless ERCIM first acquires a clearer identity and formulates astrategic plan for growth, innovation and competitiveness,premature and rapid expansion may simply dilute its effortsand make the issue of ‘ERCIM added value’ more difficult toresolve. In the meantime, ERCIM member institutions ought tobe strongly encouraged to promote ERCIM links within theresearch community of the country they represent.

Additionally, ERCIM should further develop its own researchstrategy and act as a catalyst for enhancing the research effortsof its members, while striving to improve and sustain its humanresource potential and research infrastructure. ERCIM canbecome an incubator for the creation and validation of newresearch ideas in Europe, particularly in interdisciplinary areas

where complementarity of skills and ideasis essential. Teams of excellence in selectedthematic areas already exist within ERCIMand could be opened to researchers in otheracademic and research institutions, thuscontributing to the creation of long-lastingrelationships in European and globalresearch. Given adequate financial support,ERCIM has the potential to play a signifi-cant part in a research infrastructure, whichis greatly needed if the European ResearchArea is to become a reality.

This is a difficult period for fundamentalresearch worldwide, and in Europe inparticular. We need to ensure that inEurope, with its long-standing tradition ininnovative research, fundamental researchwill remain alive and will be sustained.ERCIM is in a position to play a significantrole in this, as it can provide the necessarycritical mass and generate much-needed

momentum. Fundamental research supports the R&D effortthat is needed to sustain the competitiveness of Europeanindustry in the long term, and it cannot be allowed to lagbehind. ERCIM can serve as an example for cooperationbetween research institutions and industry in Europe, and inaddition should continue to work towards convincing industry,which is also going through a difficult period, that there arebenefits in such a long-term cooperation.

ERCIM can refine its strategy for cooperation with theEuropean Commission, the European Science Foundation, theNational Science Foundation and other international organisa-tions. In order for such partnerships to be mutually beneficial, itis necessary to define in more detail what ERCIM has to offer,as well as what it can expect. Finally, ERCIM is uniquely posi-tioned to play an important role in technology and research-based education and training, thus contributing to the employa-bility of Europe’s workforce in the emerging InformationSociety. A strategy and a timetable are required here as well, ifERCIM is to take advantage of this opportunity to establish theERCIM label in the education and training arena at Europeanlevel.

ERCIM is ready to meet the scientific, organisational, andfinancial challenges lying ahead and to act as a catalyst forstrengthening European excellence in current and emergingresearch areas.

Stelios Orphanoudakis

Stelios Orphanoudakis, Director of

the Foundation for Research and

Technology - Hellas (FORTH), and

President of ERCIM.

4 ERCIM News No. 56, January 2004

JOINT ERCIM ACTIONS

On 31 December, 2003, Gerard van Oortmerssen completedhis term as President of ERCIM, after five and a half years ofdistinguished service to the organisation and its member insti-tutions. In May 1998, Gerard van Oortmerssen succeeded CorBaayen and Dennis Tsichritzis to become ERCIM’s third pres-ident, and was re-elected for a second term in 2001. InNovember 1999, Gerard hosted and presided over ERCIM’s10th Anniversary celebration in Amsterdam, a memorableevent whose success was characterised by a high-levelsymposium celebrating ERCIM’s significant contributions toEuropean research.

During Gerard van Oortmerssen’s presidency, ERCIMexpanded its membership and geographical coverage toSlovakia, Ireland, Austria, Norway, Luxembourg, Spain andBelgium, with the joining of SRCIM (May 1998), Trinity College(May 2000), AARIT (October 2001), NTNU (January 2002),FNR (July 2002), SpaRCIM (July 2003) and FWO-FNRS(December 2003) respectively. Furthermore, Gerard activelypursued stronger ties with the European Commission, theEuropean Science Foundation and the US National ScienceFoundation in order to promote ERCIM’s role as a distributedthink tank in information and communications technology andapplied mathematics, as well as a de facto premier Network ofExcellence in Europe and throughout the world. In pursuit ofthe same goal, Gerard adopted a proposal made by the thenPresident of INRIA and Manager of ERCIM, BernardLarrouturou, for ERCIM to become the European host of W3C,which came into effect in January 2003. Within ERCIM,Gerard established the Strategy and Human Resource

Management Taskforces, which have become instrumental instrategic planning within ERCIM and in promoting the sharingof information and experience among ERCIM member institu-tions on the mobility of researchers and the management ofhuman resources.

Gerard van Oortmerssen leaves ERCIM a stronger organisa-tion and an important player in the emerging EuropeanResearch Area. Under his leadership, ERCIM has takenimportant steps towards becoming an open ICT network inEurope and bridging the gap between the research communityand industry. This was achieved through its strong relationswith startup and other spin-off companies affiliated with itsmember institutions. Gerard has undoubtedly put his mark onERCIM, and I am certain that I speak on behalf of the entireERCIM community when I say he will be missed.

Stelios Orphanoudakis, ERCIM President

Gerard van

Oortmerssen was

ERCIM's President

from May 1998 to

December 2003. He

has left CWI as

managing director as

per 1 May 2003. He is

now director of TNO

Telecom.

Tribute to Gerard van Oortmerssen

Ronald de Wolf did his PhD in the areaknown as quantum informationprocessing and quantum computing. It isremarkable that as an undergraduate hehad already written, together withNienhuys-Cheng, the standard text on acompletely different topic calledInductive Logic Programming. Thisbook alone figures as a very impressivePhD thesis. De Wolf received PhDplacement offers from MIT and CMU,but decided instead to study at the CWI.

Quantum information processing is anew and important field that combinesthe physical laws of quantum mechanicswith those of computation, resulting in anew model of computation: the quantumcomputer. The field gained momentumwhen Peter Shor demonstrated the speedwith which such quantum computerscould factor large numbers. Most currentcryptography, including electroniccommerce, is based on the fact that nosufficiently fast factorisation method isknown for classical computers. Shor thus

demonstrated that if one could build aquantum computer then most of our cryp-tographic protocols could be broken.

It is this societal relevance that makesquantum information processing soimportant. Both North American andEuropean funding agencies have setaside significant amounts of money toinvest in this new technology.

Ronald's work received the highest praisein the international community and

The 2003 ERCIM Cor Baayen Award was presented to Ronald de Wolf during aceremony in Luxembourg on 5 November 2003. The 5000 Euro annual award isgiven to a most promising young researcher having completed a PhD thesis incomputer science and applied mathematics, in one of the ‘ERCIM countries’.

Ronald de Wolf — Winner of the 2003 Cor Baayen Award


JOINT ERCIM ACTIONS

during his PhD he received frequentinvitations to speak at internationalmeetings and conferences. Amongrelated fields, his PhD thesis ranksamong the best worldwide over thelast ten years. Ronald's work had abig impact on the field of quantuminformation processing, and parts ofhis PhD thesis are now standardinclusions in textbooks on thesubject. For example, the impossi-bility method he developed to showthe limitations of quantumcomputers has become a standardtechnique and is taught around theworld in any class on quantumcomputing. New protocols and algo-rithms, such as the quantum finger-printing technique, have been usedby groups such as those at MIT andBerkeley to develop new cryptographicprimitives like quantum digital signa-tures. These advances also attractedsignificant attention from the press, witharticles in Physics News Update, Wired,Technology Research News,Süddeutsche Zeitung and others. The

work on locally decodable error-correcting codes baffled the best-knownresearchers in this classical area ofcomputer science. The elegant idea ofusing quantum mechanics as a way ofproving a classical statement is highlyoriginal, and was considered a break-

through result in the area of locallydecodable codes.

Ronald is one of the very few trulybrilliant students that his professorhas seen. He is able to combine greatmathematical skill with intensecreativity in order to solve some ofthe hardest problems in his field.

The following researchers werenominated for the 2003 Cor BaayenAward: Pekka Abrahamsson,Finland; Giuseppe Amato, Italy;Steffi Beckhaus, Germany; PatriciaBouyer, France; Fredrik Espinoza,Sweden; Andras Gyorgy, Hungary;Keijo Heljanko, Finland; VictorKhomenko, UK; Sebastian Mika,Germany; Lars Rasmusson, Sweden;

Zoltan Szabo, Hungary; GerardoHermosillo Valadez, France; RuiliWang, Ireland; Ronald de Wolf, TheNetherlands; Albertus P. Zwart, TheNetherlands.

More information: http://www.ercim.org/activity/cor-baayen.html

The mantra of quantum computing is thestatement 'information is physical'.Information processing and computationare physical processes, and hence subjectto the laws of physics. To the best of ourknowledge, these laws are quantummechanical - and quite weird. In contrast,most current computer science is basedon models of classical physics. Forexample, a Turing machine's read-writehead is always in a specific position andits memory cells contain definite infor-mation (0 or 1). In the 1980s, RichardFeynman and David Deutsch incorpo-rated the laws of quantum mechanics intothe model. This gave quite a differentpicture. In quantum computing the

memory cells can be in a 'superposition'of 0 and 1 at the same time, and differentcomputational paths can 'interfere' witheach other destructively or construc-tively, much like waves.

Roughly speaking, a quantum algorithmworks by going through many computa-tional paths simultaneously, in superposi-tion, and using interference to strengthenthe paths leading to a correct output.Halfway through the 1990s researcherslike Shor and Grover discovered quantumalgorithms for factoring integers andsearching databases that are more efficientthan their classical equivalents. Thesediscoveries made quantum computing one

of the hottest fields around. Whetherquantum computers can actually be built isthe focus of much research in experi-mental physics. However, many peoplefeel this effort will bear fruit regardless ofthe outcome. Even if the construction of aquantum computer is impossible, theresearch has led to valuable new insightsin quantum physics.

The CWI group centred around PaulVitányi and Harry Buhrman was one ofthe first in Europe to take up quantumcomputing as a serious research topic. In1997 de Wolf became a PhD studentthere and at the University ofAmsterdam. Around that time nobody

Possibilities and Limitations of Quantum ComputingThis year's Cor Baayen Award was given to CWI researcher Ronald de Wolf. Hereceived the award for his work on the potential computer of the future: thequantum computer. De Wolf proved several limitations of this revolutionarycomputation model. He discovered new quantum algorithms that are more efficientthan classical algorithms. Furthermore, he successfully applied quantum prooftechniques on classical algorithms.

Ronald de Wolf (left) receives the Cor Baayen Award

from ERCIM President Gerard van Oortmerssen. On

the right is Erna Hennicot-Schoepges, Luxembourg

Minister for Culture, Higher Education and

Research, who welcomed the participants to the

ERCIM meetings in Luxembourg.


JOINT ERCIM ACTIONS

knew whether quantum computers weresuperior to their classical equivalents.Together with co-authors, particularlyHarry Buhrman, de Wolf proved variousstrong limitations on quantumcomputers. For most problems they arenot significantly faster than classicalcomputers. These limitations wereproved by reducing complexity theoreticquestions to algebraic questions aboutdegrees of multivariate polynomials.Sufficient prove of strong -often optimal-lower bounds on the time a quantumalgorithm needs to compute a Booleanfunction, can be given by proving a lowerbound on the degree of an n-variate poly-nomial approximating that function.

Moreover, de Wolf also contributed tothe discovery of some quantum algo-rithms and protocols that outperformtheir classical counterparts. One exampleof this is a 'quantum fingerprinting'scheme. It allows two separated parties tocompare large chunks of data more effi-ciently than classical computers. Byassigning small quantum states to longclassical strings, the amount of data thathas to be exchanged for this operationcan be exponentially reduced. In the

future this technique could for examplebe used to create digital autographs.

After receiving his PhD in September2001, de Wolf moved to the USA tobecome a postdoc at UC Berkeley. Therehe showed, together with Berkeley gradstudent Iordanis Kerenidis, that quantumcomputing techniques could be used as atool to prove new results in classicalcomputer science. Specifically, theyanalyzed 2-query locally decodableerror-correcting codes. This schemeallows each encoded bit to be individu-

ally decoded by looking at only two bitsof the codeword, instead of firstdecoding the entire codeword. Suchlocally decodable codes have been thefocus of much research recently, but theprevious best general lower bounds ontheir length were only polynomial. Usingtechniques from quantum computing,Kerenidis and de Wolf established forthe first time an exponential lowerbound. No proof of this result is knownthat uses just classical proof techniques.

This classical-theorem-with-a-quantum-proof significantly adds to the relevanceof quantum computing within computerscience in general, since its impact doesnot depend on the actual building of aquantum computer. As a bonus, thequantum techniques yield quantumprotocols for the cryptographic task of'private information retrieval' that aremore efficient than the best known clas-sical protocols. Currently, de Wolf worksat CWI as a postdoc. His main researchfocus is still quantum computing.

Please contact:Ronald de WolfTel: +31205924078E-mail: [email protected]://homepages.cwi.nl/~rdewolf

Quantum computers strengthen the

computational paths leading to correct

solutions. This can be compared to

constructive interference of waves.

Pho

to c

ourt

esy

of J

ohn

K.N

. Mur

phy,

ww

w.h

otqu

anta

.com

Call for Candidates

The Cor Baayen Award, awarded to a most promising youngresearcher in computer science and applied mathematics, wascreated in 1995 to honour the first ERCIM President, is opento any young researcher having completed their PhD thesis inone of the 'ERCIM countries': Austria, Belgium, CzechRepublic, Finland, France, Germany, Greece, Hungary,Ireland, Italy, Luxembourg, Norway, Slovakia, Spain, Sweden,Switzerland, The Netherlands and the United Kingdom.

The award consists of a cheque for 5000 Euro together withan award certificate. The selected fellow will be invited to theERCIM meetings in autumn 2004. A short article on thewinner, together with the list of all candidates nominated, willbe published in the ERCIM News.

Rules for NominationNominations from each country are made by thecorresponding ERCIM Executive Committee member (alsoreferred to as 'national contact'). Those who wish that aparticular candidate be nominated should therefore contactthe ERCIM Executive Committee member for their country(see www.ercim.org/ contacts/ execom/). Nominees musthave carried out their work in one of the 'ERCIM countries'.

Nominees must have been awarded their PhD (or equivalent)not more than two years prior to the date of nomination. EachERCIM institute is allowed to nominate up to two candidatesfrom its country. A person can only be nominated once for theCor Baayen Award. The selection of the Cor Baayen award isthe responsibility of the ERCIM Executive Committee.

Submitting a NominationTo submit a nomination to your national contact, fill out theCor Baayen Award nomination form (available at the URLgiven below) and provide a copy of the candidate's PhDthesis as well as copies of the candidate's best papers (max.5), preferably provided as links to electronic documents.

Deadlines• Nominations are to be received by the national contacts by

15 April, 2004.• National contacts are to send their two selected

nominations to the coordinator by 30 April, 2004.

Further information can be obtained from your national contactor from the Cor Baayen Award coordinator Lubos Brim,Masaryk University Brno/CRCIM, ([email protected]).

Cor Baayen Award 2004


JOINT ERCIM ACTIONS

Participants attended from Canada,Chile, France, Germany, Greece, Italy,Spain, Switzerland and the UK. The setof presentations covered the necessarytopics involved in the development ofmedical simulators.

Medical image processing and virtualreality was discussed by L. Soler, whopointed out the usefulness of navigationtools and medical robotics to ourcommunity, a fact which had not hithertobeen widely appreciated.

Three-dimensional reconstructions ofanatomical organs are a popular researchtheme (talks from N. Gabrielides,National Technical University ofAthens, D. Manoussaki working atFORTH, J. Peiro, Imperial CollegeLondon and M. Thiriet for the INRIAteams EPIDAURE and REO). Medicallyoriented flow computations must beperformed in image-based discretiseddomains due to the huge between-subject variability in vessel anatomy andto the patient-dependent shape of thevessel-wall pathologies. Once the three-dimensional reconstruction is made, themesh must be compatible with theconstraints of scientific computing(presentation of L. Baffico, Universidadde Chile and M. Thiriet). Furthermore,mesh adaptation and adaptivity arenecessary in the case of unsteadyphenomena such as cryotherapy for livertumours (talk from Youssef Belhamadia,GIREF, Université Laval, Quebec,Canada).

Biofluid flow modelling must also takeinto account the deformation of thevessel wall during pressure-wave propa-gation or environment loading, in partic-ular changes in external pressure duringrespiration, and especially in thoracicvessels. The rheology of the movingfluid and of the bio-conduit wall must be

determined in vivo. At the same time,suitable software must take into accountthe complete set of rheological proper-ties of the composite bio-tissues, withpossible structure changes (presentationof A. Pascau, University of Zaragoza,Spain on Non-Newtonian fluids and M.Thiriet). Due to the geometry of thevessels and the mechanics of the walland flow genesis, the bioflow modelsbeing developed are three-dimensional,generally laminar and quasi-periodic(talk from Giuseppe Pontrelli, CNR,Roma and M. Thiriet). Moreover,although it belongs to a network with thecardiac pump and its input and outputimpedances, the three-dimensionalregion of interest is rather limited.Multiphysics multiscale modelling isalso being investigated (presentation ofM. Fernandez, Ecole PolytechniqueFédérale de Lausanne and J. Peiro).Lumped parameter models may be usedfor various regions of the vesselnetwork.

New procedures are also being devel-oped however, including the reducedbasis element method with fractal andmultiphysics homogenisation (talk fromY. Maday, Université Pierre et MarieCurie, Paris). Such complete modellingis associated with a large computationalcost. Fast solvers are thereforenecessary, using geometric and algebraicmultigrid and domain decompositiontechniques, and coupling of sequentialand parallel codes (presentation of W.Joppich, Fraunhofer-Institute forAlgorithms and Scientific Computing(SCAI), Germany).

Mass transport by the fluid conveyed inbio-conduits is investigated in order tounderstand vessel-wall lesions and todevelop new local or bulk therapystrategy. Aerosol transport and deposi-tion in the airways still needs to be accu-

rately modelled in the physical sense(talk from P.-E. Jabin, Ecole NormaleSupérieure, Paris). Mini-invasivemedical procedures use the body’snatural paths, the vessels. Interventionalmedicine treats vessel-wall damageswith medical devices, which aredisplaced through the vascular bed anddeployed within the lesion usingcatheter-based techniques. The designand shape optimisation of such devices,as well as ventricule-assist pumps iscurrently of major interest (presentationof M.-I. Farinas, Ecole Polytechnique deMontréal, Canada).

During the final session it was decidedthat the next IM2IM workshop would beheld in Marseille in August 2004, inorder to coincide with a six-weeksummer school in biomathematics. It isplanned to hold the following workshopat IRCAD in Strasbourg in December2005. The Working Group intends tosubmit a proposal under the EuropeanCommission’s Future and EmergingTechnologies scheme to drive theWorking Group in a precise direction.

Link: http://www-rocq1.inria.fr/Marc.Thiriet/Im2im/

Please contact:Marc Thiriet, INRIAE-mail: [email protected]

First ERCIM Workshop on Informatics andMathematics Applied to Interventional Medicineby Marc Thiriet

The first workshop of the IM2IM ERCIM Working Group was held from 1-2 December2003 at CRP Henri Tudor in Luxembourg, as part of the ERCIM biannual meetingshosted by the Luxembourg National Research Fund (FNR).


XForms: the Next Generation of Web Formsby Steven Pemberton, CWI and W3C

HTML Forms formed the basis of the e-commerce revolution.Now after 10 years experience with them, W3C has releasedXForms, the next generation of Web Forms. In contrast toHTML forms, in which functional and presentation markup areintertwined, XForms lets forms authors distinguish the descrip-tions of the purpose of the form, the presentation of the form,and how the results (the instance data) are written in XML.

The design of XForms is based on an analysis of how HTMLForms have been used in the last ten years, and what could bechanged to make them even more useful. In brief, XForms letsyou do everything that HTML Forms can do, and more, such as:• check data values while the user is typing them in• indicate that certain fields are required, and that the form

cannot be submitted without them• constrain values in certain ways, such as requiring them to

be in a certain range• submit forms data as XML• integrate with Web services, for instance by using SOAP

and XML RPC• submit the same form to different servers (for instance a

search string to different search engines)• save and restore a form to and from a file• use the result of a submit as input to a further form• get the initial data for a form from an external document• calculate submitted values from other values• build 'shopping basket' and 'wizard' style forms without

needing to resort to scripting.

Properties of XFormsIn detail, XForms has several important properties:• XForms improves the user experience:XForms has been

designed to allow much to be checked by the browser, suchas types of fields being filled in, that a particular field isrequired, or that one date is later than another. This reducesthe need for round trips to the server or for extensive script-based solutions, and improves the user experience by givingimmediate feedback on what is being filled in.

• It makes it easier to author complicated forms:BecauseXForms uses declarative markup to declare properties ofvalues, and to build relationships between values, it is mucheasier for the author to create complicated, adaptive forms,without having to resort to scripting. An HTML Formconverted to XForms looks pretty much the same, but whenyou start to build forms that HTML wasn't designed for,XForms becomes much simpler. See XForms for HTMLAuthors for some examples.

• It is XML, and it can submit XML: XForms is properly inte-grated into XML: it is in XML, the data it collects in theform is XML, it can load external XML documents as initialdata, and can submit the results as XML. By including theuser in the XML pipeline, it at last means you can have end-to-end XML, right up to the user's desktop.

• It combines existing XML technologies:Rather than rein-venting the wheel, XForms uses a number of existing XMLtechnologies, such as XPath for addressing and calculatingvalues, and XML Schema for defining data types. This has adual benefit: ease of learning for people who already knowthese technologies, and the ability for implementors to useoff-the-shelf components to build their systems.

• It is device independent:The same form can be deliveredwithoutchange to a traditional browser, a PDA, a mobilephone, a voice browser, and even some more exoticemerging clients such as an Instant Messenger. This greatlyeases providing forms to a wide audience, since forms onlyneed to be authored once.

• It is internationalized:Because the data submitted is XML,it is properly internationalized.

• It is accessible:XForms has been designed so that it willwork as well with accessible technologies (for instance forblind users) as with traditional visual browsers.

ImplementationsThe XForms Working Group that created XForms containsrepresentatives from many major computing companies. Ontop of this there are a large number of implementationsemerging. In fact XForms is the most-implemented W3Cspecification ever at this stage in its life-cycle.

There are several implementations that let you use XFormswith existing browsers, either with plugins or by transformingthe forms at the server, as well as a number of browsers thatimplement XForms natively.

At the time of writing, XForms has been demonstrated onPDAs and mobile phones using proxies, using the same tech-niques that are often used to present web content on smallerdevices. We anticipate that native XForms clients will appearon mobile devices soon.

There is a list of XForms Implementations that contains detailsof many of these implementations; an editorial reviewincluding screen shots of several of them can be found atwww.xml.com/pub/a/ 2003/09/10/xforms.html. XForms doesnot need any extra support from servers: it works with allcurrent servers. Although XForms has several new submissionmethods compared with HTML Forms, these are all standardHTTP methods for sending data over the network. XForms canuse 'PUT' to put data to a server, can talk to a SOAP or XMLRPC servers, but also supports legacy formats allowing formsto work with all existing forms servers.

Finally, some large user populations are beginning to emerge,including some government e-government initiatives, a coun-try's tax service, and a country's insurance industry.

Steven Pemberton is a senior researcher at the CWI inAmsterdam. He is chair of the W3C XForms Working Group.

Links:XForms home page: http://www.w3.org/MarkUp/FormsIntroduction to XForms for HTML Authors:http://www.w3.org/MarkUp/Forms/2003/xforms-for-html-authors.htmXForms implementations:http://www.w3.org/MarkUp/Forms/#implementations

News from W3C


W3C Launched Spanish Office W3C is pleased to announce the opening the W3C SpanishOffice, based at the Fundación para el Fomento de laInvestigación Científica y la Tecnología (FICYT) in Oviedo.Although the primary goal of this Office is to be dedicated tooutreach in Spain, this is also the first W3C Office with activecontacts with Spanish-speakers worldwide. The new W3CSpanish Office will be of help in coordinating with LatinAmerican communities by also disseminating information inLatin America, encouraging and co-ordinating new transla-tions to Spanish, etc.

The W3C Spanish Office joins the complement of W3CEuropean Offices in The Benelux Countries, Finland, Germany& Austria; Greece; Hungary, Italy; Sweden; and the UnitedKingdom and Ireland. Francisco Sanchez is Office Manager.The opening ceremonies were held on 20 October in Oviedo.

Links:W3C Spanish Office Web site: http://www.w3c-es.org/Press release: http://www.w3.org/2003/10/spoffice-pressrelease

W3C Requests Eolas Patent ReviewOn 29 October the W3C has presented the United StatesPatent and Trademark Office with prior art establishing thatUS Patent No. 5,838,906 (the '906 patent) is invalid andshould therefore be re-examined in order to eliminate thisunjustified impediment to the operation of the Web, becausethe critical prior art was neither considered at the time thepatent was initially examined and granted, nor during recentpatent infringement litigation.

In an unprecedented step, Tim Berners-Lee, W3C Directorand inventor of the Web, sent a letter to Under SecretaryRogan requesting that his office reinvestigate the matter."W3C urges the USPTO to initiate a reexamination of the '906patent in order to prevent substantial economic and technicaldamage to the operation of World Wide Web," stated Berners-Lee. "The impact of this patent will be felt not only by thosewho are alleged to directly infringe, but all whose web pagesand application rely on the stable, standards-based operationof browsers threatened by this patent. In many cases, thosewho will be forced to incur the cost of modifying Web pagesor software applications do not even themselves infringe thepatent - assuming it is even valid."

US Patent 5,838,906 'Distributed hypermedia method forautomatically invoking external application providing interac-tion and display of embedded objects within a hypermediadocument' was filed in October 1994 and granted inNovember 1998. The patent, owned by the University ofCalifornia with licensing handled through EolasTechnologies, covers mechanisms for embedding objectswithin distributed hypermedia documents, where at leastsome of the object's data is located external to the document,and there is a control path to the object's implementation to

support user interaction with the object. The implementationcan be local or distributed across a network, and is automati-cally invoked based upon type information in the document orassociated with the object's data.

The practical impact of withholding unrestricted access to thepatented technology from use by the Web community will beto substantially impair the usability of the Web for hundredsof millions of individuals in the United States and around theworld. Nearly every Web user relies on plug-in applicationsthat add services such as streaming audio and video, advancedgraphics and a variety of special purpose tools. On 30October, the US PTO have announced that they will beconducting a re-examination of all claims relating to USPatent 5,838,906 citing "a substantial outcry from awidespread segment of the affected industry."

Links: 20 Oct. 03 press release: http://www.w3.org/2003/10/28-906-briefingTimBerners-Lee's letter: http://www.w3.org/2003/10/27-rogan.htmlPrior art filing: http://www.w3.org/2003/10/301-filing.htmlFAQ: http://www.w3.org/2003/09/public-faq.htmlW3C Holds Ad Hoc Meeting on Recent Court Decision, Launches Public Discussion List: http://www.w3.org/2003/08/patent

W3C Held its First Outreach Event in Mainland ChinaThe China International Forum on WWW's Development2003 was held in Beijing on 12-13 November. Attendeesdiscussed the future of the Web, accessibility, SVG, themobile Web, the Semantic Web and internationalization. Theevent was co-organized by the China Computer Federationand the W3C Office in Hong Kong.

Links:China International Forum on WWW's Development 2003 Web site: http://www.w3c.org.hk/CIFWeb03/index.htmlPress Release: http://www.w3.org/2003/10/beijingevent-pressrelease

Last W3C Recommendations• Portable Network Graphics (PNG) Specification (Second

Edition): First published 1 October 1996, revised 10November 2003, David Duce

• XForms 1.0: 14 October 2003, Micah Dubinko, Leigh L.Klotz, Roland Merrick, T. V. Raman

• XML Events: 14 October 2003, Shane McCarron, StevenPemberton, T. V. Raman

An exhaustive list of all W3C Technical Reports: http://www.w3.org/TR/

ERCIM is the European host of W3C.

SPECIAL THEME: Industrial Diagnosis, Planning and Simulation



Recently at SICS, the Industrial Applications (IAM) Laboratorywas formed around a novel topic in applied research. We broughttogether a number of researchers with expertise in mathematicsand computer science, who share the conviction that the studyof the applicability of techniques and methods developed withinthe scientific community should itself be the subject of research. The activities of such a group will, of course, always be influ-enced by the backgrounds of the people participating in it, notonly with regard to their specific scientific expertise but also,and in this case significantly, by their familiarity with practicalproblems outside the scope of their scientific schooling. Thegroup formed at SICS brings together researchers from thefields of combinatory optimisation, scheduling, flow theory,constraint programming, statistical methods for data analysisand learning systems. They have domain expertise in processand manufacturing industry and resource management, moni-toring and optimisation in transportation and logistics.

The initiative behind the formation of this group came from theobservation that methods and techniques from these scientificdisciplines often have the potential to improve both the effi-ciency of resource usage and the quality of products andservices, as well as reducing the costs associated with produc-tion. On the other hand, not all methods proposed byresearchers scale well to the size and diversity of industrialproblems.

The fact that techniques and methods developed in science areused to solve practical problems is, of course, nothing new.However, while this is the basis of all technology, the quality ofthe interplay between science and technology within a societydetermines to a large extent its potential for growth and devel-opment. In the case of computing science, which is the study ofalgorithms, their properties and use, one must always be carefulto verify results against practical problems. This is a time-consuming and, at times, frustrating experience for manyresearchers, since their experience lies in a different field to thatin which the problem occurs. Even so, this process is essential

for the growth of technology and provides valuable feedback tothe research discipline.

For example, within the field of optimisation and combinatorialreasoning there exist many elegant and highly applicable resultsfor a wide selection of specific problems. In a few cases, anindustrial problem will exactly match a certain problem ideali-sation and the mathematical model used to study the propertiesof a particular class of algorithms. This is, however, only rarelythe case. Real practical problems in, for example, industrialproduction, are invariably complex combinations of severalsmaller problems and may have numerous related idiosyncraticconditions. The ability to understand and describe a practicalproblem in terms of the types of models and methods used inthis field is usually referred to as ‘modelling expertise’.Interestingly, this expertise is rarely described in the scientificliterature. The ability to use outside a scientific field the algo-rithmic methods developed within that field exists in parallelwith the science, and is indeed rather more of an artform.

A systematic method for solving a class of practical problems iscalled a methodology. However, most methodologies used incomputing science are based around a fairly limited set of algo-rithmic methods and quickly become useless whenever theproblem changes too much or is combined with other problemsinto a more complex situation. This raises issues of methodgenerality and scalability, which should be of prime concern tothe scientific field developing the method as well as of greatpractical importance to the society in which the scientist works.A systematic study of these issues will lead to two types ofresults: • better understanding of typical models for a selection of

important real-life industrial problems• better understanding of the properties of a selection of practi-

cally useful algorithmic methods.

Industrial applications of state-of-the-art techniques andmethods in computing science allow researchers to test, inpractice, the flexibility, scalability and utility of their tech-niques and methods. In addition, it is an opportunity to push themature parts of the scientific field out into practical use for thebenefit of the industry and society at large, be it in manufac-turing, transportation, processing, telecommunications,biotechnology or the service industry. This is true for manyfields of computer science, but particularly for those directlyinvolved in the modelling and solving of typical problemsoccurring in industry, eg process simulation, monitoring andprediction, capacity analysis and allocation, fault detection and

by Per Kreuger

The special theme of industrial diagnosis, planning andsimulation is structured around a selection of industrialproblems rather than a particular type of technology ormethodology. We have chosen contributions describing avariety of techniques and applications that are of relevanceto these problems, with the intent of highlighting state-of-the-art applied research within the ERCIM institutes and theirassociates.

Industrial Diagnosis, Planning and Simulation — Introduction


diagnosis, production planning, flow optimisation, resourcescheduling and allocation, structure detection and matching. Webelieve there is a strong and fundamental need to systematicallystudy the practical utility of such methods by applying them to alarge number of real cases.

For the special theme we have aimed to highlight successful orpromising applications of advanced techniques from these fields,including some in full-scale industrial settings, which we hope willcontribute to our understanding of the strengths and weaknesses ofthe various techniques and methods. The articles have been groupedinto four sections: 1.Electronics and Networks 2.Process Monitoring and Optimisation 3.Transportation and Logistics 4.Process Design and Management.

The first section contains two articles on the use of magnetic fieldmodels in the electronics industry, one on the management and opti-misation of communications network resources, and a fourth onintrusion prevention.

The second – and largest – contains articles on monitoring, diag-nosing and optimising production processes. Of these, several areconcerned with the detection of deviations from normal processparameters, while others address issues in connection with identi-fying the cause of, and correcting, faulty behaviour. In a few cases,optimisation of process parameters and redundancy of monitoringsystems are also relevant.

The third section contains articles on problems in transportation andlogistics. This type of issue often contains instances of specific prob-lems that are comparatively well understood, but in practice arefrequently made very difficult to solve by combinations of sub-prob-lems such as resource allocation, routing and scheduling. One of thearticles describes an analytical method for capacity analysis andassessment using techniques from discrete event systems. Three areconcerned with management and routing of vehicles, which is one ofthe most cost-intensive operations in transportation. Others describeapproaches to solving packing, placement and storage problems.

The articles in the final section describe methods for process designand project management — enjoy!

Link: http://www.sics.se/iam

Please contact:Per Kreuger, SICS, E-mail: [email protected]

10 Industrial Diagnosis, Planning and Simulation -Introductionby Per Kreuger, SICS

12 Efficient Algorithms for Inverse Magnetic FieldProblemsby Domenico Lahaye, CWI

13 Deviation Detection of Industrial Processes by Anders Holst, Jan Ekman and Daniel Gillblad, SICS

14 Bidding Agents Acquire Cargo in Online Auctionsby Pieter Jan 't Hoen, CWI

15 Decision Support System for Process Analysis and Supervisionby Galia Weidl, University of Stuttgart, Germany

17 Mathematical Challenges in the Electronics Industryby Wil Schilders, Eindhoven University of Technology, The Netherlands

18 Contpack - Maximising the Volume Utilisation of Containersby Onno Garms, Ralf Heckmann and Stefan Rank, Fraunhofer SCAI

20 First Commercial Bayesian Software for IntelligentTroubleshooting and Diagnosticsby Claus Skaanning, Dezide, Denmark

22 Innovating Shoe Manufacturing using AdvancedSimulation Techniquesby Emanuele Carpanzano and Andrea Cataldo, ITIA-CNR

23 Billy Goat Detects Worms and Virusesby James Riordan and Diego Zamboni, IBM ZurichResearch Lab/SARIT

24 Decision Support for Personnel and ResourcesManagement in Multi-Project Environmentsby Antonia Bertolino, Eda Marchetti and Raffaela Mirandola,ISTI-CNR

26 Towards Useful Tools for Estimating RailwayInfrastructure Capacity by Jan Ekman, SICS

27 Efficient Utilisation of Resources in the RailwayIndustry by Levelling Departures and Arrivals of Trainsby Martin Aronsson, SICS

28 Real-Time Optimisation of Empty Rail Freight CarDistributionby Martin Joborn, Carmen Consulting, Sweden

29 KUORMA: A Collection of APS-Algorithms for ForestIndustry Wood Transportby Juha Savola, VTT

31 Parc Bandwidth on Demandby Helmut Simonis, Parc Technologies Ltd, UK

32 DesParO - A Design Parameter Optimisation Toolboxusing an Iterative Kriging Algorithmby Christof Bäuerle, Clemens-August Thole and Ulrich Trottenberg, Fraunhofer SCAI

33 On-line Testing of the Reactor Protection System in the Paks Nuclear Power Plantby István Varga, Tamás Bartha and Alexandros Soumelidis,SZTAKI

35 Process Systems: Theory and Applications from Different Aspectsby Katalin Hangos and Gábor Szederkényi, SZTAKI

37 Multi-Agent System Technology in a Port ContainerTerminal Automationby Vicent J. Botti, Universidad Politécnica de Valencia/SpaRCIM

39 Contribution to Fault Detection and Diagnosis: Mixture-based Modellingby Tatiana V.Guy, Dani Juriãiã, Miroslav Kárny and AndrejRakar, Institute Information Theory and Automation,Academy of Sciences of the Czech Republic/CRCIM

40 Data Mining Applied in Anaerobic Wastewater Treatmentby Simon Lambert, CCLRC

41 Integrated Multivariate Statistical Process Control and Condition Monitoring for Fed-batch FermentationProcessesby Hongwei Zhang, Northeast Wales Institute of Higher Education, UK

43 Saving Energyby Claude Lemaréchal, INRIA

ARTICLES IN THIS SECTION



In recent decades, the use of mathemat-ical modelling and numerical techniquesin the industrial design of electricalmachines and transformers has gainedpopularity. Mathematical models forelectromagnetic energy transducers arebuilt starting from the Maxwell equa-tions - a set of coupled partial

differential equations describing electricand magnetic fields. These equations aretreated numerically by discretising them,for example, by the finite elementmethod.

In industrial design practice, so-calledinverse magnetic field problems arisefrequently. These are optimisation prob-lems in which one seeks a source distri-bution that produces an a priori specifiedmagnetic field. These hard, nonlinearoptimisation problems are mathemati-cally intriguing due to their ill-posednessand non-unique solutions. Furthermore,finding solutions efficiently is critical,and is a key ingredient in the design ofhigh-gain electromagnetic energy trans-ducers that are less susceptible to unde-sirable parasitic effects such as heat,noise and mechanical vibrations.

These inverse magnetic problemsinclude design variables such as descrip-tions of the geometry of a device (eg thewidth of the air gap in an electricalmachine), and external loading condi-tions in terms of applied currents andvoltages (eg the current through the coilsin an MRI scanning instrument). Theevaluation of the function to beminimised (the cost function) for givenvalues of the design variables requires afinite element simulation. These simula-tions form a computational bottleneck,

especially in performing full three-dimensional or transient modelling.Recently, the MAS2 research group atCWI started a research project with theaim of improving existing solution tech-niques for inverse magnetic field prob-lems. The goal of this research project isto construct and apply new optimisationalgorithms that reduce the requirednumber of cost function evaluations.These algorithms are based on a recentlyintroduced idea known as ‘spacemapping’.

Space-mapping algorithms exploit amultiscale representation of the search

space in order to accelerate the conver-gence of the iterative optimisationprocess. While fine-scale models corre-spond to the previously mentioned finiteelement models, coarse-scale models aresimple analytic models for the deviceunder consideration. Examples includemagnetic equivalent circuits (equivalentto electrical circuits) or analyticalexpressions for the electromagneticforce and torque (such as the Biot-Savartlaw). The coarse-scale representationallows well-established rules of thumb tobe incorporated in the optimisationprocess. The mapping from fine tocoarse space is performed through a

by Domenico Lahaye

The design of electromagnetic devices can be stated in mathematical terms asinverse magnetic field problems. Solving these hard optimisation problemsefficiently is of prime industrial relevance. At CWI, research has started on a newclass of algorithms for inverse problems based on the space-mapping algorithm.This algorithm exploits a multiscale representation of the space of design variables,in order to speed up the convergence of the optimisation process. Applicationsinclude the design of low-frequency electromagnetic energy transducers.

Efficient Algorithms for Inverse Magnetic Field Problems

For current-carrying coil and a permanent magnet, the first figure (left) shows the

equipotential distribution computed by the finite element method.

The second figure shows the triangular mesh used for the computation.


ERCIM News No. 56, January 2004

13

parameter extraction process. The space-mapping technique was introduced in thecontext of microwave engineering.

In our project, we must solve quasi-stationary inverse magnetic field prob-lems. The aims of the project aretwofold. Firstly, we want to provide atheoretical framework for the space-mapping technique in order to gaininsight into its convergence properties.We will do so by establishing linksbetween the space-mapping techniqueand seemingly closely related defectcorrection techniques. For the latter, asound mathematical theory exists.Secondly, we want to apply the space-mapping technique for the solution ofquasi-stationary inverse magnetic fieldproblems. A typical application that wehave in mind is the design of a linear

actuator to be used in a high-accuracypositioning system. This research effortis a collaboration between the MAS2research group and the Electromechanicsand Power Electronics (EPE) researchgroup at the Eindhoven University ofTechnology, who will provide theirexpertise in modelling and design ofelectromagnetic devices and in the vali-dation of design through experiments.The Dutch Ministry of Economic Affairsis funding the project through an inten-sive research program in electromag-netic power conversion technology(IOP-EMVT).

We envisage two possible extensions ofthe current project. In the first of these,we plan to apply the space technique inthe design of contactless energy transfersystems. In these systems, electrical

energy is transmitted by magnetic induc-tion, and electrical connections andphysical contact are avoided. Recentapplications of CETS have emerged, forexample, in the transfer of energy tomobile platforms, in the intervention-free charging of household appliancesand in medical scanning systems. In thesecond project extension, we plan tosolve inverse problems in whichmagnetic and thermal aspcets arecoupled. Solving these problems is animportant step in the design of electro-magnetic energy transformers thatminimise heat loss.

Link:http://www.cwi.nl/mas2

Please contact: Domenico Lahaye, CWITel: +31 20 592 4122E-mail: [email protected]

The general approach for deviationdetection is to construct a statisticalmodel of the relevant features of thedomain being monitored. The parame-ters of the statistical model are estimatedusing data consisting of (only or predom-inantly) normal cases. When used fordetecting deviations, the model calcu-lates the likelihood of a new sample case,ie, the probability that the model wouldgenerate that sample. A low likelihoodsignals that the sample is not likely to bedrawn from the model of the normaldata, and thus may be abnormal.

The area where this approach hasreceived the most attention is in intrusiondetection systems (IDS). Here, statisticalmethods are used to characterise trafficor user behaviour in various ways, basedon variables such as traffic volume,distribution over time, types of accessed

documents, multiple accesses todifferent servers etc. At SICS we haveused a similar approach to create asystem to detect fraud in ADSL-basedpay-per-view systems in a project withTeliaSonera. A number of features fromthe domain, such as the number ofordered and billed films and the amount

of outgoing and incoming networktraffic, was used to train a mixture modelto recognise normal (non-fraudulent)users. When tested on another set ofnormal users, plus a number of fraudu-lent cases of three different types (billingfraud, subscription fraud, and illegalredistribution fraud), the model success-

One of the major problems in complex processes is that it is impossible to foreseeevery possible type of fault or abnormal situation. This makes it hard to constructautomatic fault detection systems, since it is impossible to know what to lookfor. An interesting approach to this problem is deviation detection, which turnsthe problem around by using learning systems to model ‘normal’ behaviour andthen measuring the abnormality of a new situation against this.

Deviation Detection of Industrial Processesby Anders Holst, Jan Ekman and Daniel Gillblad

A projection of the

fraud detection

data. The redness

of the points

indicates how

unlikely they are

(the negative log

likelihood).



CWI's Evolutionary Systems andApplied Algorithms group headed byProf. J.A. La Poutré is a participant in theDistributed Engine for AdvancedLogistics (DEAL) project. The goal ofthis project is to optimise the use of roadtransport capacity in a logistics setting,using agent-based technology. Amongother things, this involves providing anopen market where business participantscan conduct online electronic acquisi-tions and place orders.

As real-world scenarios in logistics andplanning grow in size and complexity,classical centralised approaches becomeunwieldy and brittle. In a logisticssetting, a centralised planner can be over-whelmed by the stochastic nature of real-world events. For example, new businessopportunities arise that cannot be effi-

ciently exploited. Furthermore, schedulescan go awry and their repair can requireextensive rescheduling . A Multi-AgentSystem (MAS) offers a robust alternativeto a centralised planning approach. Anindividual agent within the system is anautonomous and adaptive piece of soft-ware. It is therefore able to learn torespond effectively to changingscenarios, and can efficiently solve tasksrequiring multiple agents by communi-cating with and coordinating otheragents. For example, MASs have beensuccessfully applied to difficult supply-chain management problems wherecentralised control has failed. In thetransportation setting, agents, which actfor individual trucks or coalitions oftransporters, are able to continuouslyoptimise and update planning as inci-dents occur and new opportunities for

profit arise. An agent can quickly assessthe value of new cargo or inform relevantagents if a planned delivery is delayedbecause of traffic or an unexpected over-head in the unloading of a truck.

For application to transportation settings,research at CWI focuses on possiblemarket mechanisms and bidding andnegotiation strategies of the agents. CWIis developing robust, distributed marketmechanisms as part of MASs for usagein (simulations of) the transportationsector. Online decentralised auctionsserve as the model, where agents - repre-senting trucks - bid for cargo. Currentcentralised planning is only able toutilise 40 to 60% of the total truckhaulage capacity. Preliminary experi-ments show that continuous acquisitionof new orders by agents bidding in

by Pieter Jan 't Hoen

A growing field of applications for Multi-Agent Systems is the world of logisticsand planning. CWI's research in the DEAL project focuses on optimising the useof road transport capacity through acquisition of cargo by agents bidding in onlineauctions. This promises a substantial gain in profit.

Bidding Agents Acquire Cargo in Online Auctions

fully picked almost all of the billing andredistribution frauds, and the majority ofthe subscription frauds.

Another application in which SICS isapplying the same techniques is thedetection of abnormal behaviour in shipmovements based on radar images. Thegoal is to be able to generate a warning ifa ship is moving in a strange, dangerous,or suspicious way. The relevant featuresthat are modelled in this case includelocal properties, such as current positionand speed, as well as more qualitativefeatures of the route travelled so far, suchas the number of turns, stops, possibleencounters with other boats, and so on.

It is also becoming possible to embedsuch normality modelling into industrialsystems such as network routers, indus-trial robots, the components of papermills etc. This will make it possible forthe components themselves to generate a

warning when something is beginning toget out of order; for instance, in thecontext of condition-based maintenance,the components could indicate whenthey are close to wearing out and needreplacement or service. Making compo-nents self-monitoring in this way hasseveral advantages in terms of reliability,uptime, reduced waste of materials, andlower maintenance and production costs.

We are taking the first steps towards thisin a project funded by SSF (SwedishFoundation for Strategic Research) andentitled ‘An 'English butler' for theprocess industry’. One of the test cases isa Swedish hot steel rolling mill. Thereare many factors affecting the finalquality of the steel, such as the form ofthe heating curve. By monitoring thesefactors, it is possible to detect when theprocess moves outside the normal regionin state space, and hence when a risk ofdamage to the steel exists.

Critical for the success of deviationdetection is the use of sufficientlypowerful statistical learning models tocatch relevant behaviour in often highlycomplex domains. At SICS we havedeveloped a general statistical modellingtool, ‘Hierarchical Graph Mixtures’,based on a combination of mixturemodels, probabilistic graph models, andBayesian statistics. This has given us avery powerful and flexible frameworkfor modelling, which is applicable to awide range of real-world tasks.

Links:http://www.sics.se/iam/projects/butler.htmlhttp://www.sics.se/iam/projects/adaf.htmlhttp://www.sics.se/iam/projects/dallas.html

Please contact:Anders Holst, Jan Ekman, Daniel Gillblad,SICSE-mail: [email protected], [email protected], [email protected]



online auctions (spot markets) can raisethis to 80%.

A bidding strategy under further devel-opment at CWI is the use of decommit-ment as a tool for handling unknownfuture orders available for auction. Usinga decommitment strategy, an agent isable to postpone the transportation of anacquired load to a more opportune time.This allows the agent/truck to bid inauctions it would not otherwise considerdue to prior commitments. Experimentsshow that substantial increases in perfor-mance through the use of a decommit-ment strategy can be expected due toadditional flexibility in the planningcapabilities of the MAS. This increase inperformance for the (abstract) model ofthe transportation sector can be seen as alower bound for expected additionalperformance in practice. This claim issubstantiated through experiments (seethe links below for publications), whichhave shown that the relative impact of adecommitment strategy increases withthe complexity of the world.

A full-scale MAS also facilitates commu-nication and therefore advanced negotia-tion between agents. Logistics planningis faced with complex constraints and the

varying goals and capabilities of thetransporters. CWI has demonstratedthrough evolutionary simulation thatagents can learn to negotiate successfullyover complex multi-issue goals (see thelinks below for publications). Agents canalso learn to exchange tasks/orders so asto improve performance for both parties.Negotiations between agents allow forcontinuous optimisation, while adaptivitypromotes efficient negotiation strategieswith specific opponents. Furthermore,agents can negotiate to form coalitionsfor the execution of complex tasks. Forexample, an order too large to be trans-ported by one truck can be split up over a

temporarily cooperating groupof agents.

The DEAL project intends todeliver a working prototype of aplatform supporting onlineacquisition of cargo using MASsin three years’ time. Researchwill continue at CWI into theprecise conditions and parame-ters required in the underlyingmodel of decentralised auctionsfor successful application of anMAS in such a logistics setting.The research focuses on themarket mechanisms required, ie

which rules guide the bidding behaviourof the agents and how can agents bestnegotiate over a transport? This iscomparable to the study of rules as foundin a market economy. The application ofthe right type of intelligent software willplay a major role.

Links:http://www.almende.com/dealhttp://www.cwi.nl/sen4http://www.cwi.nl/~hoen

Please contact:Pieter Jan 't Hoen, CWITel: +31 20 592 4276E-mail: [email protected]

Agents increase profits by continuous acquiring

new cargo for transportation in online auctions.

As system complexity increases, condi-tion monitoring and fault diagnosisbecome demanding tasks for processoperators, who face an overflow of dataand may have to undertake extensivesearches for unexpected faults. This canlead to work overload and high stresslevels. Simultaneously, technical,economical and time constraints onproduction efficiency and quality requirethat condition monitoring and adaptationof production and service processes beundertaken on a continuous basis.

In this context the ABB group, alongwith its customers from process indus-tries and the Danish company HuginExpert, have collaborated on a project(which commenced in 2000) targetingthe development of a methodology forroot cause analysis (RCA) and decisionsupport on process operation. Theauthor, who was involved in technologyevaluation and methodology develop-ment while associated with ABB, is nowassociated with IFF – StuttgartUniversity and is working on an exten-

sion of the methodology to automotiveand manufacturing industries andservices. It is expected that this extensionwill be implemented within a Europeanproject known as ‘The Butler Concept’(see page 16 in this issue).

Motivation and the Techniques EmployedIn finding the origin of a process distur-bance or fault, there exists a need for aquick and flexible guidance tool for deci-sion support at higher automation levels,

by Galia Weidl

The increasing complexity of industrial processes and the struggle for costreduction and higher profitability means automated systems for processesdiagnosis in plant operation and maintenance are required. The system proposedhere is designed to address this issue.

Decision Support System for Process Analysis and Supervision



including analysis of process conditionsand advice on cost-efficient actions.

The technology of probabilistic graph-ical models such as Bayesian networksand influence diagrams have turned outto be the best of a number of alternatives,in the case where high diagnostics capa-bilities, explanation of conclusions fortransparency in reasoning, and trust-worthy decision support are expected bythe users (process engineers, operatorsand maintenance crew).

Due to the existence of a number of first-level diagnostic tools, the aim has beento provide decision support in processoperation. The framework of Bayesiannetworks has been found to be an effi-cient and flexible tool in overall-levelprocess operation analysis, since not allconditions are measurable or computablein real time, and the combinatorialreasoning procedure is subject to uncer-tainties.

The development of the methodologyincorporated the following systemrequirements and modelling issues:• root cause analysis of industrial

processes with adaptation to processoperation/grade changes, aging andwear

• reusable system design for variousprocess applications

• reusable modelling of repetitive struc-tures (eg sensors, control loops) andassets (pumps, valves)

• risk assessment of disturbances byanalysis of signals’ level-trend, while

adapting to changes in process opera-tion mode

• ease of communication and explana-tions of conclusions at different processlevels, eg process overview,(sub)sections, units and instrumenta-tion.

These modelling and system require-ments have been met in the method-ology. This is supported by the integra-tion of the methodology and the Hugin-tool into the ABB Industrial IT platform.This RCA-integration allows efficientdata exchange with all available IT-applications, eg distributed controlsystems, diagnostics of sensors andcontrol loops, and physical-modelcomputations. The infrastructure forapplying this methodology in differentdomains is therefore ready for imme-diate use. The method and the analyserfor producing information have been the

subject of five separate patent applica-tions.

The first prototype of the root causeanalysis system was tested by ABB withreal process data during 2002. The moni-toring and root cause analysis of thedigester operating conditions in a pulpplant have been chosen as a real-worldapplication (see Figure 1). The structureof one of the developed BayesianNetworks is shown in Figure 2.

The real application development hasbeen closely related with its integrationinto the Industrial IT platform. Theextension includes the use of object-oriented Bayesian networks (OOBN).OOBN facilitate the modelling of largeand complex domains and allowreusability.

Figure 2: An example of a Bayesian Network for root cause

analysis of process operation.

Figure 1: Digester Fiber-line. Case-study: Monitoring of the

digester operating conditions.

The Butler Concept

This article by Galia Weidl as well as the articles 'Deviation Detection of IndustrialProcesses' on page 14 and 'First Commercial Bayesian Software for IntelligentTroubleshooting and Diagnostics' on page 20 are related to the 'Embedded Butler Concept'.The idea is to create an embedded and self-learning system for detecting, diagnosing, andcorrecting faults and deviations in complex systems.

A consortium containing among others SICS (coordinator), ABB, Aalborg University,University of Stuttgart, Tampere University of Technology, Mälardalen University, WarsawUniversity of Technology, Hugin Expert A/S, and Dezide ApS, has been formed and anapplication for a STREP project has been submitted under the Embedded Systems call inthe 6th Framework Programme of the European Commission. The goal of the project is todevelop advanced functionality within each of the sub-components of fault detection, faultdiagnosis, and corrective actions, and to combine that functionality into a complete system.

http://www.sics.se/iam/

Please contact: Björn Levin, SICS, [email protected]



Benefits, Results and ConsequencesThe resulting RCA-system providesprocess operators with information oncondition overview and advice on themost efficient sequence of correctiveactions. In addition, the advantages forprocess industries include the seamlessintegration of all relevant informationsources on the Industrial IT platform,resulting in automated and early assess-ment of abnormal conditions with flex-ible diagnosis and advice. Futureenhancements will involve integration ofthe functionalities providing advice on asuitable time for maintenance activitiesunder technical constraints and orderdeadlines in process scheduling, as wellas simulation of the impact of intended

corrective actions on process efficiency.This allows pro-active (instead of reac-tive) troubleshooting to be undertaken,which increases the process perfor-mance, availability and output, avoidspotential process breakdowns and cutsboth operation downtime and mainte-nance cost.

Expected Function – RCA as a Powerful Complement to System Control We believe that this methodology willprovide a powerful complement tosystem control in process industries. Ourexpectations and the proof of thesystem’s potential are based onsuccessful tests at ABB with real plant

data. Future challenges are primarilyrelated to the application of hybridlearning systems for abnormality detec-tion, diagnosis and advice in large-scaleindustrial settings.

Links: http://www.abb.comhttp://www.hugin.comhttp://www.iff.uni-stutgart.de

Press-release Web Site:http://www.hugin.com/cases/Industry/ABB/ABB18122003.pdf

Please contact:Galia Weidl, University SuttgartE-mail: [email protected], [email protected]

The electronics industry has always beena very fertile working environment formathematicians. In the 80s, the simula-tion of semiconductor devices providedan extremely challenging problem fromthe numerical point of view, and asimilar observation can be made for thearea of electronic circuit simulation inthe 80s and 90s. Owing to the advent ofnew numerical methods, these simula-tions are now carried out routinely.

A relatively new development within theelectronics industry is the use of soft-ware for the simulation of interconnectsstructures connecting the differentcomponents in a large integrated circuit.Figure 1 shows such a structure. Inpresent-day technologies, this structuremay contain up to seven layers.

With shrinking dimensions andincreasing frequencies, nearby metaltracks and layers can have an influenceon each other. So far, these effects could

be neglected when analysing thebehaviour of an integrated circuit.However, there is clear evidence that thiswill not hold for future generations.Electromagnetic effects (see Figure 2)may influence the functioning of acircuit to the extent that signals aredelayed intolerably. Clearly, this must beavoided, explaining the large interest inthe topic of signal integrity.

In close cooperation with PhilipsResearch in Eindhoven, the ScientificComputing Group at the TU Eindhovenstarted an investigation into numericaltechniques for the simulation of theproblem. A European project(CODESTAR) was set up with two chal-lenging tasks. Firstly, to develop effi-cient numerical techniques for solvingthe three-dimensional electromagneticproblem. Secondly, based on thesecalculations, to construct models with asmall number of parameters thatadequately describe the electromagnetic

behaviour. This step is commonlyreferred to as 'reduced order modelling'.It has become clear that methods avail-able for these tasks need a thorough revi-sion in order to be able to cope withpresent demands of the electronicsindustry, and so the ScientificComputing Group concentrates on bothaspects.

The time-dependent three-dimensionalelectromagnetic problem is commonlysolved using the finite difference timedomain method (FDTD), developed by

by Wil Schilders

New generations of electronic devices and integrated circuits are characterisedby shrinking dimensions and increasing frequencies. Electromagnetic effectsthat were previously neglected must now be taken into account in order to reliablypredict behaviour. The Scientific Computing Group of the TU Eindhoven isdeveloping new numerical techniques for coping with this challenging problem.

Mathematical Challenges in the Electronics Industry

Figure 1: Interconnect structure.



Yee in 1966. The method is simple, andprogramming it is rather straightforward.However, there are some clear disadvan-tages. First of all, the time stepping isexplicit, so the smallest spatial dimensiondictates the time step. In addition, thespatial grid is often unnecessarily finedue to the finite difference character. TheScientific Computing Group has devel-oped a general framework of time step-ping methods, which relieve the strictstability conditions. This is extremelyimportant for future simulations of signalintegrity. Also, we are now working onother improvements, such as the use ofparallel computing, combining two solu-tions on different staggered grids toimprove the accuracy, and the introduc-tion of local mesh refinement using localdefect correction (LDC).

The second main task is even moreimportant than the first. In fact, reducedorder modelling is a discipline, whichreceives growing attention in manyapplication areas as well as in mathe-matics. To stress the importance, theScientific Computing Group has organ-ised several international workshopswithin the framework of the Europeannetwork MACSI-net. Reduced ordermodelling aims at capturing complexbehaviour into a compact model with asmall number of parameters. Thefollowing sequence of figures clearlyillustrates the intention: although only afew straight lines and planes describe thebunny on the right, it is clearly recog-nized as being a bunny.

In the past year, we have developed arobust reduced order modelling tech-nique based on an expansion in Laguerrefunctions, combined with Krylovsubspace techniques. The method leadsto low order models that are stable bothin the frequency domain and the timedomain, and has been implemented andtested in a Philips Research software toolfor the simulation of electromagneticcompatibility. In the next year, it will beused within the framework of theCODESTAR project to generatecompact representations of the electro-magnetic behaviour of interconnects.

Links:http://www.win.tue.nl/scghttp://www.imec.be/codestarhttp://www.macsinet.org

Please contact: Wil Schilders, Eindhoven University of Technology, The NetherlandsTel: +31 40 247 8389E-mail: [email protected]

Figure 2: Electromagnetic effects.

Figure 3:

Principle of reduced order modelling.

Today’s industrial production is oftenspread over many sites and locationsworldwide. Manufacturing of variousparts and intermediate or final assemblycan all occur in different places, and theintegration of these production sitesrequires efficient transportation of theturnout between them. This transporta-tion of parts or products is often done in

containers or boxes. In order to minimisethe cost of transport and logistics, it isbeneficial to many companies to packtheir products as tightly as possible.

The optimisation group withinFraunhofer SCAI is currently developingautomatic packing algorithms and soft-ware that can be used to pack complex

three-dimensional parts (each of whichmay have many hundreds of thousandsof facets) into containers as efficiently aspossible, thereby maximising theirvolume utilisation.

Apart from volume utilisation, theresults calculated by our packing algo-rithms are subject to other constraints

by Onno Garms, Ralf Heckmann and Stefan Rank

The Fraunhofer Institute for Algorithms and Scientific Computing (SCAI) hasdeveloped three-dimensional packing software for optimising the volume utilisationof containers. The software automatically chooses from a given range of containersand quickly generates highly efficient three-dimensional packings of complexobjects.

Contpack — Maximising the Volume Utilisation of Containers



and demands from industry. Since theloading and unloading of containers isoften done by humans, it is important togenerate layouts which are easily repro-ducible by humans. The number ofdifferent rotational positions of thecomponents is therefore limited to asmall number. These rotations are notprovided in advance, but must be foundand optimised by the software.Additionally, the containers are filled intiers, many of which are identical.

In current industrial practice, the genera-tion of packing layouts for containers isdone by human beings, who require asufficient number of physically availablecomponents and containers, andcomplete the task by trial and error. Sucha procedure is impossible if the compo-nents have not yet been produced.Increasingly however, companies wouldlike to generate container packings fortheir parts while still in the design stage,in case there exists a way of modifyingthe outer hull of a part slightly such that itremains equivalent from a technologicalpoint of view but can be packed moreefficiently.

Finding optimal packing methods withphysical objects is a difficult and time-consuming task, because both compo-nent geometry and a large variety ofpossible container dimensions must beconsidered. Another common methoduses CAD systems to generate the layout,but these lack algorithmic support for thetask. As things stand, the packing qualityof a part into a container improves overtime as the insights and suggestions ofemployees are implemented.

Our Contpack software chooses a propercontainer size from a given list of sizesand generates a high-density regularlayout for this container. This reduces theoverall number of containers needed inthe supply chain, including stocking,transportation and handling. In the stageof production planning, the automaticpacking generation helps estimate therequired number of containers. It canalso help in recognising the need todevelop containers with new dimensions.

In contrast to existing algorithms, whichare often limited to the packing ofcuboids using bin-packing strategies, ouralgorithms can handle components ofarbitrary shapes. The software can beused at an early stage in componentdevelopment, because the algorithms can

handle CAD data of different quality.The integration of the Contpack programinto existing CAD systems allows boththe design of new components and theplanning of their transportation to beperformed in a single step within auniform environment.

The user can select a triangulated VRMLrepresentation of the component to bepacked and a list of container sizes to betested, and can define the stable posi-tions. The stable positions restrict thedegrees of freedom for the componentsto rotations around the vertical axis andto arbitrary translations. This accountsfor the stability of the generated layout.

There are a number of packing options,including lattice packing. The lattice is atwo-dimensional grid made up ofrepeating units such as rectangles, trape-zoids or hexagons. The lattice maintainsthe layout during transportation and thusprotects the components. The use of alattice results in a completely regularpacking of the container. For achievingdenser layouts there are options withouta given lattice structure. The resultinglayout can be viewed in three dimen-sions, allowing the user to browsethrough the tiers of the filled container.

Packing tasks belong to the class of NP-hard problems, meaning that no algo-

Figure 1:

Workers loading

containers with parts.

Cou

rtes

y by

the

onlin

e m

agaz

ine

'logi

stic

s pi

lot'.

Figure 2: Computer-generated packing of loudspeakers into a container.



Dezide’s product, Dezide Advisor,implements state-of-the-art researchwithin the area of intelligent trou-bleshooting and diagnostics, and is wellsuited for support and maintenance ofcomplex industrial machinery.

Dezide is a small Danish companystaffed by former employees from anR&D department of Hewlett-Packardthat was ‘spun off’ in 2001. There arecurrently eight employees, including onedoctorate and three Masters, all special-ists in computer science with a focus onBayesian networks.

Bayesian networks allow the structuringof complex information in a graphicalnetwork, where arrows depict the direc-tion of causality, that is, from cause toeffect. The exact computation methods

of Bayesian networks then allow the userto fix any variable, and continually re-calculate the probabilities of other vari-ables given the new observation. Incomparison with fuzzy systems andother AI technologies, Bayesiannetworks are mathematically sound,allowing them to be used for far morecomplex and uncertain problems.

Dezide’s software is currently being usedfor troubleshooting in many differentareas, including Internet connection prob-lems and email setup problems (TDC,Denmark’s largest telco), network routerproblems (Telenor), and common ITproblems (SAS). In addition to this,Dezide is currently involved in activitieswhere the software is being used for themaintenance of complex bridge structuresand diagnostics of windmill systems.

Dezide Advisor is based on Bayesiannetworks and implements seven patentsin this field. In short, a field expert storeshis knowledge of problems, resolutionsand symptoms using a special authoringtool. This knowledge can then benefitend-users, technicians, and operators intrying to more quickly and preciselyresolve problems. In addition, DezideAdvisor can help with detecting devia-tions from normal operating behaviour,can predict future difficulties based oncurrent measurements, and can find theprobable cause of a problem and theneither sort it out itself or call for thenecessary assistance.

Dezide Advisor is particularly wellsuited for troubleshooting and diagnosisin complex technical systems. In thiscase, Dezide Advisor will guide the user

by Claus Skaanning

Dezide Advisor is the first system of its kind based on Bayesian networks – amodern approach for artificial intelligence, providing better solutions for complexand uncertain problem domains. The system is able to provide intelligent advicefor solving complex technical problems efficiently and precisely, and no specialexpertise or skills are required to develop the underlying knowledge base.

First Commercial Bayesian Software for Intelligent Troubleshooting and Diagnostics

rithm for the optimal, time-efficientsolution of the problem is known. Thetechniques employed include exact andheuristic optimisation methodsdepending on the complexity of the taskor sub-task being performed by the algo-rithm. For packings into given latticestructures we have implemented veryfast optimal methods. For packingswithout lattice structures, the problem isNP-hard. We also apply strategies forfinding regular packings, but we addheuristic methods for finding optimalcombinations of regular packings indifferent areas of the box.

Future work will include relaxing thestrict separation of the tiers inside thecontainer by robust intermediate tiers.Also, more than one stable position pertier will be allowed. The user will be able

to define taboo areas between thecomponents, and we will fully integratethe software into the user’s CAD envi-ronment, eg into CATIA.

Fraunhofer SCAI has a strong back-ground in the solution of cutting andpacking problems. Experts in the depart-ment for optimisation have beenworking on packing problems for overten years. The application of the soft-ware to three-dimensional containerpacking started in 2002.

Fraunhofer SCAI is collaborating on thisproject together with and by the commis-sion of Audi and the BMW Group. Ourpartner Solve-IT MVI in Munich tookover the project management and devel-oped the interface to the container andVRML data, the dialogues, the visualisa-

tion in 3D and the generation of thepacking instruction reports. While theprogram is tailor-made for the needs ofthe automotive industry, it can be used inmany areas of the manufacturingindustry. Fraunhofer SCAI intends torelease a standard product for thecontainer-packing problem in the nearfuture. We also plan to extend the func-tionality of the software and to make itusable in other fields of application.

Link:http://www.scai.fraunhofer.de/opt

Please contact:Stefan Rank, Fraunhofer SCAITel: +49 2241 14 1503E-mail: [email protected]



through an optimal sequence of stepsleading to a solution or diagnosis in asfew steps as possible. Dezide Advisorcan run interactive question-answersessions, but can also run fully automat-ically, embedded in technical systems.

The next step in an interactive questionsequence is calculated based on allprevious answers, probabilities that stepswill help, and their costs (time, risk,money etc).

Dezide Advisor is an improvement overexisting solutions in several ways. It isable to reach a solution or diagnosis in asfew steps as possible and with the lowestpossible cost, and when it concludes, it

presents a list of the potential root causessorted by probability. The system is veryflexible, as the user is not required tofollow a fixed path but can skip steps thathe cannot answer, and jump directly tosteps that he wants to answer straightaway. In addition to this, the Advisor isself-adapting and adjusts itself to the situ-ation over time.

Dezide Advisor is based on severalbreakthrough inventions. These include: • an authoring tool that allows anyone to

construct Bayesian networks efficientlyfor diagnostics and troubleshooting

• algorithms for calculating the optimalsequence of actions and questions

towards solving a problem or finding adiagnosis

• validation tools for testing and qualityassurance.

Dezide Advisor includes a diagnosticengine that can be embedded in diagnos-tics applications, desktop tools forconstructing and maintaining the know-ledge base, and a scalable, secure androbust server system for Web-basedaccess by thousands of simultaneoususers.

It is possible to learn to use the know-ledge engineering tool in only two days,and no knowledge of Bayesian networksis required. The tool includes a graphicalvisualisation functionality to get anoverview of the knowledge base (seeFigure 2).

Dezide is now looking at improving itssoftware to handle more complex prob-lems in industrial plants. Together withABB, SICS, Aalborg University, theUniversity of Stuttgart, Renault andseveral other research and industrialpartners, Dezide is a member of aconsortium proposing an EU researchproject known as ‘The Butler Concept’(see information box on page 16).

In the Butler project, Dezide will beinvolved in conducting general researchfor improved knowledge acquisition andtroubleshooting algorithms withinhighly complex industrial plants. Inparticular, Dezide will work on methodsfor tuning the knowledge base based onexpert feedback, generating easilyunderstandable explanations of thesystem’s suggestions and conclusions,and importing of diagnostic data fromother formats such as log files, decisiontress, case bases and so on.

Link: http://www.dezide.com

Please contact: Claus Skaanning, Dezide, DenmarkTel: +45 9655 7370E-mail: [email protected]

Figure 1: Troubleshooting printer problems.

Figure 2:

A graphical

view of an

example

knowledge

base.



Nowadays, many different approaches,methodologies and computer aideddesign tools can be employed in the anal-ysis and optimization of complex indus-trial systems. As a consequence, theselection of the most appropriate tech-niques and tools for a particular systemis critical. Before making any final deci-sion, it is necessary to investigate thefunctionalities offered, compliancy withexisting standards, and capacity forinteroperability with non-proprietaryinstruments. We are currently workingon a methodological approach to thestructured design and simulation-basedanalysis and optimisation of manufac-turing plants. The methods and toolsused during the different phases of thedesign are illustrated with reference to anapplication in the shoe manufacturingsector. In particular, we describe aninnovative plant for the production ofcustomized shoes. This plant is locatedat the ITIA-CNR laboratory in Vigevano(Italy) and constitutes the pilot plant of alarge European research project calledEUROShoe, which began in March 2001and is scheduled to finish in June 2004.

EUROShoe is a complex and ambitiousproject with thirty-five academic andindustrial partners from ten differentEuropean countries. It aims at a dramaticrenovation of the concept of shoes asproducts and of their production, basedon a transformation from mass-producedto mass-customised goods. This productevolution goes in parallel with a transfor-mation of footwear companies intodistributed and flexible enterprisescapable of handling the complexity thatsuch a radical change in the nature of theproduct implies and of mastering theassociated new technological challenges.This implies a complete revision not

only of the entire manufacturing processbut also of the tools used to analyse andoptimize the resulting innovative indus-trial system.

For the sake of brevity, the methods andtools used for the design, analysis and

optimization of the plant automationsystem are outlined with reference to thecontrol and supervision of the transportline. In the shoe manufacturing plant thatwe consider (Figure 1), an innovativetransport line is used to move the semi-finished shoes from one machiningstation to another according to a prede-fined operating schedule. In particular,the innovative molecular structure of thetransport line (Figure 2) stronglyenhances the modularity, scalability,integrability and reconfigurability of theproduction system, thus increasing theoverall flexibility of the plant.

At the beginning of the design life cycle,the control and supervision system of thetransport line was specified by means ofUML (Unified Modelling Language)diagrams; in particular use-case, class,sequence and state diagrams were usedto specify the systems modules, theirrelationships and their dynamicbehaviour. The system architecture andfunctions were designed using theFunction Block formalism defined bythe IEC 61499 standard, which inte-grates object-oriented concepts anddiscrete event models to suitably supportcontrol software design. The functionalmodel obtained was analysed and opti-mized by means of closed-loop discreteevent simulations performed in theSimulink/Stateflow environment, whereboth the controller and the controlledprocess were modeled and simulatedthrough state charts, as was thebehaviour of the overall system. In thisway, the correctness and performance ofthe solution proposed could be evalu-ated. In particular, a bottom up method-ology was defined and exploited to studythe system according to a modularapproach. This enabled us to simplify theoverall analysis process, to verify thecorrectness of the automation functional-ities easily and to optimize the systemperformance from the very first phasesof the design life cycle. Once the func-tional modules had been verified, thecontrol and supervision algorithms weredeveloped using the SFC (SequentialFunctional Chart) formalism, which is anadvanced discrete event modellinggraphical language, directly derivedfrom Petri nets, and included in the IEC61131 part 3 standard, which definesprogramming languages for industrialPLCs (Programmable LogicControllers). The SFC algorithms were

by Emanuele Carpanzano and Andrea Cataldo

In order to handle the increasing complexity of distributed industrial manufacturingsystems, there is a strong demand for methods and tools that support the designerin the analysis and optimisation of flexible infrastructures for the automation ofproduction processes. We are currently investigating how emerging standardsand advanced simulation techniques can be exploited successfully in theproduction of custom made shoes.

Innovating Shoe Manufacturing using Advanced Simulation Techniques

Figure 1: The Innovative Shoe

Manufacturing Plant of the ITIA-CNR

laboratory of Vigevano (Italy).



analysed through closed loop discretesimulations in the ISaGRAF environ-ment in order to verify their correctnessand to optimize their performance.ISaGRAF is a CACSD (Computer AidedControl System Design) tool that allowsthe automation software to be structuredusing all five IEC 61131 languages andsupports simulation functionalities fortesting purposes, as well as automaticcode generation facilities for differentindustrial systems. In order to performthe closed loop simulations, simplifiedmodels of the plant devices have beenrepresented in ISaGRAF by means ofsuitable SFC modules and data struc-

tures. Moreover, to simplify the analysisof the simulation results, a simple 2Dgraphic animation was realized inISaGRAF (Figure 2). Simulations havebeen performed by considering typicaloperating conditions, ie, typical produc-tion orders, and the results obtainedshow that the system is deadlock freeand that the plant is well balanced, ie itsresources are all used effectively. Afterverification and optimization of theautomation functions and algorithms, thecorresponding code was generated andimplemented on the target industrialdevices. We found that the simulation-based analysis techniques reduced the

plant rump-up times and costs andimproved overall system performance.Future work will concern the exploita-tion of the standards, methods and toolsproposed for other industrial plants andtheir integration within an advancedCACSD tool for manufacturing systems.

Links: http://www.itia.cnr.it/http://www.euro-shoe.net/

Please contact:Emanuele Carpanzano, ITIA-CNRE-mail: [email protected]

Figure 2: Layout of the Shoe Plant (left) and ISaGRAF Simulation Graphical Interface (right).

The Billy Goat is specialized to addressthe problems posed by network serviceworms. As an intrusion detection sensor,its most important property is that it isfree from the high rate of false alarmsproduced by many other sensors. Itachieves this property through the use ofa novel architecture that combines anextensive view of the network, spoofedservice interaction with potentialattackers, and a clear focus on detectingautomated attacks.

One of the greatest threats to security hascome from automatic, self-propagatingattacks such as viruses and worms.These attacks scan at random until theyare able to place a program on the serverusing a maliciously crafted request. Theprogram uses the now-infected server asa base from which to attack otherservers. The direct result is rapid expo-nential growth in the number of attackersleading to load-induced network failure.

While the presence of these attacks is byno means new, the damage that they areable to inflict and the speed with whichthey are able to propagate has becomeparamount. Further increases in connec-tivity and service complexity onlythreaten to exacerbate their virulence.

ArchitectureThe Billy Goat functions by spoofing theexistence of machines and services atotherwise unused IP addresses. Becausethe addresses are otherwise unused, all

Researchers at the IBM Zurich Research Laboratory have designed andimplemented a new intrusion detection tool that not only facilitates early detectionof attacks but also sharply reduces the false alarm rate.

Billy Goat Detects Worms and Virusesby James Riordan and Diego Zamboni



traffic destined to them is a priori suspi-cious. The sensor spoofs services, ratherthan merely recording attempted connec-tions, to determine the intention behindthe traffic.

Billy Goat is built atop a security-hard-ened Linux machine that offers no realservices beyond very restricted login. Itis configured in conjunction with thenetwork on which it runs so that trafficdirected toward address subnets that arenot used is routed to the Billy Goat.

The Billy Goat itself offers a virtualiza-tion infrastructure that allows individualsensors to be written as if they wererunning on a single host. It also providesa logging infrastructure based on a rela-tional database facilitating correlationand analysis of the copious data producedby the large number of virtual sensors.

The spoofed services include HTTP,Microsoft RPC (remote procedure call),Microsoft SQL (database), and SMB(file sharing and printing).Vulnerabilities in these services arecommonly used as vectors for worm andvirus propagation.

One very important requirement of theBilly Goat is that it continue to functionin times of heavy worm activity. Inparticular, it must retain its utility evenwhen the performance of the network isdramatically diminished or evencompletely unreliable. This requirementimplies a distributed system of BillyGoats on the network. Each local BillyGoat serves to inform administrators oflocal infection so that it can be elimi-nated. In order to support the distributedarchitecture, the Billy Goat distributioncontains an automatic update mecha-

nism. This mechanism ensures that adeployed Billy Goat has all the latestsensors and signatures.

It is worth contrasting the Billy GoatBox with another security tool called aHoney Pot. A Honey Pot, as its namesuggests, offers "something desirable" tolure attacks to a particular machine.Unlike a Honey Pot, the Billy Goat doesnot advertise any services, and does notallow any of its services to be corrupted.The Billy Goat is designed to detectlarge-scale automated attacks rather thantrying to deceive sentient humanattackers.

Please contact:James Riordan, Diego ZamboniIBM Zurich Research Lab/SARITTel.: +41 1 724 89 81E-mail: [email protected]

Propean (Figure 1) relies on themetaphor that project teams correspondto the processing resources of classicalperformance models, and project activi-ties are the tasks to be performed withinestablished time intervals.

According to this metaphor, well knowntechniques from performance analysiscan be usefully adapted to managers’needs, for eg assessing the time forcompletion of specified tasks, handlingpersonnel multitasking over concurrentprojects, distributing the workloads indevelopment cycles, deciding aboutproduct release, and similar issues.

However, a key aspect for the effectivedeployment of Propean is to adopt aninterface that appears familiar tomanagers. Hence, we adopt as the inputmodelling notation the UML (UnifiedModelling Language) Profile forSchedulability, Performance and Time(RT-UML) recently approved as anOMG standard. In this way, managers donot need to become knowledgeable ofperformance notations, but can develop amodel of the flow of activities and oftasks distribution using the widespreadUML notations and tools. The transla-tion into a format that can be processedby performance analysis algorithms isthen performed automatically.

The SPE ApproachThe Propean technology is based on is theSoftware Performance Engineering (SPE)approach, which is a systematic, quantita-tive approach to software systemconstruction in order to meet performanceobjectives. Its basic concept is the separa-tion of the Software Model (SM) from itsenvironment or Machinery Model (MM).

The SM captures the essential aspects ofsoftware behavior; it can be representedby means of Execution Graphs, in whichnodes are software workload componentsand edges represent transfers of control.Each node is weighted by use of ademand vector that represents the

by Antonia Bertolino, Eda Marchetti and Raffaela Mirandola

Decision making in project management requires accurate and automated supportto monitor and control the development process. Managers must make reliableschedule predictions and optimize personnel utilization. They thus must be ableto dynamically evaluate whether the resources assigned to a job are sufficientand whether the organization structure is adequate to meet the scheduleddeadlines. To support managers in these tasks, scientists at ISTI-CNR are studyingthe potential of classical performance engineering and propose the Propean(Project Performance Analysis) methodology.

Decision Support for Personnel and ResourcesManagement in Multi-Project Environments



resource usage of the node. The MMmodels the hardware platform and isbased on Queueing Network (QN)models.

The MethodologyStarting from the above metaphor wecapture the aspects relative to the activityplanning in the SM, and those relative topeople utilization and distribution in theMM. These models are derived fromSequence Diagrams (SD) andDeployment Diagrams (DD) respec-tively, and are automatically combinedby Propean which generates a uniqueQN, that can be analysed by traditionalperformance analysis tools.

The methodology is currently beingimplemented. Figure 2 illustrates the toolarchitecture including: a UML case tool(specifically ARGO) which processes theinput diagrams and generates an XMLfile; a component which takes as input theXML file and produces the correspondingQN; a QN analytical solver or simulator,which derives the performance analysisresults; and finally, a results converterwhich analyses the performance resultsand converts them back into UML.

Here below, we outline the basic steps ofthe Propean methodology:1.Analysis: definition of project activi-

ties (manager)2.Modelling: definition of the SDs and

the DD (manager)3.Model annotation: specification of

parameters and values (manager)4.SPE models generation: derivation of

the SM and MM models (automatic)

5.Model evaluation: resolution of theQN and derivation of the relevantpredictions (automatic)

6.Analysis of results: evaluation of theresults obtained (manager).

ApplicationsOne of the advantages of using perfor-mance techniques for project manage-ment is their natural capability to handlemultiple projects and their mutualinvolvement in schedule and resourcesusage. In fact, Propean results particularlyuseful in environments where multipleprojects compete for common resources,which is for instance the case in most soft-ware development environments.

Propean is the result of a collaborationbetween ISTI-CNR in Pisa, theUniversity of Rome Tor Vergata andEricsson Lab Italy (ERI) in Rome. It hasbeen applied with encouraging results tosome industrial case studies. Here wemention two of them: one is thewidespread Rational Unified Process(RUP) customised for a specific ERIproject. Considering RUP itself as aproduct, Propean provided answers to thefollowing questions: How long will RUPtake to process a certain project? Howwill RUP utilize the available resources?How is the RUP schedule affected by theconcurrent processing of severalprojects? The application of Propean toRUP was divided into two steps: the first,called RUP modelling, consisted of thedescription of the functionality and archi-tecture of the RUP product by means ofUML diagrams annotated according tothe RT-UML profile. The second step,called RUP customisation, representedthe core of Propean application andincluded the refinement of the UMLdiagrams according to personnel avail-ability and process requirements.

In a second case study, Propean was usedto support the decision to release a soft-ware product, after testing is completedwith no trouble report left open. Propeancan assist managers in the early predictionof the expected time until release, consid-ering the actual resource availability (withpersonnel multitasking). Alternatively, ifthe release time is fixed with thecustomer, Propean can help decide themost adequate personnel configuration tofulfil the time constraints.

Propean can provide useful support todecision makers in project management.The tool can provide valuable input foranalysis, but human expertise remainsessential to model the processes, to tunethem with the correct values and tointerpret the results obtained.

Links:http://www.isti.cnr.it/ResearchUnits/Labs/se-lab/index.html

http://www.omg.org/technology/documents/formal/schedulability.htm

http://www.perfeng.com/

Please contact:Eda Marchetti, ISTI-CNRTel: +39 050 315 3467E-mail: [email protected]

Figure 1: Different approaches

to decision-making.

Figure 2:

Propean Tool

Architecture.

Performance Model

UML to QN Component

UML Tools

UML Model (XML)

Results Converter

Analysis of Performance Results

QN Solver



Traffic on the Swedish railway networkhas increased considerably in recentyears, and large investments in railwayinfrastructure are planned in order tomeet the expected need for higherrailway capacity. When new infrastruc-ture is designed, it is important for thequality of the transportation system thatthe infrastructure and the traffic solu-tions are adapted to each other. For thispurpose, new standards, methods andtools are required for precise analysis ofinfrastructure capacity.

Our starting point is that railway traffic isto be regarded as a set of traffic patternsthat are repeated on sections of the tracksystem. From this standpoint we havedeveloped a method for analysingcapacity in signalling design. Considerthe left side of Figure 1, which depicts aninstance of periodic traffic, by which wewill describe the method. Each period ofthis repeated traffic pattern consists ofthree trains (w, e and f) traversing asingle line with three meeting stations(ms1, ms2 and ms3). Each of theadvances of the two trains w and e areseparated into two parts, w1, w2 and e1,e2 respectively, by the waiting points at

which the trains wait for other trains topass. We use the term ‘cleared paths’ toindicate the paths of the trains producedby this separation. Waiting points andcleared paths are the basic concepts onwhich the method relies.

The 'traffic pattern graph' on the right inFigure 1, shows the priorities of themovements on the cleared paths in thetraffic pattern. There is one node in thegraph, for each 'cleared path'. Straightarcs concern prpriorities in one and thesame period, and bent arcs concernpriorities in two consecutive periods. Forinstance, the bent arc from e2 to w1means that the movement e2 in period nprecedes the movement w1 in periodn+1. The arcs only concern direct priori-ties.

The compact model of the traffic patterngiven by the traffic pattern graph is thefirst step in the construction of a set ofconditions to be used for estimatingcapacity. The next step in this construc-tion is the addition of arcs for move-ments that may directly impede on eachother. This is followed by the decorationof each arc with the shortest length of

time between the starting times of thetwo movements. The result, for someinvented shortest durations betweenstarting times, is the 'conflict graph'shown in Figure 2.

Each arc in the conflict graph is a condi-tion. For instance, the bent arc of weight25 from f to e1 represents the conditionthat the movement e in period n + 1 maynot start earlier than 25 time units afterthe start of the movement f in period n.Given the conflict graph, the minimumperiod is obtained by analysing thecircuits in the conflict graph, where acircuit consists of one or more arcs. Thecycle mean of a circuit is the sum of theweights divided by the number of bentarcs, and the minimum period of thetraffic pattern equals the maximum ofthe cycle mean for all circuits in theconflict graph. The maximum cyclemean in a conflict graph can be effi-ciently computed, with running timeO(n3) , at worst, with n as the number ofcleared paths. The minimum period forthe considered traffic pattern is 50,where the circuit of weight 100consisting of the edges (w1,w2),(w2,e1), (e1,e2) and (e2,w1), two of

by Jan Ekman

With the aim of supporting the process of adapting the railway infrastructure topresent and future traffic needs, SICS is developing methods and tools forestimating railway capacity. The project is supported by Banverket, the SwedishNational Rail Administration.

Towards Useful Tools for Estimating Railway Infrastructure Capacity

Figure 2: Conflict graph.Figure 1: Traffic pattern and traffic pattern graph.



In the railway industry, the traditionalmethod has been to first produce thetimetable, and then to assign resourcessuch as engines, cars and personnel tothe scheduled transports. Nowadaysassignments are done with optimisationtools that are based, for example, onnetwork flow models.

However, since the deregulation of therailways in Sweden, and due to thelimited capacity of the track resources,competition has increased among opera-tors for attractive timetable slots.Although the commission of theEuropean Union has stated that theEuropean railway shall be deregulated,different countries have reached differentdevelopmental stages. Sweden has dereg-ulated cargo traffic and some passengertraffic, and has separated railwaymanagement and slot allocation fromtraffic operations.

The competition for track slots comespartly from varying customer demands

on the railway operators, and partly fromthe growing interest in the efficientusage of expensive resources such asengines (see Figure 2). It is increasinglyapparent that there are two ways to effi-ciently use resources, namely, optimalassignment of resources to tasks, andwell-scheduled departure and arrivaltimes, which facilitate efficient assign-ment. Before the application fortimetable slots is made, each operator

must find the best compromise betweenresource usage and customer demands.This stage in the planning process, madecritical by deregulation, currently suffersfrom a lack of good decision-supporttools, and improvements are necessary inorder to facilitate good assignment inlater stages.

Over the last few years, SICS hasaddressed this early process stage. For ahuman, the variables of geography, timeand assignment mean the problemrapidly becomes very complex. To helppractitioners, SICS has developed aseries of computational models for thisprocess, in particular one constraint-programming model and one integer-programming model. The latter iscurrently being integrated into Sweden’slargest cargo operator. The model is anextension of a network flow model, inwhich time windows have been incorpo-rated, leading to a mixed-integerprogram. By relaxing the departure andarrival times to be within lower and

by Martin Aronsson

The raison d’etre of the transportation industry – moving things between locations– is also the source of a key problem; namely, that the resources necessary foroperation are geographically distributed. This means scheduling can become anextremely complex problem, particularly in comparison to stationary industries,specifically traditional job-shop scheduling.

Efficient Utilisation of Resources in theRailway Industry by Levelling Departures and Arrivals of Trains

Figure 1: Efficient resource utilisation is

a key factor in the railway industry.

which are bent, is a critical circuit. Thatis, a critical circuit is a circuit with cyclemean equal to the maximum cycle mean.

To sum up, this method allows us todecide the minimum period for a giventraffic pattern on a railway section. Thisis in itself useful for deciding on the bestinfrastructure for the planned traffic,given, for instance, by a timetable. Themethod applies to precise examination ofinfrastructure, such as considering theeffects of adding and altering the posi-tions of track signals. The methodsuggests that capacity of a railway

section can be estimated as the overallminimum period for a collection of auto-matically generated traffic patterns, eachof which satisfies some requirements onthe traffic. It may also be of interest toknow, for a given fixed infrastructure,which traffic solutions are suitable andwhich are not. In such a case, rather thansimply estimating the capacity, it is alsonecessary to determine which trafficpatterns are badly adapted to the infras-tructure, ie, those traffic patterns with along minimum period. Without payingany attention to robustness of the trafficsystem, analysis of capacity would be

merely of theoretical interest. One wayof dealing with robustness is to adddefault buffer times to the course ofevents on which the model is built,making this set of buffer times therobustness criterion. However, the modelof the traffic pattern given by the conflictgraph is well suited as the basis forrobustness analysis, as for instancestability analysis and delay propagation.

Please contact: Jan Ekman, SICSTel: +46 8 633 15 65E-mail: [email protected]

Pho

to: G

reen

Car

go.



upper bounds, and adding constraintsthat ensure that certain relations betweentrains still hold (eg for cars that should bedecoupled and attached to another train),the tool can produce efficient rotationsfor the engines while moving the trans-port operations inside the time window.In the general case, a network flowmodel with time windows becomes verycomplex, but in practice it turns out thatwe can compute solutions to the gener-alised problem as long as the timewindows are not too large. Using itera-

tive application of the process, we havesuccessfully computed solutions for timewindows up to sixty minutes.

In the stationary industry, this process ofmoving tasks in time to reduce theresources needed is sometimes referredto as ‘levelling’, since it tries to level theresource usage over time. While this is asimpler task in stationary industries, it isjust as important in the transportationindustry. In Figure 3, a typical resourcegraph is presented for a railway cargo

company. It is the peaks that set thenumber of resources (engines) neededfor the transportation stock. By using thelevelling software, the resource peakscan be considerably reduced. Our full-scale tests have shown that 5-10%resource savings are within reach.

Link: http://www.sics.se/iam/

Please contact:Martin Aronsson, SICSTel: +46 8 63315 87 E-mail: [email protected]

In railway systems, it often occurs thatthere is an imbalance in the freight flows,meaning empty freight cars must bemoved in order to enable new transports.An efficient empty freight car distribu-tion process is vital for good utilisationof the freight cars. A bad distributionprocess can result in delayed transports,good-will losses and an unnecessarilylarge car fleet with the associated capitaland maintenance costs.

Green Cargo is the largest railway cargocarrier in Sweden. In collaboration withCarmen Consulting and LinköpingUniversity, Green Cargo has developedan advanced new system for planningtheir empty freight car distribution. Theaim of the system is to improve thequality of the empty freight car distribu-tion process. The system handles about7000 freight cars, divided into 25 types.The railway network has about 1500pick-up and delivery points for freightcars, and the system produces detailed

transportation plans between thesepoints.

The kernel of the new system is an opti-misation model that calculates distribu-tion plans in real time. The model isconstantly updated and re-optimised asnew information becomes available, egupdated freight car status, transportorders or train capacities. The responsetime of the system for re-optimising theplans is less than one second. The plansalways reflect the optimal distribution

by Martin Joborn

The rail freight carrier Green Cargo has a new system for rail freight car distributionplanning. It is an automatic, real-time planning system, based on a kernel ofoptimisation models and methods. The new system has resulted in both significantsavings and quality improvements.

Real-Time Optimisation of Empty Rail Freight Car Distribution

Figure 2: Process flow, customer demands and resource utilisation requirements

restrict the timetable process.

Railway Administration

Operator Customer Demands

Traffic Plans

Timetableconstruction

Levelling

Optimised assignement

Figure 3: Resource utilisation as a function

of time (5 days) with peaks every night.



for the current situation. The objective ofthe distribution planning is to maximisethe number of filled transport orderswhile minimising transportation costs, aswell as to generate distribution planswith a preferred structure and to ensurethat cars arrive punctually to customers.

The high performance of the systemmakes it possible to respond immedi-ately to customers’ transportationdemands. While on the phone placingthe transport order, the customer willalso be informed of whether freight carsare available for the transport. The distri-bution system generates an online opti-misation to calculate and guarantee thedelivery of the empty freight car to thetransport. However, regarding themethod of satisfying a customer’s freightcar demand, the system maintains flexi-bility for as long as possible, whilesimultaneously ensuring that the orderwill be filled.

The system is fully automatic and needsno manual interaction to execute thecalculated plans. At the correct time,stations, terminals and operating crews

receive instructions relevant to their carsand transport orders - bookings, shuntingorders etc – from the system.

The introduction of the new system hasmade empty freight car distribution atrustworthy and highly precise process,which has lead to loyalty from bothpersonnel and customers. The reliabilityof the process decreases the need forbackup solutions such as local freight car

inventories. Furthermore, the freight carcontrollers have a good overview of thefreight car supply-and-demand situationover the system’s planning period of oneweek.

The improved utilisation of freight carshas enabled Green Cargo to reduce itscar fleet and the associated capital andmaintenance costs. Green Cargo esti-mates its yearly savings at €7.6 million,excluding the benefits from improvedcustomer relations.

The concept used at Green Cargo isapplicable to other problems as well.Interesting potential applications includeloaded freight car planning, taxi plan-ning, truck allocation, and containerdistribution.

Link: http://www.carmenconsulting.com/includes/ProjectSheets/RealTimeOptimization.pdf

Please contact:Martin Joborn, Carmen Consulting, SwedenTel: +46 8 555 79 438E-mail: [email protected]

Full control over freight car planning at

Green Cargo.

Pho

to: P

eter

Lyd

én

These algorithms enable the forest industry tooptimise the transport of wood to sawmillsand paper and pulp mills using a nationwidecalculation area. The entire operation includesabout 250 trucks, about 100 points of deliveryand about 10 000 wood batches daily. TheKUORMA system is part of a more extensiveinformation system, which incorporates opti-misation algorithms for wood harvesting,transport planning and routing.

The KUORMA system was developed in acommercial project headed by TietoEnator asa systems integrator. VTT InformationTechnology was a sub-contractor responsiblefor the development of transportation algo-

rithms. The customer was Stora Enso Finland.Stora Enso uses about 17.3 Mm3 of domesticwood annually, corresponding to 288 000truckloads.

Transportation is a remarkable source of costin the forest industry, especially in Finlandwhere the distances are long and the woodownership includes private citizens. Not onlyare wood batches of varying sizes aredistributed over some 3 000 000 roadsegments, but the price for standing timber isrelatively high. In order therefore to reducewood procurement costs for competitivenessreasons, one must concentrate on transporta-tion costs. Stora Enso had a system developed

by the coalition in the early 1990s that usedgreedy algorithms. Its performance and theacceptability of the solutions it provided hadreached their limits. Furthermore, improve-ments in technology had made it possible touse a low-cost Linux workstation cluster witha large amount of essential RAM memory.

The development project started in July 2001,and through October 2002 the system wasgradually improved through operational use.

The Techniques EmployedThe main components of the KUORMAsystem are the road network infrastructure, the

by Juha Savola, Hannu Rummukainen and Olli Jokinen

The KUORMA system (English acronym LOAD) is a combination of APS (AdvancedPlanning and Scheduling) algorithms and the system infrastructure services usedin the field of forest industry.

KUORMA: A Collection of APS-Algorithms for Forest Industry Wood Transport



load cluster model, the load assignmentmodel and the routing model.

The key information for the road networkinfrastructure is the nationally administratedand updated road material, which covers thewhole of Finland and contains about 3 000000 segments of road, including small, privateforest roads. The central attributes are roadclass, x-y geometry and length. The basic roadmaterial is organised as an undirected graphstructure. Because the set of current objectswith a coordinate varies all the time, the calcu-lation of distances had to be performeduniquely for each data set. Nevertheless, theoperative calculation effort can be reducedusing so-called supermatrices (size 15 000 –20 000 squared) based on all the distinctpoints at any given time. Supermatrices arecalculated in a Linux cluster during idle hours.When the actual matrices are calculated util-ising supermatrises, the hit ratio can exceed 95

%, depending of course on the time betweenthose two calculations. The calculation time istypically a few minutes using a 2.2GHz dual-processor PC with 2Gb RAM. The basic solu-tion given by the road infrastructure is thefastest or shortest distance between any twoobjects using three different road weightclasses.

Load Cluster ModelIn the first stage of the KUORMA system,batches of wood in the forest are clustered intoloads that fit into trucks. This is done regularlyas newly cut stands are entered into thedatabase, and old wood is transported away.

Large batches are easy to divide into truck-loads, but due to the large number of privateforest owners in Finland, many batchescontaining several wood types are quite small,and may only be accessible by low-quality,temporary roads. The road quality is modelled

by a weight limit, which represents a hardconstraint for the order in which the batches ofa load can be collected. The primary objectivein the optimisation model is the minimisationof total driving time for load collection.

Load Assignment ModelThe assignment model calculates the bestdestinations for all loads to be transported.The destinations are factories, sawmills andrailway stations (for final transportation bytrain to the factories). Assignment is taken asinput transportation costs based on distances,the daily wood requirements of the destina-tions, the availability of loads per day andregional and team goals.

Wood requirements are described as amountsfor different kinds of wood and penalty costsfor deviation from the desired amount. Thecalculation period is divided into an opera-tional period (a few days) and a tactical period(some weeks). In the operational period, desti-nations are assigned for every load. In thetactical period, standing wood marked forcutting is optionally reserved for certain desti-nations because the loads are not yet to betransported.

The results generated by the model show thedestinations for loads, the transportation daywithin an operative period, and the openingday for every stand marked for cutting in thetactical period.

Load Routing ModelThe third stage comprises the routing andscheduling of trucks for the immediate future.At this stage, the load destination can nolonger be changed, but the choice of whichloads to transport in the operational period is

Figure 1:

A truck

unloading

at a paper

mill.

Figure 2: Snapshop from road network of a load cluster model.

Before reduction.

Figure 3: Snapshop from road network of a load cluster model.

After reduction.



The system allows customers to requestlarge amounts of bandwidth for shortperiods of time, with a firm guarantee ofhigh quality of service (QoS). Requestsare processed automatically and, ifaccepted, are provisioned automatically.The system ensures that requests areonly accepted if all QoS guarantees canbe met.

A typical example would be a two-hourvideo conference starting at some fixedtime, which requires a reliable (lowpacket-loss), low latency connectionbetween a number of locations aroundthe world. A requirement might beimposed that the accepted demands mustnot cause the service level agreements(SLAs) of any existing customer tobreak. To meet this requirement it isnecessary to account for the bandwidthuse in the network. The system does thiswith an automated workflow.

The customer books the BoD service on-line via a portal up to 24 hours inadvance. If the request can be acceptedhe receives an email confirmation withintwo hours. At the requested start time, atraffic-engineered (MPLS-TE) tunnel is

set up between the service end-pointsand the access control bandwidth limitsare modified. When the service is over,the tunnel is automatically removed andaccess control limits are reinstated forthe customer. All these operations areperformed automatically without humanintervention: the customer is ‘self-provi-sioning’ the request on the network.

At all times, the BoD system controls theavailable bandwidth and rejects userdemands that exceed the availablecapacity for the requested time period. Inthis way QoS is guaranteed for allservices at all times.

The system uses different hybrid solvers: • the Route Generator module finds

routes in the network which satisfydelay and capacity limits

• the Commitment Solver compares thecommitted bandwidth in the networkwith current and historical trafficmeasurements and determines theavailable capacity for additionalservices.

These solvers are implemented inECLiPSe, our constraint-programming

environment, and use a combination ofconstraint, local search and mathematicalprogramming techniques. The hybridisa-tion allows us to find near-optimal solu-tions in a limited time, and to scale thealgorithms even for large networks.

The system was developed jointly byParc Technologies and IC-Parc, ImperialCollege London for an industrial partner,and has been implemented on a globalMPLS network. Parc Technologies is astart-up company developing decision-support systems for network problems.IC Parc is a research department in thefaculty of Engineering, Imperial CollegeLondon. The Parc Bandwidth onDemand Solution is only one of severalconstraint-based systems that have beendeveloped by Parc Technologies in thenetworking area in recent years.

Link:http://www.parc-technologies.com

Please contact:Helmut Simonis, Parc Technologies Ltd, UKTel: +44 207 594 8463E-mail: [email protected]

Running demanding services like video conferencing over the Internet is noteasy, with packet loss and jitter making it a less than enjoyable experience. TheParc Bandwidth on Demand (BoD) system helps Internet service providers to selland differentiate high-quality, short-term connectivity for video conferences andother services, thereby avoiding these problems.

Parc Bandwidth on Demandby Helmut Simonis

free subject to certain constraints. The primaryobjective is to minimise the total driving costs.Trucks may operate all over Finland, but theymust regularly return to their home areas toexchange drivers, and may also have sched-uled days off. The paper mills and sawmillsoperate on exacting schedules, and musttherefore be provided with specific numbersof loads of different types within given timeintervals, although deviations are possiblewith associated penalty costs. There are anumber of other constraints and secondarygoals, making this the most complicated opti-misation model in the KUORMA system.

Future ActivitiesThe transportation branch of the forestindustry stands to benefit greatly from optimi-sation algorithms. Stora Enso anticipates a5% saving in wood transportation costscompared to the old, greedy-algorithmapproach. Our approach can also be applied toother transportation segments.

Further work on the KUORMA system couldinclude evaluating multimodal transports,adaptation of new road material containing z-coordinates (hills) and making use of dailyweather data.

Links:http://www.vtt.fi/tte/projects/fleet_management/kuorma.html

http://www.vtt.fi/tte/projects/fleet_management/dynatrans.html

Please contact:Juha Savola, Hannu Rummukainen, Olli Jokinen, VTT Information TechnologyE-mail: [email protected]



Nowadays simulation programs arewidely used in the car industry to opti-mise production processes or to test thebehaviour of a car before a real prototypeis built or real tests are performed.Simulation helps to reduce the number ofreal prototypes and the time betweenmodel changes compared to real tests.Crash simulation is one of the most time-consuming simulation tasks and is there-fore computationally expensive. Manysimulation runs with different parametervalues are required to optimise construc-tion parameters, and the results have tobe analysed by an engineer. In this casethe use of a numerical optimisationtoolbox, which analyses different para-meter constellations, can support thedevelopment process significantly. Theoptimisation toolbox uses the PAMCRASH simulation program to calculatethe crash properties of a car model. Foreach given set of parameters the simula-tion results have to be evaluated by anobjective function. This article focuseson a box beam test example whichdemonstrates the applicability of thetoolbox to crash simulation problems.

The definition of an objective functionwhich evaluates the crash simulationscan be difficult for technical applica-tions. In the case of the box beamexample the objective function isconstructed as a comparison between areference crash and a crash of a modelwith modified material parameters. Theaim of this procedure is to define an opti-misation problem with a well-knownsolution to test the toolbox and the opti-misation algorithms. In the referencecase the acceleration of a box beam withgiven Youngs modulus and Poisson ratiois calculated and the objective function isdefined as the integral of the quadraticdifferences to the acceleration of a model

with modified parameters over the time.This integral can be approximated by asum over discrete time steps:

Thus the minimum of this objectivefunction is reached if the material param-eters of the modified model reach thevalues of the reference case.

Figure 2 shows a plot of the objectivefunction versus change of the Youngsmodulus. The minimum of this functionis at 210 kN/mm2 and is identical withthe reference case. Close to its minimum

the function is smooth but on the left sidethe behaviour is rather chaotic. Underrealistic conditions, almost all propertiesof crash experiments are subject toscatter, although it is desirable to avoidthe scatter of a model as much aspossible. Numerical properties of thesimulation codes as well as certainfeatures of the crash model may beresponsible for these in-stabilities.Typical sources of scatter which arecaused by the model are buckling andcontact under an angle of 90°. Furtherinvestigation of this effect with the boxbeam example indicated that the scatterresults from an instability between two

by Christof Bäuerle, Clemens-August Thole and Ulrich Trottenberg

DesParO is an optimisation toolbox designed for industrial simulation. The toolboxcontains a collection of efficient algorithms for computer-based optimisation andcan easily be adapted to any industrial simulation code. DesParO was especiallydesigned for computationally expensive simulation codes and supports parallelcomputation.

DesParO - A Design Parameter Optimisation Toolbox using an Iterative Kriging Algorithm

Figure 1: Crash example box beam. Figure 2: Scan of the objective function

for the test problem.

Figure 3: One possible mode of the crash

simulation.

Figure 4: Convergence of the different

optimisation algorithms.



different modes. Figure 3 shows onesuch mode. In this case the upper edgeon the right side is shifted outward andthe lower edge is shifted inward. In thesecond mode of the model both edgeswere displaced outside the box beam.

Scatter of a model is a difficult conditionfor optimisation and an engineer tries toavoid instabilities of a construction. Inthe test case of the box beam example itturned out to be possible to stabilize themodel by a small modification of theconstruction. In particular, if both edgesof the original model are bent to one sideit is possible to suppress the asymmetricmode during the crash simulation and thescatter of the objective function can bereduced significantly. This procedurenow leads to a well defined optimisationproblem.

For the solution of nonlinear problems awide literature on optimisation strategiesis available. Popular strategies are conju-

gate gradient, downhill simplex, andsurrogate models. An important charac-teristic of the optimisation of crash simu-lations is that it is usually very time-consuming for a realistic car model.Optimisation strategies based on surro-gate models are especially suited for opti-misation under such conditions. In thefirst step design of experiment methodsare used to choose parameter values, forwhich a simulation has to be executed.The values of the objective function forthese parameters can now be used toconstruct a surrogate function as an inter-polation of the real problem. The evalua-tion of the surrogate function is very fastand it is much easier to search for theminimum of this interpolation. Duringthe next steps the design of experimentpoints can be refined iteratively nearbythe minimum of the surrogate function.In Kriging the surrogate model consistsof a sum of basis functions, very often ofGauss type. Figure 4 shows the relativeconvergence of different optimisation

algorithms at the stabilized box beamexample which was described in thisarticle. Although the Kriging algorithmconverges very fast during the first itera-tions at this problem, the optimisationstrategy does not work so reliably if theobjective function is not smooth enough.

Future work will concentrate on theaspect of improving the robustness of thesurrogate models by replacing the designof experiment methods by pattern searchstrategies. Furthermore it is planned toextend the surrogate models to unstablecrash constructions. In this case surro-gate functions are used for approxima-tion (instead of interpolation) of theobjective function.

Links:http://www.scai.fraunhofer.de/35.0.htmlhttp://www.auto-opt.de

Please contact:Christof Bäuerle, Fraunhofer SCAITel: +49 2241 14 27 94E-mail: [email protected]

Instrumentation and control (I&C)systems are used in safety-critical appli-cations to monitor issues such as opera-tional safety and service availability.They are gradually being replaced bysoftware-based digital systems, and inorder to enhance reliability in the designand operation phases of these software-based I&C systems, a number of tech-niques must be applied. One of the best-known of these is periodic testing duringnormal operation.

We have created a testing methodologyfor safety-critical supervisory systems,which has been implemented at the PaksNuclear Power Plant located in Hungary.Each of four reactor units has a dedicated

Reactor Protection System (RPS) for theintervention and safe shut-down of theunit in an emergency situation. Thissystem contains three redundancies, alsocalled ‘trains’. An RPS, like all physicalsystems, is also susceptible to faults,including sensor failures and componenthardware faults. Naturally, each RPS hasa fault-tolerant, highly redundant hard-ware and software architecture.However, faults can cause latent errorsthat may become activated only in emer-gency conditions. To eliminate these andensure reliable operation in all situations,an RPS must be periodically testedduring normal operation by special testequipment.

The Test Concept It is necessary that the test equipment beable to create experimental scenarios ofinput signals in one part of the protectionsystem, without affecting the operationof the other parts or the RPS as a whole.While testing a certain train in a 2/3redundancy system, only two trainsremain in normal operation. This situa-tion is very sensitive, since an erroroccurring in either of the two operatingredundancies could initiate an emer-gency protection action. The train to betested must be selected by a corre-sponding ‘Test Enable’ (TE) signal,while the TE signals of other trains aresimultaneously disabled. During the test,

by István Varga, Tamás Bartha and Alexandros Soumelidis

The most important issue for safety-critical supervisory systems is the correctoperation of every component. The Systems and Control Laboratory (SCL) ofSZTAKI has developed a new validation and testing concept to ensure the correctfunctionality of safety elements. Three implementations have been completedand are regularly used in the Paks Nuclear Power Plant (Hungary).

On-line Testing of the Reactor ProtectionSystem in the Paks Nuclear Power Plant



the process parameters are set in such away that safety actions are initiated.

The test execution makes the 2/3 redun-dancy system sensitive to errors. For thisreason, the two trains not being tested areconstantly monitored for errors duringthe test. If an error is detected in thesesystems, the test stop criterion disablesthe TE signal of the train being tested.This causes the test to cease immedi-ately, and the train returns to normaloperation. This ensures that two fault-free trains are always available forcorrect voting, and thus we avoid theactivation of unnecessary EP actions.

The Test EquipmentThe principles shown above have beenimplemented in the RPS of the PaksNuclear Power Plant. We created threerealisations of the test concept: testmachines for the start-up and operationalphases of the reactor (the Start-up Test

Machine and Periodic Test Machinerespectively), and an improved testsystem, which includes the completefunctionality of the earlier machines(Universal Test System).

The periodic testing concept of the RPSsystem suggests a stand-alone TestMachine, which is temporarilyconnected to test inputs from the systemin order to perform test procedures.However, a Periodic Test Machinedesigned and built on this basis provedinconvenient and difficult to use in prac-tice. Therefore, the revised test conceptincludes a distributed, general-purposetest system called the Universal TestSystem, or UTS for short.

Universal Test SystemThe basis of the distributed test equip-ment is an RS485-type industrial bussystem (CAN), which provides the inter-face to the task-specific ‘active’ test

plugs (analogue test plug, binary testplug, control plug etc; see Figure 2). Themicroprocessor-controlled analogue andbinary active plugs contain all the devicesnecessary to perform a full test of theanalogue and binary input modules. Theactive control plug contains transceiverand receiver components to send andreceive control signals to/from the RPSsystem. The design of the active testplugs, the microcomputers and theRS485 bus interface requires implemen-tation in high density SMD technology.

The active test plugs are powered by theRPS system. When all the safety require-ments are equally satisfied, galvanicisolation must only be realised on theCAN interface. The analogue and binaryplugs can be applied simultaneously tomultiple input modules, even to a wholecabinet or a complete train. An industrialcomputer (Local Test Machine) controlsthe active test plugs using a CANcontroller card. The Central TestMachine (CTM) supervises the threeLocal Test Machines, each of which isdedicated to a single RPS train, as shownin Figure 1.

The Universal Test System is in regularuse at the Paks NPP, and has helped toimprove the reliability of the RPSsystem, while reducing the time, andtherefore cost, dedicated to testing.

Link: http://www.sztaki.hu/scl

Please contact:Tamás Bartha, István Varga, SZTAKITel: +36 1 279 6227E-mail: [email protected], [email protected]

Figure 1:

Structure

of the

Universal

Test

System.

Figure 2:

UTS plugs.

Figure 3: The CTM software during a test.

Local TM LTM-Y

Central Test Machine CTM

Y Train X Train W Train

Universal Test

System

Intelligent UTS Plugs

L2 CAN

Local TM LTM-X


L2

Local TM LTM-W


L2 CAN CAN

Nuclear System

Uni

t Com

pute

r



A rough definition for process systems isthat the operation of such systems isbased on the laws of thermodynamics.Well-known examples of processsystems include heat exchangers, chem-ical reactors, distillation columns andwastewater treatment systems. TheProcess Control Research Group is partof the Systems and Control Laboratoryof SZTAKI, and studies the modelling,analysis and control of process systems.

Process Modelling and Model AnalysisProcess models for control and diag-nostic purposes most often take the formof differential-algebraic equations(DAEs), with the conservation balancesbeing the differential part and the consti-tutive equations the algebraic part. Anumber of hidden relationships existbetween the equations and their terms,as dictated by the laws of thermody-namics, unit operation, reaction kinetics,heat and mass transfer. Together withthese relationships, the equations can beseen as structured texts (like computer

programs) with a syntax and semanticsdefined implicitly by the underlyinglaws of thermodynamics and chemicalengineering. This approach enables usto use formal methods of computerscience to construct, verify, analyse andsimplify process models in a rigorousand automated way. Our latest results inthis field are an assumption-drivenmodel-building procedure, a systematicmodel-simplification method and anassumption-retrieval procedure for thedetermination of simplification assump-tions of two related process models.Furthermore, the prototype of anassumption-driven model editor hasbeen developed in Prolog language.Cooperating partners in this researcharea are the Department of ComputerScience at the University of Veszprémin Hungary and the Computer AidedProcess Engineering Centre in theDepartment of Chemical Engineering atthe University of Queensland inAustralia. Several years of joint workresulted in a book entitled ’ProcessModelling and Model Analysis’, written

by Katalin Hangos and Ian Cameron andpublished by Academic Press in 2001.

Intelligent and Discrete Process Control Intelligent and discrete process controland diagnostic methods apply so-calleddiscrete event system models, which arecurrently one of the hottest researchtopics in systems and control theory.Such systems are widely found in prac-tical applications and are characterisedby the presence of discrete valued vari-ables: state, input and output variableswith a finite number of possible values.At present, our research in this field isfocussed on the hierarchical decomposi-tion and analysis of process modelsusing hierarchical coloured Petri Nets.Our previous results on.the invarianceinheritence in Petri Nets and the applica-tion of Artificial Intelligence methods ingenerating operating procedures serve asa starting point for this work. Recently,we successfully utilised hierarchicalcoloured Petri Nets for multiscale-modelling, deadlock-detection and fault

by Katalin Hangos and Gábor Szederkényi

Combining a priori phyisical and engineering knowledge with theories of artificialintelligence, discrete event systems and modern nonlinear control has led to newtheoretical results and successful applications in the field of modelling, analysisand control of process systems at the Process Control Research Group at SZTAKI.

Process Systems: Theory and Applications from Different Aspects

Figure1: Reachability surface in the state space of a simple fed-

batch fermentation process.

Figure 2: Screenshot from the assumption-driven model editor

software (analysis of model transformations).



diagnosis in an industrial granulatorsystem. The simulation results run on thevalidated system model are promising,making the next short-term goal toimplement the developed techniques onthe real system.

In 2001, Kluwer Academic Publisherspublished a book in this field with thetitle ’Intelligent Control Systems (AnIntroduction with Examples)’, written byKatalin Hangos, Rozália Lakner andMiklós Gerzson.

Nonlinear Process Control Nonlinear system theory and thenonlinear control and diagnosis methodsbased thereon form one of the most chal-lenging and fast-developing areas inmodern systems and control theory. Itssignificance is explained by the fact thatmost engineering systems (and processsystems in particular) show nonlinearbehaviour, with a few exceptions ofcontinuously operated systems near awell-defined and stable operation point.

Apart from its practical importance,nonlinear systems and control theoryrequires the understanding and applica-tion of advanced mathematical conceptsand tools from the areas of nonlineardifferential equations, functional anal-ysis, differential manifolds and algebraand symbolic computations. It turns out,however, that the principles of the engi-neering discipline governing thebehaviour of the nonlinear system play akey role in the understanding, analysisand control of engineering systems. Thusthe use of engineering knowledge deter-mines certain structural properties of anonlinear system, and this can make itfeasible or even easy to handle otherwisevery difficult problems. This type ofinterdisciplinary approach to controlproblems is commonly called a ’grey-box’ approach. Recent research resultsshow that this approach can be success-fully applied for the nonlinear analysis ofcontrol-related features such as local andglobal stability, reachability, observ-ability, passivity and zero dynamics.These analysis results together withphysical and engineering knowledgeallow us to understand the dynamicbehaviour of process systems in moredepth, and to select and design nonlinear

controller structures with the propertieswe require.

In the field of nonlinear mechanicalsystems, the Hamiltonian description ofthese systems gives rise to their specialcharacterisation and the development oftheir very powerful passivity-basedcontrol. The analogy with theHamiltonian description of mechanicalsystems has led to the Hamiltoniandescription of process systems.

The developed methods were success-fully tested on a variety of processsystems: • fermentation processes with highly

nonlinear reaction kinetics operated incontinuous or fed-batch mode

• a low-power nonlinear gas-turbine (incooperation with the Department ofAircraft and Ships at the BudapestUniversity of Technology andEconomics and the János BolyaiMilitary Technical College Faculty ofthe Miklós Zrínyi National DefenceUniversity)

• a nonlinear electro-pneumatic valvewithin the brake system of trucks (incooperation with the Research andDevelopment Center of Knorr-BremseBrake-systems Ltd. in Budapest).

Our most important partner in this areaof activity is the Systems and ControlLaboratory of SZTAKI, and a significant

result of this cooperation is anotherbook, entitled ’Analysis and Control ofNonlinear Process Systems’ and writtenby Katalin Hangos, József Bokor andGábor Szederkényi, which is to bepublished by Springer-Verlag in early2004.

Links: http://www.sztaki.hu/sclhttp://daedalus.scl.sztaki.hu

Please contact:Katalin Hangos, SZTAKITel: +36 1 279 6101 E-mail: [email protected]

Figure 3: Coloured Petri Net model of a valve sub-system.

Agent/multi-agent systems have becomean important field within artificial intelli-gence research. They have found anumber of applications, includingcontrol processes, mobile robots, air-traffic management and intelligent infor-mation retrieval.

Here we present a multi-agent system forsolving the automatic allocation problemin a container terminal. The operationscarried out in such terminals are amongthe most complex tasks in the transportindustry. This is due to three factors: thediverse range of entities performingoperations, their interaction with adynamic environment, and thedistributed nature of the problem, inwhich independently functioningsystems directly affect each otherthrough their actions.

Existing centralised and sequentialapplications for container terminalmanagement are now insufficiently flex-ible to respond to changing managementstyles and the highly dynamic variationsin loading/unloading requirements.Traditionally, the entire terminal iscontrolled by centralised software,which limits the expansion and reconfig-uration capabilities of the system. Usinghierarchical organisation forces thegrouping of resources into permanent,

tightly coupled sub-groups, withinwhich information is processed sequen-tially by a centralised software super-visor. This may result in much of thesystem being shut down by a single pointof failure, as well as contributing to planfragility and increased response over-heads.

The multi-agent system model seems anadequate framework to overcome suchproblems through the design and devel-opment of an application that is flexible,adaptable, versatile and robust enoughfor the efficient management of acontainer terminal.

It is critical that once in port, the turn-around of a cargo ship be as fast aspossible. An average cargo liner spends60% of its time in port, and a cost in theorder of US$1000 per hour is associatedwith this. Consequently, the containerallocation process must be directedtowards minimising cargo ship stowagetime. This is the main objective of theoptimisation of the global performanceallocation process.

Container terminal management is avery complex system, and traditionalsolutions decompose the problem intoseveral sub-problems, each representingone particular aspect. While the set of

operations to be conducted in theterminal is very extensive, existingapproaches share some commonfeatures:• the Marine Side Interface focuses on

the loading and unloading ofcontainers. Normally two or threegantry cranes (GC) are used to movecontainers for each ship

• the Transfer System transferscontainers between the apron and thecontainer storage yard. Yard trucks(YT) perform transports within theterminal. Transtainers are used to pickup or to put down a container on thestorage area of the yard (see Figure 1)

• the Container Storage System allo-cates and controls containers in theyard (see Figure 2)

• the Land Side Interface handles inter-actions with the land transportationmodes.

System ArchitectureFigure 3 shows the system architecture,in which we divide the problem into sub-problems, each of which is solved by aspecific agent. These agents are mainlycharacterised by their independencefrom the rest of the system elements.They are able to coordinate and tocommunicate decisions to the rest of thesystem. Communication between agentsis done by means of asynchronous

by Vicent J. Botti

Scientists at Universidad Politécnica de Valencia have been working on a solutionto a port container terminal management, specifically the automatic allocation ofcontainers. Given its inherent complexity, we propose that multi-agent systemsare the most suitable method for addressing this problem.

Multi-Agent System Technology in a Port Container Terminal Automation

SPECIAL THEME: MACHINE PERCEPTION

Figure 1: Yard map.

Figure 2: Transtainer.



messages, which are based upon theFIPA-ACL standard. The proposeddistributed approach enhances flexi-bility, efficiency and robustness. Fiveagent classes can be found in thissystem:• ship agentsdetermine the ships’

loading/unloading sequence • stevedore agentsmanage the ships’

loading/unloading process • service agentsdistribute the containers

in the port terminal• transtainer agentsoptimise the use of

these machines• gate agentsinteract with the land

transport (I/O of containers by land).

Ship AgentsIn response to the arrival of a ship (shipagent creation event), the system createsa new ship agent for this ship and its loadprofile. Its goals are to minimise gantrycrane idle time, the ship’sloading/unloading time, and the derivedcosts from the stowage process. Thiswork is closely related to that of thestevedore agents, with which the shipagent must coordinate.

Each ship agent faces a schedulingproblem in which a set of resources (thecranes) must be assigned to the differentoperations (loading/unloading ofcontainers), thereby establishing aresource use time (loading/unloadingtime). This requires that the various shipagents active at any time must coordinatewith each other to minimise clashesbetween assigned cranes.

Stevedore AgentsFor any given gantry crane, stevedoreagents try to obtain the most appropiatescheduling in order to manage containerstowage. The agent must know thegantry crane loading/unloadingsequence, which yard trucks are assignedto this crane and the positions of thevarious containers within the terminal(this information is provided by the rele-vant service agents). The agent thereforecoordinates with active ship agents andservice agents, and attempts to minimiseboth empty movements of the machineryemployed and the number of machinesnecessary for the internal transfer.

Service AgentsThe terminal is divided into services,with each being assigned specificstacking ranges. The main goal of aservice agent is to determine the appro-priate allocation for containers arrivingin the terminal from a specific service,and the most suitable configuration forthat portion of the yard controlled by theagent. To do this, the agent must knowthe yard map of its assigned portion, thecontainer characteristics (type, length,weight, destination port, ship) and thestacking factor. The agent must alsocoordinate with other service agents inorder to resolve conflicts, such as reallo-cation of containers where their assignedstacks are full.

In solving the configuration problem,service agents must maximise stackingdensity in their yard portions. The

service agent launches this process auto-matically when it considers it to benecessary, based on criteria such as time,stack allocation conflicts (slots withoutuse or cargo ships that run out ofreserved areas), low stacking density etc.

Transtainer AgentsEach transtainer is modelled as anautonomous agent whose goal is toperform container-stacking operationsefficiently. A transtainer agent mustminimise the transtainer’s empty move-ments. To do this, it obtains the mostefficient sequence for moving thecontainer to or from its correct positionin the yard. Each agent waits for stackingrequests from the various service agents,which inform the transtainer agent of thelocation of containers to be loaded tovessels or external trucks, or wherecontainers being unloaded from vesselsor trucks are to be placed.

Gate AgentsA gate agent controls the arrival ordeparture of a container by land. Theagent manages the assigned terminalgate, informing the correspondingservice agent of the arrival of newcontainers (in order to store them) and ofthe arrival of trucks (in order to retirecontainers from the yard). For instance,when a container arrives, the gate agentchecks the accuracy of its data, and if thedata is correct, asks for a container loca-tion from the service agent responsiblefor the service to which the container isassigned. Once the location is known,

Figure 4: System Interactions.Figure 3: System Architecture.



Complexity of modern manufacturingand process industries emphasises theimportance of the fault detection andisolation for the purpose of processcondition monitoring. Unlike occasionalinspections, periodic checks, etc, on-linecondition monitoring seems to be themost promising way to decrease thenumber of unanticipated system shut-downs while optimising the productioncosts. Extension of recently developedprobabilistic mixture-based approach tothe domain of systems' fault detectionand isolation provides an adequatesupport to solution of process moni-toring problem. Performance of theapproach was verified on a real processof gas conditioning.

Within an EC IST-99-12058 projectProDaCTool, a powerful generic tool fordescribing and optimising complexuncertain multi-dimensional systems bymultiple models was developed andsuccessfully applied. Hypothesis onexistence of several operational regimes- normal as well as faulty - makes multi-model approach a natural framework forsolving the fault detection and isolation(FDI) problems. This has motivated us totest its applicability to FDI domain.

The whole framework relies on a proba-bilistic mixture modelling that providessuitable description of non-lineardynamic systems. Each of the operationmodes as well as each of the faulty statesis represented via probabilistic model. Aconvex probability density functions ˆcalled components - describing respec-tive normal dynamic regressions, eachassociated within some operation mode,models the process data. Quasi-Bayesestimation provides the model of thesupervised process. Provided richprocess data exist, the model learnedrepresents all the operation statesoccurred in the past, each associatedwith the corresponding component. Theprobability of occurrence of a particularoperational state is expressed by theweight of the respective component.During on-line operation, the learnedcomponents' weights are updated usingavailable process measurements andexploited for effective estimation of theactual operation state. A proper recom-mendation is then generated in a timelyfashion. Effectiveness of the outlinedapproach has been verified on a gas-conditioning unit, a part of an industrial-scale pilot installation in Josef StefanInstitute for treatment of technological

wastewater. Real-time diagnostic resultsshow that each component correctlyrepresents faulty states of the system.Apart from short transient periods,during changes of process states, allfaults were successfully detected anduniquely isolated. It is important to notethat the presented diagnostic approachrelies on availability of rich past data.These should also include all faultystates that have to be diagnosed.Unfortunately this is not always the casein real applications, especially if a diag-nosis system is designed from scratch.The on-line applicability of the quasi-Bayes estimation allows to design adap-tive FDI version. The available self-learning capability will enable toconsider newly met faulty states.

Please contact: Tatiana V. Guy, Institute Information Theory and Automation, Academy of Sciences of the Czech Republic /CRCIMTel: +420 2 6605 2254 E-mail: [email protected]

by Tatiana V.Guy, Dani Juri i , Miroslav Kárny and Andrej Rakar

A multi-model approach complemented by probabilistic mixture modelling hasbeen applied to solve fault detection and isolation problem. The resulting algorithmprovides efficient tool for process condition monitoring.

Contribution to Fault Detection and Diagnosis: Mixture-based Modelling

this is communicated to the truck, whichdelivers it to the appropiate stack. Asimilar process occurs for containersleaving the terminal.

ImplementationInternal and external interactions(between the various agents and withsystem users respectively) are shown inFigure 4. FIPA standards have beenfollowed as far as possible in imple-

menting this prototype. In doing so, a setof auxiliary agents has been created: • wrapper agents provide access to the

database and communicate withexternal software

• the UDMA (User DialogueManagement Agent) is an interfaceagent with human users

• the UPA (User Personalisation Agent)manages the explicit profiles and pref-erences of registered human users.

There is also an agent platform (AP) tosupport the entire architecture. Itprovides the following services: yellowpages, white pages, a communicationchannel and an agent platform securitymanager.

Please contact: Vicent J. Botti, Universidad Politécnica de Valencia/SpaRCIME-mail: [email protected]



Coordinated by ERCIM, TELEMAC is aproject within the European ISTprogram that is developing innovativeapproaches to managing anaerobicwastewater treatment plants, with partic-ular application to the wine industry (seealso ERCIM News No. 48). One of thefeatures of the project is the integrationof a variety of approaches, including softsensors, fault detection and isolation, andremote monitoring and access ofmultiple plants. The project has nowbeen running for two years and has madesignificant progress.

Within TELEMAC, CCLRC has beenworking specifically on data mining.This is an important approach, sincedata, such as pH, temperature, and moreadvanced measurements such as volatilefatty acids (VFA), is constantly beingaccumulated. Data mining opens up theprospect of learning from this data inorder to manage plants better. A numberof possibilities are being studied:• developing models or rules that help to

predict dangerous conditions on theplant from trends in sensor readings

• detecting faulty sensors throughinconsistent sets of readings

• partially substituting for expensivesensors by combining readings frommore commonly available sensors.

Within the TELEMAC project, data isavailable from a number of plants ofdiffering types and sizes. These rangefrom large industrial waste-processingplants, through pilot-scale plants with afull range of instrumentation, to smalllaboratory-scale set-ups used for runningspecific experiments.

Data mining is often regarded as one partof the broader problem of knowledgediscovery. Knowledge Discovery inDatabases (KDD) is defined as ‘a non-

trivial process of identifying valid, novel,potentially useful, and ultimately under-standable patterns in data’ and datamining as ‘exploration and analysis, byautomatic or semi-automatic means, oflarge quantities of data in order todiscover meaningful patterns’ (Kumar, V.& Joshi, M. [1999], Tutorial on high-performance data mining, at http://www-users.cs.umn.edu/~mjoshi/hpdmtut/, Uni-versity of Minnesota). Although much ofthe activity is data mining, the goal isusually knowledge discovery. Datamining is not simply a matter of runningalgorithms for rule induction or neuralnetworks; a considerable amount ofpreliminary work is usually required, andTELEMAC is no exception. The overallprocess includes stages such as data selec-tion, data cleaning (for example, dealingwith missing values or outliers), datareduction or enrichment, data preparation,the data mining itself, and reporting(through visualisation, statistics etc).

The software being used at CCLRC fordata mining is Clementine, a productavailable from SPSS. Clementine wasdeveloped in Europe as a commercial,general-purpose data mining tool andwas adopted by the Business andInformation Technology Department of

CCLRC in 1998 for scientific applica-tions such as the EC-funded projectDECAIR (http://www-air.inria.fr/decair/). Like a number of visualisationsystems (for example AVS, Nag IrisExplorer, IBM Open DX), it presents avisual interface that allows a user toconnect modules together, allowing datato flow from one end to the other.Visualisation software such as XMDV isalso being used.

Initial work focussed on the use of datamining to recover a known model fromsynthetic data, thereby giving prima facieassurance of the applicability of theapproach. Fitting was generated using asimultaneous prediction of three keyvariables. A Pruned Neural Netmodelling tool was employed. The datawas split into a test set and a training/vali-dation set. The training/validation set israndomly split 50/50. The reason for thisthree-way split is to avoid over-trainingthe neural net — a common difficultyfaced in practice. It has proved possibleto reproduce the synthetic data to a goodlevel of accuracy.

Exploratory work has also been done onconfidence and prediction intervals forTELEMAC data. This refers to thedistinction between the accuracy of thefit itself (the regression) and in thepredictions arising from its use. Theseare different, since an individual valuewill have extra variability due to noise.

Simple modelling of target variables forsome of the industrial data is also underway. The objective is to determinewhether a neural net model constructedin one time period could be used to makepredictions of a target variable in anothertime period. The technique used is theExtended Pruning approach with ameasurement of VFA as the target, and it

by Simon Lambert

The research and development project TELEMAC is creating an adaptable,customisable system for the monitoring and control of anaerobic wastewatertreatment plants. One of the techniques being employed is data mining — theextraction of useful knowledge from data using a variety of techniques.

Data Mining Applied in Anaerobic Wastewater Treatment

Anaerobic digester at industrial scale

(Sauza, Mexico).



As fermentation processes are respon-sible for the production of high valueadded substances such as enzymes,antibiotics and vitamins, they have greatimportance in the biotechnologyindustry. From the control engineer’spoint of view, the fed-batch processespresent the greatest challenge. At thesame time, there is an enormouseconomic incentive behind the work oncontrol, optimisation and conditionmonitoring of fed-batch fermentationprocesses.

To achieve high performance operation,the optimisation of the primary factorsinfluencing fed-batch fermentationprocesses is an important task. The

ability of optimal control techniques toassist in this optimisation has been inves-tigated. Optimal control techniques relyupon an accurate model of the processand for many years mechanistic modelshave been used to develop optimalcontrol strategies for fed-batchprocesses. However, mechanisticmodels of fed-batch processes areusually very difficult to develop due tothe complexity and nonlinear nature ofthe processes. An attractive alternative todeveloping a mechanistic model of aprocess is to build an empirical modelusing input-output data collected fromthe process. Among all the existingmethods, the multivariate statisticalprocess control techniques, such as PCA

and PLS have been proved to be verypromising approaches for application tofed-batch fermentation systems. Multi-way Partial Least Squares (MPLS) isused to identify empirical models forfed-batch fermentation processes. Theobjective function of the optimal controlproblem for a batch or fed-batch processneeds to reflect the performance of thefermenter. Maximisation of yield, selec-tivity or conversion and minimisation ofbatch cycle time are some examples ofpossible objectives. As the optimalcontrol of fed-batch fermentationprocesses is a challenging dynamic opti-misation problem, it is usually difficultto solve because of the nonlinear systemdynamics and the constraints on the

by Hongwei Zhang

Multivariate Statistical Process Control is seen by many as the key to real-timeprocess monitoring which results in significant improvements in operationalsafety and efficiency. Members of the Control Engineering Research Group (CERG)at Manchester School of Engineering are developing MSPC technologies, includingPrincipal Component Analysis (PCA), Partial Least Squares (PLS) and KernelDensity Estimation (KDE), and combining them to form an integrated robustprocess control and monitoring package. Recently, research has been carriedout at CERG with particular interests in the applications to fed-batch fermentationprocesses.

Integrated Multivariate Statistical Process Control and Condition Monitoring for Fed-batch Fermentation Processes

gave 84% accuracy when applied to datafrom another time period from a partic-ular plant.

Techniques of rule induction are beingapplied to estimate values of sensor read-ings based on more easily obtained values,and to determine how reliable the modelsso developed remain over time. Rules aregenerated in forms such as the following:

Variable X falls in a particularrange of high values if variable Yfalls within a particular range of lowvalues – this rule scores (N, p).

Here N and p are indications of thedegree of satisfaction of the rule. Ruleinduction has the advantage that the rulesgenerated are often meaningful to human

domain experts, and can be critiqued andvalidated by them.

A key question for future work is thehandling of the evolution of the state ofthe plant over time. This will begin withthe approach of Dorffner to neuralnetworks for time series processing, inwhich a model is built up from trainingrecords that contain both prior timevalues and current values. This tech-nique is often known as lagging.

A further important question is the place ofdata mining in the final TELEMACsystem. The project as a whole is nowconsidering the form of the finalTELEMAC system and its tools, but at thetime of writing it seems that data miningwill operate as a component at the

Telecontrol Centre (responsible for moni-toring several plants), and will run periodi-cally to update the knowledge base. Itremains an open question as to how todetect when it is necessary to re-run the datamining algorithms for a particular plant.

Link: http://www.ercim.org/telemac/

Please contact:Simon Lambert, CCLRC Responsible for data mining workTel: +44 1235 445716E-mail: [email protected]

Bruno Le Dantec, ERCIM Administrative coordinatorTel: +33 4 9238 5013E-mail: [email protected]

Olivier Bernard, INRIA Scientific coordinatorTel: +33 4 9238 7785E-mail: [email protected]



control and state variables. Typicalchoices for the objective function in theoptimisation of a fed-batch fermentationprocess include the maximisation ofeither biomass or metabolite productionwith respect to the substrate feed rate.Based on the techniques of MPLS andFGA, members at CERG have devel-oped optimal control algorithms that aresuitable for application to batchprocesses. The algorithm utilise linearmodels of the process that are identifiedusing the multi-way partial least squarestechnique. The linear models are usedwithin optimal control law to regulatethe productivity of the batch by manipu-lating the substrates that are fed into the

fermentation vessel. The proposedcontrol algorithms are successfullyapplied to different fed-batch fermenta-tion process and their performancescompared with alternative control algo-rithms that have been applied to theapplications in the past. The proposedcontrollers are found to comparefavourably with these alternative algo-rithms.

As the objective of control systems forbatch and fed-batch processes is alwaysto ensure high quality of the products. Afurther issue in the successful operationof fed-batch fermentation processes isthe early and accurate detection of fault

conditions, such as sensor failures ordrifts that may occur. The early detectionof fault conditions is of great benefit infed-batch fermentation processes sincethe earlier that a fault can be detected andacted upon, the lower its impact will beon the process. In some situations thiscan be critical, for example, a drift on apH sensor could have catastrophicresults on biomass growth if thismeasurement is used within a feedbackcontrol scheme. PCA and PLS have alsobeen seen as very promising approachesfor fault detection and isolation. Inprevious studies condition monitoringand control of fed-batch fermentationprocesses have been viewed as indepen-dent problems. In the research carriedout at CERG, these problems are consid-ered together and integrated fault detec-tion and process control schemes arebeing developed. These schemes relyheavily on the successful developmentof a PLS model to provide soft-sensingcapabilities in a fed-batch fermentationprocesses. In research implemented atCERG, it is demonstrated throughseveral different case studies that PLSmodels can be used, not only in a predic-tive control framework to regulate thegrowth of biomass within the fermenter,but to also provide fault detection andisolation capabilities. It is further shownfrom simulations that the integration ofthe predictive controller and fault detec-tion capabilities provides a useful diag-nostic tool for the control system.

This research so far has seen some verygood results in various simulations,efforts now are being made to verifythese techniques through lab experi-ments. Future research also includes theintegration of artificial intelligencemethods such as expert systems and themultivariate statistical process controlmethods. This project is funded byEPSRC (Grant Number GR/N24858).

Links: http://travis.eng.man.ac.uk/http://www.newi.ac.uk/zhangh/

Please contact: Barry Lennox, University of Manchester, UKTel: +44 161 275 4324E-mail: [email protected]

Hongwei Zhang, Northeast Wales Institute of Higher Education, UK Tel: +44 1978 293 392E-mail: [email protected]

Figure 1:

Data Unfolding.

Figure 2: Detection of a pH sensor fault and pH measurement inferred by the PLS model.

Figure 3:

Integrated Process

Control and

Condition

Monitoring.



To meet a demand that oscillatesbetween 40 and 60 gigawatts, EdF oper-ates about 150 thermal units (nuclear andconventional) and 50 hydro-valleys.Since electricity can hardly be stored,production must closely follow demand.Changing the operation mode of a unithowever, especially a thermal one, isdifficult and costly. As a result, opti-mising the schedules is a criticalproblem, involving large amounts ofmoney. It is also a tough nut to crack,and is often referred to as the unit-commitment problem: large-scale,mixed-integer, heterogeneous. Itscomplexity can be illustrated by a super-ficial description of the production units.

A thermal unit must produce a certainminimum power (positive). This intro-duces a combinatorial aspect into itsoperation - the unit is ‘on’ or ‘off’. Thisis not the only one, however: whenswitched off or on, the unit must respec-tively stay off or maintain its productionlevel, for a minimum period of time.Also, when switching on, the unit mustfollow specific start-up curves thatdepend, among other things, on howlong the unit has been off. Many similarconstraints exist, and we will notenumerate them here. Thermal plantscan be suitably modelled such that theresulting optimisation problem would betractable if there were only one unit tomeet the demand – but this is certainlynot the case with 150.

A hydro-valley is a set of interconnectedreservoirs and associated power plants,each of which has a certain number ofturbines. Some plants are equipped topump water from their downstreamreservoir into their upstream reservoir(thus allowing this water to be reusedlater during peak hours), but a plantcannot simultaneously discharge(producing energy) and pump(consuming energy). Using convenientsimplifying assumptions, hydro-valleys

can be modelled as an ordinary linearprogramming problem, which is onceagain easy to solve with one valley, butnot with fifty.

One way of tackling the unit-commit-ment problem is price decomposition:the constraint of meeting demand isreplaced by a ‘dummy benefit’rewarding each megawatt produced by aunit (the marginal benefit is the same forall units). This replacement, calledLagrangian relaxation, allows a separateoptimisation of the 150+50 productionunits. As a result, we are faced with acoordination problem, which is to findoptimal dummy prices. Since October2002, EdF has been using the so-calledbundle method, which was invented atINRIA in the mid-seventies, and whichis particularly efficient for this sort ofproblem.

The question then arises: can it be guar-anteed that coordination produces equi-librium? The answer is a definite ‘no’ -

the schedules computed by Lagrangianrelaxation are far from meeting demandwithin acceptable tolerances. While acertain combination of them does meetit, this does not produce a technicallyfeasible schedule.

To balance these two opposite outputs, aheuristic technique has been developed,also at INRIA, and is now being industri-ally evaluated by EdF. Experiments onreal datasets indicate that substantialgains are obtained, as compared to theimplementation currently in operation atEdF. In fact, the coordination phasedelivers not only optimal dummy prices,but also a lower bound on the minimumpossible cost of any feasible schedule;this is a well-known property ofLagrangian relaxation. Usually, thefeasible schedules produced by thecurrent implementation cost about 1%more than this lower bound. The newapproach typically cuts this excess by ahalf.

However, the main advantage of the newapproach is its robustness, which can beillustrated as follows. On a particular dayin 2002, an uncommon peak load of70GW occurred. The schedule obtainedby the current implementation producedan intolerable mismatch of 1.3GW; incontrast, the new heuristic techniquemanaged to find a schedule that cost lessand, more importantly, met the demandto within 50MW - quite a reasonabletolerance.

Please contact:Claude Lemaréchal, INRIATel: +33 04 76 61 52 02E-mail: [email protected]

After fifteen years of collaboration, EdF, the French Electricity Board and INRIAare finalising an efficient software which will optimise the daily production ofelectricity in France.

Saving Energyby Claude Lemaréchal

Barrage de Chaudanne, a hydro-valley

operated by EDF, the French electicity

board.

Sou

rce:

ww

w.s

truc

tura

e.ne

t; P

hoto

by

Dr.

Kla

us J

anbe

rg

ERCIM News No. 56, January 2004


Articles in this Section

45 Solving Complex Problems with AdvancedTechniquesby Enrique Alba, University of Malaga/SpaRCIM

47 Nipper - the Neat Internet Protocol Packet EditorSami Pönkänen and Marko Lyijynen, VTT

48 On Describing B2B Processes with Semantic WebTechnologiesby Santtu Toivonen, Tapio Pitkäranta, VTT and Jung UngMin, Stanford University, USA

50 Tracking Assets and Vulnerabilities in CorporateNetworks by Dieter Gantenbein, IBM Zurich Research Lab/SARIT

51 Numerical Simulation through Web Servicesby Stephan Springstubbe, Jürgen Klein and Ottmar Krämer-Fuhrmann, Fraunhofer SCAI

53 Advanced Communication Networksby Michel Mandjes, CWI

54 Information Security Research at NTNU by Stig F. Mjølsnes and Ingvild Ytrehus, NTNU

55 Supervised Term Weighting for Automated TextCategorizationby Franca Debole and Fabrizio Sebastiani, ISTI-CNR

56 Remote Monitoring of Health Conditionsby Dirk Husemann, IBM Zurich Research Lab/SARIT



Complex problems – those that can onlybe solved in non-polynomial time – arebecoming common in all the domains ofour lives: telecommunications,economics, bio-informatics, industrialenvironments etc. In these and otherfields of research it is often essential tomodel and solve optimisation, learning,or searching tasks for applications thatdo not admit an easy formulation. In fact,it is often the case that the problem athand is non-differentiable, has a largenumber of constraints or objectives, doesnot admit contour conditions, or is ill-defined. Also, most real-life problemsshow a high degree of inter-parameterlinkage (epistasis), many locally optimalsolutions (multimodality), and present ahigh dimension.

All these ideas were included in therecently completed MALLBA project(2000-2003), and are being extended inTRACER (2003-2005), a new set ofproposals for dealing with complexproblems. The MALLBA project was aneffort to develop an integrated library ofskeletons for optimisation containingexact, heuristic, and hybrid techniques.One important outcome of the projectwas the MALLBA library, which ispublicly available at http://neo.lcc.uma.es/mallba/easy-mallba. The threetarget environments considered forsolving complex problems are sequen-tial, LAN, and WAN computer plat-forms. The list of modern techniques weconsider in our studies with MALLBAand TRACER includes evolutionaryalgorithms, simulated annealing,memetic algorithms, ant-colony algo-rithms, scatter search, and a set of varia-tions on well-known techniques likebranch-and-bound, randomised localsearch, etc.

Software skeletons are somewhat similarto simplified software patterns. In

MALLBA, software skeletons for allthese algorithms have been built with acommon internal and public interface.This permits fast prototyping and trans-parent access to parallel platforms.MALLBA skeletons distinguishbetween the concrete problem to besolved and the solver technique.Skeletons are generic templates to beinstantiated by the user with the featuresof the problem. Much work has been putinto designing reliable, general algo-rithms, and a permanent interest exists inusing software engineering concepts for

correct design. All the knowledge relatedto the solver method (eg parallel consid-erations) and its interactions with theproblem are implemented by theskeleton and offered to the user.

Skeletons are implemented by a set ofrequired and provided C++ classes thatrepresent an abstraction of the entitiesparticipating in the solver method.Provided classes implement internalaspects of the skeleton in a problem-independent way. The most importantprovided classes are 'Solver' (the algo-rithm) and 'SetUpParams' (configuration

parameters). On the other hand, requiredclasses specify information related to theproblem. Each skeleton includes the'Problem' and 'Solution' required classesthat encapsulate the problem-dependententities needed by the solver method.Depending on the skeleton, other classesmay be required.

Therefore, the user of a MALLBAskeleton need only implement the partic-ular features related to the problem. Thisspeeds up considerably the creation ofnew algorithms with minimum effort,

especially if they arebuilt up as combina-tions of existingskeletons (hybrids).Figure 1 shows anexample of design fora simulated annealingalgorithm.

The focus ofMALLBA wasmainly in creating thesoftware and hard-ware infrastructure(especially an opera-tive WAN) for laterdevelopments, whileaccounting for paral-lelism and hybridisa-

tion. Many application domains havebeen addressed within MALLBA, but atthe same time, many important issuesconcerning the problems and the tech-niques themselves were deferred forTRACER.

In the new TRACER project(http://tracer.lcc.uma.es) therefore,powerful and robust algorithms arebeing developed in close connectionwith the Internet. Again, in all ourresearch we pay close attention to effi-ciency and to the software design aspectsof the algorithms. This project represents

by Enrique Alba

The area of complex problems and modern solving techniques is one of themultiple research lines in progress at the University of Málaga. It has beenaddressed in the research projects ‘MALLBA’ and ‘TRACER’.

Solving Complex Problems with Advanced Techniques

Figure 1: UML design of a simulated annealing heuristic in

MALLBA.

Stop_Condition_1

Move

Default_Move

SetUpParams User_Statistics

State_Vble

State_Center

Problem

Solution

NetStream

Solver_SeqSolver_Lan Solver_Wan

Statistics

Stop_Condition

<<abstract>>

Solver



a joint effort of five groups in Spain,coordinated from the University ofMálaga and devoted to proposing solu-tions and algorithms (in C++ and also inJava) that outperform existing cutting-edge optimisation techniques. We areworking on two fronts: achieving resultshaving the best-so-far accuracy for theconsidered problems, and developingalgorithms for these that show the best-so-far numerical efficiency. At the endof the project we expect to provide theresearch community with leading algo-rithms and accurate results, therebyenhancing scientific knowledge on opti-misation. Some successful case studiesinclude the following:• Computing optimal routes for the

vehicle-routing problem (vehiculartechnology). VRP is an important NP-hard problem for which we are able todesign minimum cost routes for a fleetof vehicles with a cellular evolutionaryalgorithm. In this algorithm, tentativesolutions evolve in overlapped neigh-bourhoods connected in a 2-D gridtopology. Using local searching ineach iteration (specifically 2-opt plusλ–interchange algorithms) has beenthe key step in outperforming otheralgorithms in wall-clock time andaccuracy. Real-life instances takenfrom the city of Málaga have beensolved to optimality quite efficiently(see Figure 2 left). In Figure 2 (centre)we show a partial solution covering85.6% of the area. The colour codesindicate whether a given location iscovered by 0, 1, 2, 3 or 4 antennae. Anoptimum solution will have all the area(100%) covered by exactly one singleantenna (colour code 1).

• Designing optimal error-correctingcodes (information theory) showingthe highest minimum Hamming

distance between codewords (see anexample in Figure 2, right). For thisproblem, distributed evolutionaryalgorithms and a new local searchalgorithm inspired by the repulsion ofcharged particles have been used. Thenumerical effort of our algorithmsmakes it possible for the first time toobtain solutions for large instances ofthis problem, in which at present theoptimum can only be said to belong toa certain range of values.

• The optimal location of antennae(telecommunications) is another hottopic of research, since UMTS,mobile, and wireless platforms need anew set of networks for their develop-ment over the whole of Europe. Wehave proposed distributed algorithmsable to compute solutions covering100% of the area with a minimum setof strategically located antennae.While in the past this problem requiredaround forty machines working formany hours to reach a solution, ouralgorithms can now provide us withoptimum solutions in a few minutesusing between one and eight CPUs. Infact, we are extending these algo-rithms to other hard problemssuggested by well-known telecommu-nication companies in Europe, inwhich a physical model of wave prop-agation plus detailed information ofthe geographical area is included.

• Optimisation in continuous domains isan additional line in our research. Thiskind of problem frequently requiresvery different algorithms to thosefound in combinatorial optimisation.We have recently proposed a parallelhypercubic evolutionary algorithmdealing with heterogeneous operationsincluding concepts from fuzzy logic tosolve continuous problems: systems of

linear equations, computing filters forsound waves, polynomial fitting(Chebyshev), and training neuralnetworks among others.

• The new and exciting field of gridcomputing is also being addressed tosolve some of the most difficult prob-lems. Condor and Globus are two ofthe most successful technologies wehave been using, just as XML andSOAP have been studied for somecases. Concretely, we have developeda new grid algorithm to compute theexact Pareto front of several hardmulti-objective problems by usingCondor in a grid with more than 100computers in our department(http://neo.lcc.uma.es/Software/ESaM).At present, we are extending the gridto other sites in Spain and Europe.

We are also working on other topics likebio-informatics or dynamical optimisa-tion (DOP). However, besides workingon the algorithmic aspect, we are aware ofcommunities of users demanding solu-tions for their own problems. This is whywe plan to release a public Internetservice that will help researchers to solvecomplex problems by feeding them(written in Java, C++, and otherlanguages) into a client/server system andthen selecting one solver from an assortedset of algorithms to obtain a solution witha short round-trip time and low effort.Any kind of comment, idea, or collabora-tion is welcome.

Link: http://neo.lcc.uma.esPlease contact:Enrique Alba, University of Malaga/SpaRCIMTel: +34 952 132803E-mail: [email protected]

Figure 2: Example

of a real VRP in

Málaga, location of

antennae (left,

centre) and error-

correcting codes.



The need for a tool like Nipper came upin a networking project when we startedtesting and debugging our software.Numerous tools exist for networktesting; those we use most frequentlyinclude tcpdump and ethereal forsniffing, ping for connectivity checkingand nmap and nessus for scanning ofsecurity holes. However, we were unableto find an elegant, simple tool that couldcreate arbitrary IP packets and send themto network or directly to applications.

Instead of coding a quick and dirty, use-and-throw-away type program, wedecided to design and implement acustomisable tool that would have asimple, user-friendly GUI. In the firstplace, it would be used quite extensivelyin the project, and therefore needed to beuser-friendly. Further, we also wanted tobe able to use it in other projects,meaning it needed to be customisableand extensible, and if possible, able torun on OSes other than Linux (althoughLinux is what we use most frequently).Java was therefore the natural languageto choose.

Nipper turned out to be very usable, andwe thought it might prove popular as astarting point for building other tools.Consequently we decided to go opensource with Nipper. Nipper was releasedunder VTT Public License.

Nipper's Architecture and FeaturesNipper has three main components:PacketEditor is the GUI, PacketEngine iswhere all the real work is done andNetwork Interfaces are the componentsthat deliver the packets. See Figure 1 fora diagram of the Nipper internals. Thereason for a three-layer architecture iscustomisability. We wanted the ability toswitch between Network Interfaces, toadd new protocols to PacketEngine andmaybe someday to use one GUI to

control many PacketEngines on differenthosts.

PacketEditor takes its input from theuser. The user can create new packets,add protocol headers and payload tothem and fill in values for differentheader fields. This input is stored in anXML format description that can besaved and loaded from a file. Anexample of an XML description isshown in Figure 2.

PacketEditor passes the XML descrip-tion to PacketEngine where the realaction happens. PacketEngine builds anumber of real packets from the XMLinput. Inside PacketEngine, differentprotocols are represented by Java classes

that perform the actual conversion ofXML parameters to byte-format protocolheaders. Each protocol class has a set ofrational default values for fields whichthe user has left blank. Tasks like check-summing and calculation of length fieldsare also done within the protocol classes.Nipper currently includes basic supportfor IPv4, IPv6, ICMPv4, TCP and UDPprotocols and payload data. Partialimplementation of Mobile IPv4 is alsoincluded.

After the packets are built, PacketEnginestarts sending them one by one. Thishappens by passing the byte array to aNetwork Interface component. Atpresent there are two Network Interfacesto choose from: PacketDumper andPacketSender. As the name implies,PacketDumper simply dumps the byte-format packets to a debug window.PacketSender uses the Java NativeInterface (JNI) and Linux raw sockets tosend the packets to network. CurrentlyPacketSender is able to send both IPv4and IPv6 packets.

Using NipperSo let's take Nipper for a test drive. As anexample we show you how to build andsend a packet with IPv4 and UDPheaders and a string 'hello world!' aspayload. Note that using raw sockets (thePacketSender) requires root privileges.

Figure 3 shows the main window ofNipper. In the lower part of the windowis the set of packets with one packetvisible. Select the packet and pick IPv4from the tabs in the top part of thewindow. Fill in values for different IPv4fields or leave some fields blank and letNipper use default values. In Figure 2 wehave set the source and destinationaddresses and the TTL field. Click on'Add to packet' to add the IPv4 header tothe packet you selected earlier. Do the

Nipper – the Neat Internet Protocol Packet EditoR – is a tool for testing networksand network applications. More specifically it is used to create arbitrary packetsand to send them to network or directly to applications. Nipper is written in Javaand uses portions of C-code and Linux raw sockets.

Nipper – the Neat Internet Protocol Packet Editorby Sami Pönkänen and Marko Lyijynen

Figure 1: Nipper architecture.

Figure 2:

XML description of a HTTP request.

<?xml version = '1.0' encoding = 'UTF-8' ?><packetset><packet stime="1"><header type="IPv4"><srcaddr>10.0.0.1</srcaddr><dstaddr>10.0.0.2</dstaddr>

</header><header type="TCP"><dstport>80</dstport>

</header><header type="Payload">

<data>GET http://www.vtt.f iHTTP/1.0</data></header>

</packet></packetset>



same for UDP and Payload (under the'Other' tab).

Next you need to select a NetworkInterface. Go to the 'PacketEngine' menuand 'Set network interface'. AvailableNetwork Interfaces are listed on thedialogue box shown in Figure 4. Select'RawSocket'.

Now you are ready to send the packet tonetwork. Open a shell and start tcpdumpif you want to see whether the packetcauses any reaction. Then click on 'Sendpacket(s)'. A debug window opens upand if everything goes as planned you

should see some activity in your tcpdumpand a report as shown in Figure 5.

The verbosity of debug messages can beadjusted via the 'PacketEngine' menu.For more information about usingNipper, consult the Nipper User's Guide,which is available at the open source siteof VTT.

Developing NipperAdding new protocols to Nipper is quitestraightforward. We suggest taking acloser look at one of the existing proto-cols and reading the files under doc/ inthe Nipper package. The source codeincludes JavaDoc documentation.

Network Interfaces can also easily beadded. One could for instance create aNetwork Interface that uses the tun/tapdevice instead of raw sockets. Again,take a look at existing source code andthe documentation bundled with Nipper!

Link: http://opensource.erve.vtt.fi

Please contact: Marko Lyijynen, VTTTel: +358 9 456 4311E-mail: [email protected]

Sami Pönkänen, VTTTel: +358 9 456 7232E-mail: [email protected]

Figure 3: Nipper mainwindow. Figure 4: Network Interface selection. Figure 5: Nipper debug window.

Two emerging trends are reshaping theWeb as we know it today. One of these isthe Semantic Web, in which content isconsumed by computer programs inaddition to humans. The other is WebServices, which provides a set of stan-dards and practices for the remote use ofcomputer applications. The combinationof these two is sometimes referred to asIntelligent Web Services. The PAXproject concentrates on this within thecontext of two specific B2B cases,namely service chain composition and

expediting process. More specifically,the cases are approached in a ‘bottom-up’ manner, meaning that serviceproviders have a degree of freedom indescribing and executing their services.This in turn means that some level ofintelligence or adaptability is expectedfrom the service consumers, that is, theservice composer and the expeditor.

Web Services carry on the tradition ofdistributed computing. Individualservices are scattered over the Internet

and are used remotely. Web Services aremodular and composite by nature,meaning that simpler services can begrouped together to form more complexones. It is necessary that the servicecomposer be able both to find theservices and also to evaluate them beforethe composition. Service discovery iseasier to achieve, since Web Services areaddressable with URIs, which can bestored in a common repository such asUDDI. Service evaluation is not sostraightforward, because it presupposes

The PAX project, carried out by VTT Information Technology in cooperation withthe Center for Integrated Facility Engineering (CIFE) at Stanford University,concentrates on Intelligent Web Services within the context of two specific B2Bcases, namely service chain composition and expediting process.

On Describing B2B Processes with Semantic Web Technologiesby Santtu Toivonen, Tapio Pitkäranta and Jung Ung Min



descriptions of the services in acommonly agreed-upon fashion. Thisproblem is especially evident if theservice composer is a computer program.Service descriptions must be serialised ina machine-understandable format, andconform to a common ontology.

The richness of service descriptions canvary from plain service-type categorisa-tion to detailed descriptions aboutservice characteristics. The PAX projectconcentrates especially on process-related service descriptions. Services aredescribed using Semantic Web technolo-gies such as RDF and OWL. Figure 1depicts a complex Web Service, which iscomposed based on the process descrip-tions of the individual services.

In the PAX project, the service composi-tion task is approached in terms of aspecific scenario from the constructionindustry. In this scenario there is asubcontractor (SC), sub-subcontractors(SSCs), banks, and material suppliers.The scenario also contains three kinds of

processes: informational (such asrequests for quotes), material (such asproduct deliveries), and financial (moneyflow). Some of the processes are one-to-one (eg product deliveries) whereasothers are one-to-many (eg biddings forSSCs and banks). The scenario can alsobe configured, based for example onwhether the project is on or behindschedule. For instance, the servicecomposer, in this case the SC, can choosethe more expensive but faster SSC if theproject is behind schedule – somethingthat it would not do if the project wasrunning on time.

Apart from this type of service composi-tion task, the adaptability of softwareagents is also investigated. More specifi-cally, agents adapt to interaction protocoldescriptions if unaware of the utilisedprotocols beforehand. Interaction proto-cols are typically used when modellingsoftware agent conversations. They tooare described using Semantic Web tech-nologies, and conform to a commoninteraction protocol ontology.

The specific case chosen for studying inter-action protocol-based adaptability is anexpediting process, depicted in Figure 2.The general contractor (GC) expectsstatus reports from SCs regularly.Suppose it has received one from someSC, containing information about animportant steel delivery that has not yetarrived. GC designates an expeditor tocontact other process participantsdirectly in order to verify the status of thesteel delivery. Traditionally this meanspaying visits to the shop floors or makingphone calls. Instead, expediting is nowapproached with novel technologies,namely software agents representing theprocess participants and Semantic Weblanguages for describing the utilisedinteraction protocols.

From SC’s status report, GC is able toextract the steel order ID as well ascontact information for the steel supplier(SS) and an external carrier (CR). Withthe order ID, it asks SS for the deliveryID of that order. Note that GC is initiallyunaware of the interaction protocolsprovided by SS and CR. Rather, it isaware of a common interaction protocolontology according to which the indi-vidual protocol descriptions aredesigned. With that knowledge, it is ableto adapt its behaviour to correspond toSS’s interaction protocol.

With the delivery ID provided by SS, GCthen contacts CR and requests a deliveryreport. Like SS, CR also offers descrip-tions of the interaction protocols it usesto communicate, and GC then adapts itsbehaviour to conform to those. Afterreceiving the delivery report, GC cancompare the information in the statusreport and the delivery report and takeaction if needed. The comparison andpossible resulting actions can beperformed either by the software agentitself or, more realistically, by a humanbeing.

Link: http://www.vtt.fi/tte/proj/pax/

Please contact: Santtu Toivonen, Tapio Pitkäranta VTTE-mail: [email protected]

Jung Ung Min, Stanford University, USAE-mail: [email protected]

Figure 1:

Composing

and executing

a complex

Web Service.

Figure 2:

Interaction

protocol-

based

expediting

process.

Request Status ReportRequest Status Report[failure]

Extract Carrier URIExtract Carrier URI

Extract Supplier URIExtract Supplier URI

Extract Order IDExtract Order ID

Query for Delivery IDQuery for Delivery ID

[success]

[success]

[failure]

Request Delivery ReportRequest Delivery Report

Compare InformationCompare Information

[success]

[failure]

Take ActionsTake Actions

[not ok] [ok]

GC to SC

GC to SS

GC to CR

GC

GC

GC

GC

GC



Tracking computing devices as assetswith usage, health, and vulnerabilityinformation facilitates the provision andmaintenance of an efficient, optimizedservice. Recent incidents with virusesand worms appear to support this. Ingeneral, a precise understanding of theoperational infrastructure is a keyelement of many corporate decisions.Examples also include the negotiation ofoutsourcing contracts, the planning ofmergers and acquisitions, server consoli-dation and business optimization.

Building an accurate inventory ofcomputing assets in heterogeneousdynamic systems and networking envi-ronments is difficult, especially whenonly limited privileges are available andno prior device instrumentation hastaken place.

Classical methods for inventory and assetmanagement quickly reach their limit intoday’s dynamic environments: Periodicphysical inventories ('wall-to-wall') havethe distinct advantage of identifying theactual location of devices but requirecostly human visits ('sneaker net') andcannot detect mobile, currently out-of-office equipment nor the existence anduse of logical assets. Financial assettracking, although an accepted process inits own right, cannot detect additionalequipment brought into or accessing theresources of an organization. Periodicself-assessment questionnaires to befilled out by individual end users or theircost-center managers are another, oftencomplementary approach. Apart from thehuman effort they require and the inaccu-rate, incomplete data they may contain,most forms pose questions the answers to

which could easily be obtained from theinfrastructure itself.

Well-managed computing infrastructurestypically equip servers and end-userdevices with software daemons to track thesystem’s resources and health. There are

many situations, however, in which thesedaemons cannot be relied upon. In manyorganizations, there are a fair number ofdevices that are brought in ad-hoc withoutappropriate instrumentation, for whichinstrumentation is not available, or onwhich instrumentation has been disabled.

It therefore appears advisable tocomplement process and policy-basedasset and security management with'automatic sensors' to recalibrate thedynamic and heterogeneous environ-ment. This is where IDD fits in.

ScenariosYour organization may already havedetailed inventories, but what about the ITinfrastructure of the company you interact

with? How can you quickly substitute inany missing data?

How many old Windows systems withless than 512 MB of memory do we needto upgrade this geography? Where doesthis device with a strange MAC address

come from? Which machines are unsafe,ie, without the latest patches and notprotected by antivirus and security soft-ware?

Computer A is a safe machine used by asecretarial employee on the mornings offour out of seven days a week. ComputerB — perhaps a student-lab workstation— has many server ports open, is notcompletely safe, and is shared by users X,Y and Z. Some server has local unpro-tected databases and no screen passwordset. A portable device currently in officeC has no hardware password set.

And those rogue servers should either beturned off or turned into a policy-basedoperation. We may want to send e-mailsto those employees whose machine

In today’s dynamic information society, organizations critically depend on theunderlying computing infrastructure. Mobile users and computing devices addto the challenge. Business operations, intellectual property, and corporate valueare quickly at risk. The IBM Zurich Research Laboratory has developed a toolcalled Intelligent Device Discovery (IDD) that collects information from all possiblesources, computes an aggregate picture of assets, and categorizes their usageand vulnerabilities.

Tracking Assets and Vulnerabilities in Corporate Networksby Dieter Gantenbein

Networks, communication paths, and sources of information.

I BM

Accountingand Inventory

Network Servicesand Syslog

SubnetTraffic ProbesEnd System

Configurations

GatewayAccess Logs

Appl. ServerUsage Logs

NetworkFlows



configurations harbor discovered oraccepted risks. Discovery provides themeans to detect early potential exposuresand track down the exact location ofassets in violation of current policies.

Information SourcesCorporate networks of large companiesspan geographies, the individual loca-tions of which typically consist of sepa-rate segments for user access, servers, aswell as intranet and extranet connectionzones. The operation of wide-area and/orcampus networking, servers, and/or userdevices may be handled by differentorganizations. In general, there areneither universal administrators nor “onesize fits all” credentials. In order, never-theless, to be able to derive an overallpicture, we propose that data beharvested from whatever sourcepossible, and then combined into an inte-grated information model. The range ofinformation sources considered encom-passes network-based autodiscovery,privilege-based additional access to on-line network and application subsys-tems, and the lean processing of manu-ally provided ledger tables containingfinancial and physical inventories andconfigurations. Bothering the end user isconsidered only as a last resort.

TechniquesThe sources of information are manifold,and the techniques to access them areeven more so. The communication stackand the device ports can be activelymapped by scanning from administrativemachines, including IP ping sweeps,UDP/TCP port scans, and remoteWindows SMB/Registry/WMI finger-printing. Should established policieslimit the yield of this technique for someof the target devices, analysis of subnettraffic can also passively map otherwisestealthy devices. Although this approachassumes probes on appropriate segmentsand VLANs, the trend to corporatenetflow architectures that collect aggre-gated traffic accounting records fromstrategic subnets provides for a morestrategic access-point for network flowinformation. Observing the communica-tion between devices can be also used tolog networking services, including(DNS) name servers containing direc-tory information about registeredresources, DHCP address servers withlease information, WLAN wirelessaccess servers, and PIX firewalls. Someof the related events may also be readilyavailable at syslog servers. Whereasmanaged end-user systems may beinstrumented with agents that interact

periodically with corporate managementservers, eg to keep software up-to-dateor to download new virus and wormdefense policies, there is also the entirerealm of other application servers formail, group communication, Web appli-cations, mobile-user access portals, etc.,that may provide detailed informationabout usage patterns.

ConclusionsIDD is a network-based IT assetdiscovery and categorization tool. Itcombines various device discovery,network and security scanning programswith enhanced data collection, miningtechniques, and distributed automationto form a single application built uponthe IBM e-business platform withWebSphere and DB2 on Windows andLinux. It is used today by consulting,outsourcing, and security teams oninternal and customer networks.

Link:http://www.zurich.ibm.com/idd

Please contact:Dieter Gantenbein, IBM Zurich Research Lab/SARITTel: +41 1 724 88 53E-mail: [email protected]

The expansion of the Internet and thedevelopment of secure access andcommunication methods enable a newand integrative usage of simulation tools.Web portals offer the integration of data,applications, services and resources in apersonalized environment.

The Internet or Intranet enable devel-opers to access central resources such asdatabases or computer systems via agraphical user interface. This does notonly reduce investment costs, but also

keeps expenditures for installations andsupport as well as licenses at a minimum.

From Authentication to SimulationImportant components of a web portalare:• user management – the basis for

authentication • object management – for the realiza-

tion of ‘user workspaces’• control of processes incl. accounting

of computer resources and licensefees.

Modern authentication methods, safecommunication and quick databaseconnections allow clients to useresources via the portal independently oftime or location.

Numerical Simulation in a Web Portal Through increased computer powernumerical simulation is becoming moreimportant for the development of turbomachines. The software TRACE (Turbo-machinery Research Aerodynamic

Simulation portals provide secure access to data and the use of different analysisand simulation tools via Internet. Benefits are a safe access independent oflocation or time, personalized services with an adequate user support, and modernsecurity methods like certificate authentication or encryption. Thus simulationportals meet industrial requirements in a ‘portable’ environment.

Numerical Simulation through Web Servicesby Stephan Springstubbe, Jürgen Klein and Ottmar Krämer-Fuhrmann

Computational Environment), devel-oped by the Institute for PropulsionTechnology of the German AerospaceCenter (DLR) in Köln-Porz, provides adetailed insight into complex fluidprocesses.

Fraunhofer SCAI integrated this soft-ware solution into a web portal, thusenabling a better use by the industry.

Individual Services for Users – the TRACE-Portal Web portals cannot only be used easilyby means of a browser, but also offer apersonalized Internet environment anduser guidance. This supports perfectly avariety of user roles like Internet surfers,registered users, licensed customers andadministrators.

In particular the role und group defini-tion of users makes it easy to supportvarious development teams in the bestpossible way. The portal solution isportable to other computer systems andcan be customized to every applicationand service. Thus portals can easily beadapted to new technologies.

Client-Server Technology on the Basis of Sun ONE™The software Sun Open NetworkEnvironment (Sun ONE™) is the basis forthe implementation of the TRACE Portal.Sun ONE™ is Sun's standards-based soft-ware platform for building Web Serviceson Demand. Originally being developedfor e-commerce applications, now Sun

ONE™ has been adapted by FraunhoferSCAI for scientific applications. SunONE™ supports web portal solutions witha modular architecture. The TRACEPortal consists of three components (seeFigure 2): Web container, Enterprise JavaBeans container and Compute server.Each of these components realize a webservice interface and is therefore easilyadaptable to new services or technologies.In the following we describe the portalarchitecture.

ClientThe simplest access to the portal serveris via a browser. The browser realizes

access to the public area by simpleHTTP protocol, and for registered usersto the restricted area by the HTTPSprotocol using ssl encryption.

Alternatively web services or applica-tions can access the simulation servicedirectly. In both cases the client systemcommunicates via Internet connectionwith the web portal server.

Web and Application ServerServer Module ‘Web Container’: Theserver module accessed by the client is theweb container. It realizes the presentationlayer forming the user interface. It isimplemented by Servlets and Java serverpages (JSP). The web container executesthe user authentication and guidesthrough all portal services by means of agraphical user interface. Different userroles are supported by the portal, so thateach user works in a personalized webenvironment with individual services,account and workspace data.

Server Module ‘EJB Container’: TheEnterprise Java Beans (EJB) containerrealizes the business logic and can beunderstood as an interface between theuser and resources, like the database orthe compute server. The EJB container isrealized in Java.

Server Module ‘Database’: The databasestores all data of the portal, like user



Figure 2: Architecture of the TRACE-Portal.

Figure 1: Time-accurate 3D-Simulation of a Propfan with TRACE.

.

Cou

rtes

y of

DLR



data, virtual user workspace, versionmanagement and account data.

Compute ServerThe most important web service for theuser is the execution and control ofnumerical simulations. This is controlledby Enterprise Java Beans of the EJBcontainer. The data access layer allows toupload data to the computing platform.TRACE simulations are executed on amulti-processor Sun-server or on acomputational GRID.

The user can monitor and control compu-tations. He is able to configure, start, stopor terminate simulation processes.Accounting is done automatically by theportal.

The portal is well suited for the use, ifseveral departments are working with thesame simulation code. Portal solutionssupport access to central resources,which minimize license fees, man powerof installation, updating, and administra-tion efforts.

Fraunhofer SCAI offers complete solu-tions for the conception, design and reali-sation of innovative Internet presenta-tions. Portal solutions offering numericalsimulation as a web service in a safe, user-friendly web environment, which can beused independent of location or time.

Links:http://www.scai.fraunhofer.de/webhttp://www.trace-portal.de

Please contact:Ottmar Krämer-FuhrmannFraunhofer SCAIE-mail: [email protected]

The current trend in telecommunications is towards ubiquitous availability of anyapplication - at any time or any place. This will eventually lead to the universalaccessibility of a common network infrastructure that supports all imaginablecommunication services. CWI’s researchers are working on the use of queueingtheory in communication networking, which is relevant in many areas, includingwireless services and network economics research. Application of the developedtechniques could lead to a more economic use of network resources.

Advanced Communication Networksby Michel Mandjes

Technological innovations innetworking are occurring at a rapid pace.Whereas early networks supported alimited number of communicationservices, nowadays a broad variety ofservices are integrated into a singleinfrastructure. This inevitably leads toperformance issues. For instance, a voiceuser asks for essentially real-timeservice, in particular minimal delay,whereas these delay requirements are farless demanding when sending an e-mail.While the network could be designed tomeet the most stringent of these Qualityof Service (QoS) requirements, this

would clearly lead to low utilisation, andconsequently a waste of resources.

CWI’s Advanced CommunicationNetworks theme (PNA2) aims to applyqueueing theory to tackle these trafficmanagement problems. Queueing theoryis all about congestion effects arisingfrom scarce resources. It helps to under-stand how congestion arises, but alsohow congestion can be avoided. Inparticular, queueing theory facilitates thedevelopment of QoS differentiationmechanisms. CWI’s research hasfocused on novel scheduling mecha-nisms, such as Generalised Processor

Sharing (GPS), that are capable ofoffering multiple QoS levels. A keyresearch issue in this area concerns theprotection of user classes against ‘misbe-having’ traffic streams – in this way GPSmitigates the impact of heavy tails. Thisstrand of research requires sophisticatedprobabilistic techniques such as large-deviations methods, as well as a thor-ough knowledge of Gaussian processes.

Recent developments in networkinghave led to challenging new researchdirections. One important field concernsperformance support for wirelessservices. Current wireless systems likeGSM are highly customised for carryingvoice traffic, and offer only limited low-bandwidth data applications, such asshort messaging services (SMS). Thenext generation of wireless networks likeUMTS are specifically designed tosupport a wide range of high-speed dataapplications, such as Web-browsingsessions, document transfers, andimaging services, in addition to conven-tional voice calls. The integration ofthese heterogeneous applications raisessimilar challenges to those describedabove for wireline systems. However in

Due to both the ongoing increase

in traffic volume and the

introduction of new network

services, major infrastructural

upgrades are necessary, such as

here in the Sciencepark

Amsterdam. The networking

group at CWI explores the trade-

off between efficient use of

network resources and the quality

of service offered.

Pho

to: C

WI



wireless environments, these issues arefurther exacerbated by the relatively lowbandwidth and scarcity of spectrum (seethe UMTS licence auctions), and thepropagation characteristics of wirelesssignals.

In particular, the transmission rate inwireless communications fluctuatesconsiderably over time due to multi-pathpropagation effects. These rate fluctua-tions provide an opportunity to achievethroughput gains by scheduling datatransmissions. Recent research in PNA2has explored the QoS as perceived byusers of such ‘channel-aware’ or ‘oppor-tunistic’ scheduling algorithms. It wasshown that user-level performance maybe evaluated by means of a processor-sharing model with a variable servicerate. The variable service rate captures

the throughput gains achieved by thechannel-aware scheduling algorithms.

A second major effort has been investedinto network economics research. Asargued above, a prerequisite for thesupport of heterogeneous applications isthe ability to offer multiple performancelevels. A next step is to decide how theseperformance levels should be priced.Evidently, with no pricing, or with asimple flat fee, all users would opt for thepremium class, disabling performancedifferentiation. This can be seen intoday’s Internet: the flat fee does notencourage ‘economical use’ ofresources. The consequence is that theInternet tends to be used by those whoare relatively insensitive to congestion,thus pushing away services that requirestringent performance guarantees (suchas voice users). This cross-disciplinary

phenomenon is usually called the‘tragedy of the commons’. A way toprevent this ‘tragedy’ from happening isby instituting charges. Performance-sensitive users are usually willing to paymore for the service than congestion-indifferent users. Hence by choosing theprices of the different performance levelsappropriately, incentives can be givensuch that only the performance-sensitiveusers join the premium class. Recentresearch in PNA2 has focused on thistype of problem, with our approachincorporating microeconomic and game-theoretic elements into the queueingframework.

Link:http://www.cwi.nl/pna2

Please contact: Michel Mandjes, CWIE-mail: [email protected]

This graduate education and researchcooperation initiative is currently run bythe five departments of Telematics,Computer and Information Science,Mathematical Sciences, PhysicalElectronics, and Industrial Economicsand Technology Management. Theoverall aim is to address problems ofICT vulnerability at the strategic level bysupervising and carrying out researchwith PhD students on a wide range oftopics essential to information securitytechnology. In this way we will trainexperts with the ability to perform effec-tive information security analyses andprovide solutions to ICT vulnerabilities,wherever they occur.

Information SecurityInformation security is vital for the accep-tance and common use of informationsystems in most sectors of activity in ourinformation-intensive society, includinghealth, finance, trade, public and privateadministration, media and entertainment,communication and transport, and within

the ICT industry itself. Open networkedsystems are particularly challenging inthis regard. Clearly, Internet services andmobile networks already suffer from thiskind of problem, resulting in a greatdemand for comprehensive research andthe training of experts within informationsecurity technology.

Currently, technology and methods forinformation security involve:• hardware-based mechanisms (tamper-

resistance, OS kernel support, andsignal processing)

• cryptology (mathematical primitives,protocols and models)

• software engineering (language/mech-anisms/tools)

• software systems (operating systems,database systems, middleware platforms)

• networked systems (communicationprotocols, naming, routing, adaptivity)

• methodology (formal logic, evaluationcriteria, threat and vulnerability anal-yses, audits, best operation practiceand policies).

The difficulty of analysis and construc-tion in information security technologygrows rapidly with increasing scala-bility, functionality, resource distribu-tion and partitioning of security policy.Information security requirements placeconditions on the design of system struc-ture, user interfaces, data storage,processing locality, communication, andmanagement. On the other hand, solu-tions must meet user expectations and beacceptable within cost restrictions.

Norwegian Research ProgramIn the financial allocation letter of 2003,the Norwegian Ministry of Trade andIndustry requested that the NorwegianResearch Council (NFR) initiate astrategic research program within the fieldof ’ICT Security and Vulnerability’ (IKT-SoS). A total budget of NOK 59 millionwas allocated, and it is to be operationalfrom 2003 to 2007 (1 Euro ª 8 NOK).Only NOK 15 million of the total budgetwere allotted for this first year by thecouncil. Nevertheless, total submissions

NTNU, the Norvegian University of Science and Technlolgy has established aninterfaculty Information Security Research Program, organised under theuniversity’s strategic focus area of Information and Communication Technologies.

Information Security Research at NTNUby Stig F. Mjølsnes and Ingvild Ytrehus



for research funding amounted to NOK 90million, indicating a strong interest byNorwegian research groups in working onthe varied problems in this field.

NTNU ObjectivesThe eight-professor committee of theNTNU Information Security ResearchProgram has created and submitted projectplans which will contribute directly to thegoals set through the National Strategy forICT Security in general, and by theNorwegian Science Council IKT–SoS.These include the following initiatives, asspecified in the call:• strengthening of national education in

information security in both depth andscope

• further strengthening of existing infor-mation security research networks inNorway and Europe, and take-up byuser organisations

• support for political strategies aimed atreducing ICT vulnerabilities

• pursuit of excellence in informationsecurity RTD in Norway.

About fifteen years ago, NTNU startededucational activity within this field in

the Department of Telematics. Studentinterest and motivation are now veryencouraging. This year (2003), one ofthe Masters degree information securitycourses produced 170 examinants, andabout 36 Masters theses in informationsecurity were completed in the fivecooperating NTNU departments. Nextyear, the number of PhD candidates isexpected to be 12-15, within a frame-work of ongoing research projects, labs,and international university cooperation.

Research ProjectsNew research projects that have beendefined and are underway are presentedas four ellipses in the depicted NTNUreference model, ranging from human-organisational issues (‘people’) to deepmathematical theories (‘machines’).Stage 1 funding by NFR enablescommencement of a PhD project in role-based access control, a post-doctoralfellowship in secure protocols, andfunding for visiting scientists in all areasindicated. Stage 2 next year will enhancethis picture. Additionally, several MarieCurie Fellowships are open for visitingPhD students. Communication security

research forms a significant part of theactivity in the recently establishedCentre of Excellence, ‘QuantifiableQuality of Service in CommunicationSystems’ at NTNU, in which around fivePhD positions are allocated to security-related studies as of now. Cooperatingwith NTNU, industrial research projectsrelated to information security areexecuted at SINTEF and UNINETT,including the National Centre forInformation Security in Trondheim.

Links: Strategic focus area of ICT at NTNU: http://www.ntnu.no/satsingsomraader/ikt/

Information Security Program: http://www.ntnu.no/satsingsomraader/ikt/fp/info_sikker/index.htm

Q2S: http://www.ntnu.no/q2s/protrust.php

Marie Curie Training Site in Reliability, Safety and Security Studies at NTNU: http://plato.maskin.ntnu.no/emner/mcts/visartikkel.php?id=18

NORSIS: http://www.norsis.no

Please contact:Stig Frode Mjølsnes, NTNUTel: +47 73 55 04 59E-mail: [email protected]

Text classification (TC) is the activity ofautomatically building, by means ofmachine learning (ML) techniques,programs ('text classifiers') capable of'pigeonholing' natural language texts, ieplacing them in categories from a prede-fined set, according to an analysis of theircontent. Instances of text classification aretopic spotting, spam filtering, genre classi-fication, or automated authorship attribu-tion, depending on the nature and meaningof the categories being considered.

The construction of an automatic textclassifier is usually articulated in threephases. The first phase is term selection,in which the most relevant terms for theclassification task are identified. The

second phase is term weighting, in whichdocument-dependent weights for theselected terms are computed so as togenerate a vectorial representation foreach document, in which each term isweighted by its contribution to the exten-sional semantics of the document. Thethird phase is classifier learning, inwhich a learning device generates a clas-sifier from the vectorial representationsof the training documents.

This entire process involves an activityof 'supervised learning', ie one in whichinformation on the membership oftraining documents in specific categoriesis used. Traditionally, supervisedlearning only enters into phases 1 and 3;

phase 2, instead, usually relies on tech-niques borrowed from text search suchas tf-idf ('text frequency * inverse docu-ment frequency'), a weighting functionbased on the distribution of the termswithin the document and within thecollection, where a high value indicatesthat the word occurs often in the docu-ment and does not occur in many otherdocuments. As a consequence, thesetechniques do not exploit the informa-tion provided by training documentssince text search does not involve anytraining documents.

In our current work we propose thatlearning from training documents shouldalso affect the term weighting phase, ie

Researchers from ISTI-CNR, Pisa, aim at producing better text classificationmethods through the use of supervised learning techniques in the generation ofthe internal representations of the texts.

Supervised Term Weighting for Automated Text Categorizationby Franca Debole and Fabrizio Sebastiani



that information on the membership oftraining documents in specific categoriesbe used to determine term weights. Wecall this idea supervised term weighting(STW). As an example of STW wepropose a number of 'supervised vari-ants' of tf*idf weighting, obtained byreplacing the idf part of tf*idf with thesame function that has previously beenused in the term selection phase. Therationale of replacing idf lies in the factthat idf represents a measure of the docu-ment-independent value of a term, but assuch it is suboptimal in a text classifica-tion context, in the sense that it relies onan intuition ("the document-independentvalue of a term is inversely proportionalto the number of documents in which it

occurs") that is valid also in informationretrieval tasks in which no training dataare available. Feature selection functionsrely instead on an intuition ('the docu-ment-independent value of a term isdirectly proportional to how differentlythe term is distributed in the positive andnegative examples of the categories ofinterest') that refers to the presence ofcategories, and that is thus specific totasks in which category data is available.

We have run STW experiments onReuters-21578, the standard benchmarkof text classification research, with threeclassifier learning methods (Rocchio,kNN, support vector machines), threeterm selection functions (information

gain, chi-square, and gain ratio), and twopolicies for addressing term selectionand weighting ("local" and "global").Results show that STW is a powerfulnotion since, when instantiated with the'gain ratio' feature selection function, itcan bring about improvements as high as11% in accuracy over a standard tf*idfrepresentation with a support vectormachine learner.

Links: http://faure.iei.pi.cnr.it/~fabrizio/Publications/SAC03b.pdf

Please contact: Fabrizio Sebastiani, ISTI-CNR Tel: +39 050 3152892 E-mail: [email protected]

This technology, if adopted, could helplimit a patient’s health risk, increase hisor her level of comfort, reduce both thenumber of check-ups and the time spentin hospital, and enhance the effectivenessof pharmaceutical field tests. The solu-tion includes a so-called 'pill box' and ahome-based blood-pressure monitor thatresembles a large wristwatch.

These new prototypes build on existingdevices available through specialistsuppliers and diagnostics companies. Thenew enhancements make it possible forthe devices to transmit secure patient data,free of tampering, through a mobilephone to an Internet portal and so to thedoctors and nurses monitoring the patient.

The wristband blood-pressure moni-toring device can also check other vitalsigns such as heart rate, and is activatedby simply pressing a button. Blood-pres-sure readings, for example, are gatheredfrom one or more sensors via a Bluetoothshort-range radio connection. Oncetransmitted, secure access ensures onlyauthorised medical personnel see thepatient's data. If an unusual readingcomes through, either a reminder can be

sent to the patient to take his or her medi-cation or a new prescription can be made,depending on the doctor's diagnosis.

The 'pill box' helps keep track of thepatient's medication by sending a signalto his or her mobile phone every time apill is removed. If a patient forgets totake medication or is taking too manypills, he or she is sent a reminder via

mobile phone to follow the prescribeddoses.

Since vital measurements are readilyavailable and can be monitored on-the-go, it is expected that both these solu-tions will increase patient mobility andhelp reduce the need for patients to visitthe doctor . Regular and frequent moni-toring can also bring peace of mind topeople with chronic illnesses or thoserecovering from an operation. Thesedevices are already of great interest tothe pharmaceutical industry, which willbe able to conduct field trials of newdrugs more accurately.

These solutions are examples of howtechnology from the emerging field ofinformation-based medicine can be usedto deliver medical information in realtime to help the medical profession andthe health industry improve the qualityof patient care.

Link:http://www.zurich.ibm.com/csc/mobile/index.html

Please Contact: Dirk Husemann, IBM Zurich Research Lab/SARITE-mail [email protected]

IBM researchers working with medical device manufacturers and mobile-phonehandset manufacturers have created a unique solution to track vital health signs.

Remote Monitoring of Health Conditionsby Dirk Husemann

Health-monitoring devices such as the

blood-pressure cuff (left) and pill box

(right) send data to the mobile phone via

Bluetooth. The mobile hub software

integrated into the mobile phone (centre)

forwards the data to a care centre for

monitoring (screen in the back) and

returns reminders or alarms in an

emergency.


FM03 was organised by FormalMethods Europe, an independent associ-ation whose purpose is to promote andsupport the industrial use of formalmethods for the development ofcomputer systems. Formal methods havebeen controversial throughout theirhistory, and the realisation of their fullpotential remains, in the eyes of manypractitioners, merely a promise. Havethey been successful in industry? If so,under which conditions? Has anyprogress been made in dispelling thescepticism that surrounds them? Arethey worth the effort? Which aspects offormal methods have become so wellestablished in industrial practice thatthey have lost the 'formal method' label?

FM 2003 aimed to answer these ques-tions by seeking contributions not onlyfrom the Formal Methods communitybut also from outsiders and even fromsceptics. The organisers were confidentthat the presentation of a wide spectrumof experiences and the opportunity forthe open discussion of contrasting opin-ions would foster a better and deeperunderstanding, if not a wider adoption ofFormal Methods.

The scientific programme chairs of thesymposium were Keijiro Araki (KyushuUniversity) and Dino Mandrioli(Politecnico di Milano). 144 submis-sions – mainly of high quality – werereceived from more than 25 countriesthroughout the world. Less than onethird were accepted, thus maintaining thetraditionally high standards of the FMsymposia. The symposium also includedtalks by four invited speakers: KouichiKishida (SRA, Japan), Brian Randell(University of Newcastle, UK), GerardHolzmann (NASA/JPL, USA), Jean-Raymond Abrial (Consultant, France)and two presentations by industrial spon-

sors. Eight tutorials, seven workshops,and tool demonstrations were associatedwith the main programme.

New to the symposium was theIndustrial Day dedicated to FormalMethods and Industry, organized byDines Bjørner (Technical University ofDenmark) and sponsored by FME andCoLogNet. There is a strong commit-ment within FME towards an increasinginvolvement of industry in the formalmethods community. The Industrial Daythus had two purposes: to focus on theindustrial use of formal methods withinFM 2003, and to launch a new associa-tion, ForTIA, the Formal TechniquesIndustrial Association. The first Chair ofForTIA is Anthony Hall from PraxisCritical Systems, UK.

Links: http://fme03.isti.cnr.it/http://fmeurope.orghttp://www.eurice.de/colognet

Please contact:Stefania Gnesi, ISTI-CNRGeneral Chair FM 03Tel: +39 050 3152918E-mail: [email protected]

More than 200 scientists from academia and industry attended the 12thInternational FME Symposium FM 2003, an ERCIM sponsored event, hosted byISTI-CNR, 8-13 September, 2003, at the CNR research campus, Pisa.

FM 2003 – 12th International Symposium on Formal Methodsby Stefania Gnesi

EVENTS

ERCIM sponsors up to tenconferences, workshops andsummer schools per year. Thefunding for all types of events isin the order of 2000 Euro.

ConferencesERCIM invites sponsorshipproposals from establishedconferences with an internationalreputation, where substantiveoverlap can be shown between theconference topic and ERCIM areasof activity. Typical cases wouldinclude annual conferences incomputer science with internationalprogramme committees, substantialinternational participation, andproceedings published with anestablished international sciencepublisher.

Workshops and Summer SchoolsERCIM sponsors workshops orsummer schools (co-) organised byan ERCIM institute. The additionalfunding provided by ERCIM shouldbe used to enhance the workshopby, for example, increasing thenumber of external speakerssupported.

Forthcoming Events Sponsoredby ERCIM:• SOFSEM 2004 – The Conference

on Current Trends in Theory andPractice of Informatics,Merin, Czech Republic, 24-30January 2004

• 9th International Conference onExtending Database Technology(EDBT'04), Heraklion, Crete,Greece, 14-18 March 2004

• 27th Annual International ACMSIGIR Conference, Sheffield, UK,25-29 July 2004

• MFCS 2004 - 29th InternationalSymposium on

Mathematical Foundations ofComputer Science, Prague,Czech Republic, 22-27 August2004

Further information:

http://www.ercim.org/activity/sponsored.html

ERCIM-Sponsored Events

Stefania Gnesi, general chair of FM'03

addresses the audience.


EVENTS

ERCIM Workshop onSoft Computing 2003by Petr Hájek

The second workshop of the ERCIMworking group on soft computing washeld from 18-20 October in Brno, CzechRepublic. It was held in parallel with theDATAKON conference, a high-profiletraditional conference focussed ontheoretical and technical backgrounds,best practices and development trendsin the deployment of informationtechnology for information systemsdevelopment, including practicalresults from industry.

The previous three workshops on SoftComputing (with the 2002 workshop thefirst to be recognised as an ERCIMworkshop) were held in parallel withSOFSEM conferences. Unfortunatelythis year SOFSEM altered its conferencedate such that it clashed with an impor-tant conference on fuzzy logic, leading

to the workshop aligning itself insteadwith DATAKON. This proved to beextremely helpful, since local adminis-trative matters were attended to by theDATAKON organisers. This includedthe opening concert and welcome recep-tion on October 18 (the arrival day). Ourthanks are due to Professor Staudek fromMasaryk University, the chair ofDATAKON, for this arrangement.

The 2003 workshop included fourinvited talks and nine contributions.Proceedings with full papers appeared asIssue 5 of Volume 13 of the internationaljournal, Neural Network World.

Invited Talks• C. G. Fermüller: ‘Theories of vague-

ness versus fuzzy logic: can logicianslearn from philosophers?’

• B Gerla: ‘Many-valued logics andsemirings’

• S Jenei and F Montagna: ‘A proof ofstandard completeness for non-commutative monoidal t-norm logic’

• P Jipsen: ‘An overview of generalisedbasic logic algebras’.

Contributed papers from Czech, Slovak,British and Italian authors ranged overvarious topics of probabilistic, possi-bilistic and fuzzy approaches tomanaging uncertainty.

To quote from my editorial in theproceedings: "The fourth workshop onsoft computing appears to continue inour good tradition and to contribute tofoundational and other aspects of thediscipline of soft computing. It is hopedthat we shall keep this tradition also inthe future."

Link: http://www.datakon.cz/datakon03/softcomp.html

Please contact: Petr Hájek Institute of Computer Science, Academy of Sciences of the Czech Republic/CRCIME-mail: [email protected]

ISC 2003 - IndustrialSimulation Conferenceby Ana Pajares, Juan C. Guerriand Carlos Palau

The first Industrial SimulationConference (ISC’2003) was held inValencia, Spain, from 9-11 June 2003.This annual international conference isintended to give a complete overviewof recent industrial simulation relatedresearch for the european communityand the rest of the world.

The meeting, hosted by theCommunications Department of theTechnical University of Valencia, wasorganized by the European SimulationSociety (Eurosis) and co-sponsored bythe Ecole Nationale Supérieure des Artset Industries Textiles (ENSAIT) and theGhent University. With the objective ofexchanging techniques and ideas amonguniversities and industry, ISC’2003 put

together both scentific research resultsfrom universities and practicalapproaches from industry, which partici-pated with a variety of contributions tothe following topics:• Methodology• Verification and Analysis• Complex System Simulation• Simulation in Electronics• Simulation in Engineering• Automotive Simulation• Simulation in Logistics• Simulation in Manufacturing• Chemical Engineering and Energy

Simulation• Simulation in Robotics and Multibody

Applications• Simulation in Aerospace and Traffic• Modelling and Simulation in the

Textile Industry.

More than 80 scientific papers werepresented at ISC’2003, which werepublished in the proceedings of theconference. Besides these contributions,the conference also included a keynote

speech about knowledge discovery indatabases and modelling, and two tuto-rials concerning parallelism in discreteevent simulation and modelling andsimulation of complex cell spaces.Together with the ISC’2003 conferencea workshop on modelling and simulationin the textile industry took place.

The second edition of ISC (ISC’2004)will be held in Malaga, Spain.

Links:ISC 2003: http://www.eurosis.orgISC 2004: http://www.eurosis.orgCommunications Dept. at UPV:http://www.dcom.upv.es

Please contact:Ana Pajares, Universidad Politécnica de Valencia E-mail: [email protected]

Philippe Geril, EUROSIS - SIM-CENTER,Ghent University, BelgiumE-mail: [email protected]


EVENTS

CALL FOR PARTICIPATION

CLEF 2004

The Cross-Language Evaluation Forumaims at promoting research into thedesign of user-friendly, multilingual,multimodal retrieval systems byproviding an infrastructure for systemtesting and tuning, comparison ofdifferent approaches, and creation ofreusable evaluation test-suits.

The CLEF 2004 campaign offers thefollowing system evaluation tracks:Mono-, Bi- and MultilingualInformation Retrieval on NewsCollections. The document collection forCLEF2004 will contain English,Finnish, French, and Russian docu-ments. A common set of topics (ie struc-tured statements of information needsfrom which queries are extracted) will beprepared in Dutch, English, Finnish,French, German, Italian, Spanish,Swedish, Russian, Japanese andChinese. Topics in other languages canbe supplied on demand.

Mono- and Cross-LanguageInformation Retrieval on StructuredScientific DataThe rationale for this track is to studyretrieval in a domain-specific contextusing the GIRT-4 German/English socialscience database with controlled vocabu-laries in English, German and Russian.Topics will be available in English,French, German and Russian.

Interactive Cross-LanguageInformation Retrieval - iCLEFThis year, the interactive CLEF trackwill study the problem of Cross-Language Question Answering from auser-inclusive perspective.

Multiple Language QuestionAnswering (QA@CLEF)Mono- and Cross-Language QA systemswill be tested. Languages involved areDutch, French, German, Italian, Spanish,English.

Cross-Language Image Retrieval(ImageCLEF)This track evaluates retrieval of imagesdescribed by text captions based onqueries in a different language; both text

and image matching techniques arepotentially exploitable. Two test collec-tions are available: St AndrewsUniversity historical photographiccollection; University Hospitals Genevamedical images.

Cross-Language Spoken DocumentRetrieval (CL-SDR)This track aims at the evaluation ofCLIR systems on noisy automatic tran-scripts of spoken documents and willaddress the following problems: bilin-gual-SDR from Dutch, French, German,Italian, and Spanish; retrievalwith/without known story boundaries;use of multiple automatic transcriptions

ParticipationRegistration is via the CLEF website.

Important Dates• Registration opens:15 January 2004• Data release: from February 2004• Submission of runs by participants

15 May 2004 • Release of individual results

from 15 July 2004• Workshop: 16-17 September 2004

More information: http://www.clef-campaign.org/


8th ERCIM Workshop "User Interfaces For All"

Vienna, Austria, 28-29 June 2004

"User-Centred Interaction Paradigmsfor Universal Access in the InformationSociety" is the special theme of the 8thERCIM Workshop "User Interfaces forAll" which builds upon the results ofthe seven previous workshops.

The vision of User Interfaces for Alladvocates the proactive realisation of the'design for all' principle in the field ofHuman-Computer Interaction (HCI),and involves the development of userinterfaces to interactive applications andtelematic services, which provideuniversal access and usability to poten-tially all users.

Areas of interest for which papers aresolicited, include, but are not limited to,the following topics:Inclusive design and evaluation:• Design methodologies for universal

access• User interface adaptation• Intelligent context-sensitive

processing• Modelling task execution contexts• Universal access in software quality

models• Learnability, tutorials, online help• Evaluation techniques.

Universal access & novel interactiontechnologies & tools:• Multimedia / multimodal interfaces• Novel interaction techniques, devices

and metaphors• Universal access in new interaction

paradigms (wearable and ubiquitouscomputing, tangible interfaces, VirtualReality)

• Interface architectures, developmenttools, interoperability

• Universal access in ubiquitous envi-ronments

• Universal access scenarios for AmIenvironments.

Internet and WWW access:• Accessibility of (public) web sites• Privacy and security• Personalized web content delivery• Universal access in e-services (e-

learning, e-government, e-health, etc)• Online communities for all.

e-Accessibility & Policies:• National, European, international poli-

cies on e-Accessibility• Standardisation & universal access• Legislation for universal access• Economics of universal access.

Important Dates• 15 February 2004: Deadline for elec-

tronic submission of all papers• 1 April 2004: Conditional notification

of acceptance (confirmation will begiven upon registration)

• 1 May 2004: Deadline for electronicsubmission of camera-ready papers

• 31 May 2004: Deadline for registration.

More information: http://www.ui4all.gr/workshop2004/




ECAI'04 —16th EuropeanConference on ArtificialIntelligenceValencia, Spain, 22-27 August 2004

The European Coordinating Committeefor Artificial Intelligence (ECCAI) andthe 'Grupo de Tecnología Informática"of the 'Departamento de SistemasInformáticos y Computación' of the'Universidad Politècnica de Valencia'cordially invite you to participate in the16th European Conference on ArtificialIntelligence (ECAI 2004). ECAIprovides a public forum for researchersfrom academy and industry and publicorganizations, offering them an invalu-able occasion to meet and exchangeideas, thus contributing to making realthe upcoming 'Information Society'.

The conference includes, from 22-24August, two days and a half of work-shops and tutorials on most advancedtopics, as well as the Second STarting AIResearchers Symposium (STAIRS),followed by the three-day technicalprogram of ECAI. Following an estab-lished tradition, the Sub-Conference onPrestigious Applications of IntelligentSystems (PAIS) will run in parallel withECAI. The technical programme willinclude refereed paper presentations,invited talks by prestigious speakers andposter sessions.

Important Dates• Paper abstract submission: 13

February 2004• Full paper submission: 18 February

2004• Notification of acceptance : 2 May

2004• Final paper due : 31 May 2004

More information:http:www.dsic.upv.es/ecai2004

CALL FOR PAPERS

European Conference on Digital Libraries (ECDL) 2004

University of Bath, UK, 12-17 September 2004

ECDL 2004 is the 4th in the series ofEuropean Digital Library Conferences.ECDL has become the major Europeanforum focusing on digital libraries andassociated technical, organisational andsocial issues.

Digital library research brings together anumber of disciplines and practitionercommunities, providing a stimulatingenvironment for debate and an opportu-nity for establishing collaboration.ECDL provides a forum for theexchange of ideas between scientificdisciplines and fosters joint initiativeswith application communities.Involvement of researchers and practi-tioners from computing and informationscience disciplines is well established atECDL. Increasingly these disciplines areengaging in discussion and co-operationwith other groups concerned with know-ledge management. ECDL 2004 encour-ages involvement from a wide range ofdisciplines and looks forward toincreasing contributions from usabilityexperts, educationalists, developers ofeLearning systems, and by workingwithin the eScience and GRID commu-nities. The continued success of ECDLrests on the exchange of ideas and expe-rience between these groups.

ECDL 2004 welcomes contribution andparticipation from scholars, researchers,practitioners and policy makers acrossthe range of disciplines related to thedevelopment of the digital library in itswidest sense.

TopicsContributions may include but are notlimited to:Strategy, management and policy making:• Human resources, economics, busi-

ness models• Evaluation, metrics, testbeds.

Digital library applications:• Digital libraries and learning

• Digital libraries and ResearchGrids/eScience

• Multilingual digital libraries• Geo-spatial, still and moving images,

sound and 3D graphics libraries.

Information architectures and inter-operability:• Infrastructures and middleware• Semantic Web technologies• Agent technologies• Development of Web Services• Application of open standards &

protocols• Ubiquitous environments and mobile

access.

Users of digital libraries:• Personalisation and agent technologies• Annotation and recommendation

services• Presentation and visualisation• Usability, user studies and user-

centred design.

Digital repositories:• Curation and preservation• Provenance and trust• Describing and managing collections• Digital rights management, licences

and schema• Open archives and self archiving.

Knowledge management:• Knowledge extraction, data and text

mining• Semantic interoperability, ontologies

and knowledge representation• Classification and indexing, thesauri

development• Metadata schema, registries and

mapping

Delivery channels:• Changing models of scholarly

communications• Collaboratories• Virtual organisations and digital

libraries.

Important Dates• 5 April 2004: Papers, panels, tutorials

and workshops submissions due.• 19 May 2004: Poster and demonstra-

tion submissions due.• 11 June 2004: Final submission date.

More information:http://www.ecdl2004.org


EVENTS


STAIRS 2004 — SecondEuropean Starting AIResearcher SymposiumValencia, Spain, 23 - 24 August 2004

STAIRS'2004 is the second EuropeanStarting AI Researcher Symposium, aninternational meeting aimed at fosteringinteraction and exchange between AIresearchers, from all countries, at thebeginning of their career. STAIRS'04

CALL FOR PAPERS

The 13th Conference on Mathematics for Industry

Eindhoven, The Netherlands, 21-25 June 2004

The conference will be held at theEindhoven University of Technologyfollowing up ECMI - 2002 in Jurmala,Latvia. In order to have a broader scopeof application of mathematics inindustry, ECMI 2004 is jointly organizedby ECMI and ENBIS (EuropeanNetwork for Business and IndustrialStatistics). The conference will bedevoted to mathematical and statisticalmodelling, analysis and simulation ofproblems arising in a practical context.In particular the following applicationareas have been chosen:• Aerospace• Electronic industry• Chemical technology• Life sciences• Materials• Geophysics• Financial mathematics• Water flow.

Important Dates • Submission of proposals for minisym-

posia: 30 January 2004• Early-bird registration: 1 February

2004• Submission of papers and posters 25

February 2004 • Notification of acceptance of papers

and posters: 25 March 2004.

More information: http://www.ecmi2004.tue.nl/


Symposium on Point-BasedGraphics

Zurich, Switzerland, 2-4 June, 2004

The Symposium on Point-BasedGraphics to be held at the Swiss FederalInstitute of Technology in Zurich,reflects the current renaissance of graph-ical methods based on point-sampledgeometry; this is a result of the dramaticincrease in computational complexityobserved when using polygonal models.

Manuscripts should be submitted by 7 April 2004.

Suggested topics • Data acquisition • Surface reconstruction • Representation of point-sampled

geometry • Sampling,approximation, and interpo-

lation • Digital geometry processing of point

models

offers PhD students and people holding aPhD for less than one year: • a first experience on submitting and

presenting a paper in an internationalforum with a broad scope and thor-ough selection process

• the opportunity to gather knowledgeand exchange ideas related to theirresearch problems and approaches

• information on European researchcareers and mobility.

STAIRS'04 will be held jointly withECAI'04 in Valencia, Spain, on August23rd and 24th. The Symposium will behosted by the 'Departamento de SistemasInformáticos y Computación (DSIC) ofthe Universidad Politécnica de Valencia'.

Important Dates • 1 April 2004: Title and abstract• 7 April 2004: Paper submission dead-

line• 21 May 2004: Notification of accep-

tance sent to authors• 4 June 2004: Camera-ready version due• 23-24 August 2004: Symposium venue

More information:http://www.dsic.upv.es/ecai2004/stairs2004

• Transmission and compression ofpoint-sampled geometry

• Geometric modelling using pointprimitives

• Rendering algorithms and renderingarchitectures for point primitives

• Animation and morphing of point-sampled geometry

• Hybrid representations and algo-rithms.

More information:http://graphics.ethz.ch/pbg/cfp.html

SPONSORED BY ERCIM

MFCS 2004 — 29th InternationalSymposium on MathematicalFoundations of ComputerScience

Prague, Czech Republic, 22-27 August , 2004

The series of MFCS symposia, orga-nized alternately in the Czech Republic,Poland and Slovakia since 1972, has along and well-established tradition. TheMFCS symposia encourage high-qualityresearch in all branches of theoreticalcomputer science.

The topics include algorithms and datastructures, automata, grammars andformal languages, complexity theory(structural and computational), computa-tional biology, computational geometry,concurrency theory, cryptography andsecurity, databases and knowledge-based systems, foundations of program-ming, formal specifications and programdevelopment, machine learning, modelsof computation (parallel and distributedcom-puted, fixed parameter tractability,quantum computing, and other), se-mantics and logics of programs, andtheoretical issues in artificial intelli-gence.

Important Dates • Submission deadline: 15 March 2004• Acceptance/Rejection: 14 May 2004• Final version due: 6 June 2004• Early registration: 15 June 2004

More information: http://mfcs.mff.cuni.cz/


ERCIM News is the magazine of ERCIM.Published quarterly, the newsletter reports onjoint actions of the ERCIM partners, and aimsto reflect the contribution made by ERCIM tothe European Community in InformationTechnology. Through short articles and newsitems, it provides a forum for the exchange ofinformation between the institutes and alsowith the wider scientific community. This issuehas a circulation of over 9000 copies.

Copyright NoticeAll authors, as identified in each article, retaincopyright of their work.

ERCIM News online edition is available athttp://www.ercim.org/publication/ERCIM_News/

ERCIM News is published by ERCIM EEIG, BP 93, F-06902 Sophia-Antipolis CedexTel: +33 4 9238 5010, E-mail: [email protected] 0926-4981

Director: Michel Cosnard

Central Editor:Peter [email protected]

Local Editors:AARIT: n.c.

CCLRC: Martin Prime [email protected]

CRCIM: Michal [email protected]

CWI: Annette Kik [email protected]

CNR: Carol Peters [email protected]

FORTH: George [email protected]

FhG: Michael [email protected]

FNR: Patrik [email protected]

INRIA: Bernard [email protected]

Irish Universities Consortium:Ray [email protected]

NTNU: Truls [email protected]

SARIT: Harry [email protected]

SICS: Kersti [email protected]

SpaRCIM: Salvador [email protected]

SRCIM: Gabriela Andrejkova [email protected]

SZTAKI: Erzsébet Csuhaj-Varjú[email protected]

VTT: Pia-Maria Linden-Linna [email protected]

Free subscriptionYou can subscribe to ERCIM News free ofcharge by: • sending e-mail to your local editor• contacting the ERCIM office (see address

above) • filling out the form at the ERCIM website at

http://www.ercim.org/

The ITU has released the fifth book in its Internet Reports series entitled "TheBirth of Broadband". The book is a fountain of statistics and analyses.

The content is best summarizing by quoting from the Foreword: “The introductorychapter of this report, Broadband dreams, explains what broadband can do forusers, society and industry. Chapter two, Technologies for broadband, explains thedifferent broadband technologies and how each can provide broadband accessunder different economic and network condi-tions. Chapter three, Supplying broadband,looks at how broadband has been successfullyprovided in certain economies and how certainpolicies can help expand the network. Chapterfour, Using broadband, discusses the currentand emerging applications that are drivingbroadband take-up along with applications andcontent models that show the most promise forthe future. Chapter five, Regulatory and policyaspects, examines regulatory and policy frame-works in successful broadband markets. Chaptersix, Promoting broadband, looks at the broad-band experiences of several countries character-ized by high penetration rates and extensivenetworks, including conclusions drawn from ITUcountry case studies on broadband, and examines why and how broadband shouldbe actively promoted. Chapter seven, Broadband and the information society, looksat broadband as a component of a society built around ubiquitous access to infor-mation, including some of the benefits and pitfalls of total connectivity. TheStatistical annex contains data and charts covering 206 economies worldwide, withoriginal data on broadband and comparative information measured against a selec-tion of variables.”

Surprising to this reviewer is how much greater broadband penetration is in some ofthe Far Eastern countries, particularly in the Republic of Korea which leads the restof the world. In Korea, over 20 of 100 inhabitants have broadband and over 40percent of Internet subscribers access the Internet via broadband. Figures for costand hours of usage are given.

The book, nearly 200 pages, may be ordered via [email protected] or the ITU Web site:http://www.itu.int/osg/spu/publications/sales/birthofbroadband/index.html

Harry Rudin, Consultant and Swiss Local Editor for ERCIM News

Book Review


IN BRIEF

Fraunhofer-Gesellschaft — ASino-German Mobile-Communi-cations-Institute, located at theFraunhofer Institute for Telecommuni-cations, Heinrich-Hertz-Institut inBerlin was established in December2003. The institute will cover variousaspects of mobile communications andalso to corroborate and initiate jointChinese and German research activities.Primary responsibility is both theoret-ical and practical research work forevolving third generation mobilecommunication systems. The work issupported by the Ministry of Scienceand Technology (MOST), P.R. Chinaand the Federal Ministry of Educationand Research (BMBF), Germany. Theinstitute is the second wing of the Sino-German research establishment after the'Sino-German Joint Software Institute'which was opened in Bejing earlier in2003. sion takes place via a network ofantennas at fixed locations, temporaryad-hoc networks can be set up usingappropriate technologies. Such tempo-rary networks can be used, for example,for car-to-car communication in vehiclefleets.

INRIA — Michel Cosnardbecamethe new president of INRIA following adecree signed by the French PresidentJacques Chirac. Michel Cosnard takesover from Bernard Larrouturou, whowas recently appointed Chief ExecutiveOfficer of the French National Centre forScientific Research (CNRS).

Michel Cosnard was born in 1952. Hereceived his engineering degree incomputer science from ENSIMAG(Grenoble, France), his Master's degreein applied mathematics from CornellUniversity (USA) in 1975 and hisDoctorat d'Etat in computer science fromthe Université de Grenoble (France) in1983. He joined the CNRS in 1979, andin 1987 he was appointed professor ofcomputer science at the 'Ecole NormaleSupérieure de Lyon' (France), where hefounded the 'Laboratoire d'Informatiquedu Parallélisme' (LIP). He served aschairman of the LIP until 1997.

In 1997, he became director of the researchunit of INRIA in Lorraine, and of LORIA('Laboratoire Lorrain de Recherche enInformatique et ses Applications'). In June2001, he was nominated director of theINRIA research unit in Sophia Antipolis(Nice). His research interests are in thedesign and analysis of parallel algorithms,and the complexity analysis of automataand neural nets. In January 2001, he wasappointed director of the French NationalResearch Program on Grid Computing bythe Ministry of Research. In 1994, MichelCosnard was awarded a national prizefrom the French National Academy ofScience. In 1995, he received the IFIPSilver Core and in 2003, the CharlesBabbage Award. He has published morethan 100 papers related to parallelprocessing.

Michel Cosnard will represent INRIA onERCIM's Board of Directors and willcontinue to act as ERCIM's manager.

©IN

RIA

/ P

hoto

G. F

avie

r -

Age

nce

Vu

NTNU — The NorwegianUniversity of Science and Technologywill spend 40 million NOK (5 millionEuro) to enhance the research andstudies in language technology. Duringthe last 6-7 years, a number of differentprojects in language technology havebeen launched at NTNU. Most of theprojects are externally financed, mainlyby the Research council of Norway, andtheir priority programme KUNSTI(development of knowledge inNorwegian language technology).Nevertheless, many of the minor projectsare financed by internal funds at NTNU.The 40 millions will be shared betweenthe various projects at NTNU. The mainprojects are BRAGE (user interface withnatural language), which focuses onhuman-machine dialogue, and the alliedproject VOCALS, which focuses on theconvergence between communicationsystems, advanced dialogue manage-ment and spoken language technology.There is also a project between NTNUand the University in Bergen calledLOGON, which is a project devoted tomachine translation from Norwegian toEnglish. In addition, there are also manyon-going minor projects. An applicationdeveloped at NTNU is already commer-cialized and available to the public. Thisapplication has a natural language inter-face, and the system can give you infor-mation about the bus schedules inTrondheim. You can either use the webto send your question to the server, oryou can choose to use SMS.

INRIA — Texas Instruments hasestablished a JavaTM competencecenter in Rennes, Francein December2003 as an extension of its effort toexpand the availability of innovativewireless multimedia applications on TIOMAPTM processors and wirelesscommunications chipsets. The centerwill open following a five-year researchprogram conducted jointly with INRIA,that was focused on developing tech-nology that will optimize the perfor-mance of Java applications whenrunning on cell phones and otherportable and converged multimedia elec-tronics. One of the primary goals of theJava competence center will be to drivethe transition of the research results intobusiness reality, thus enabling the largecommunity of Java developers an accel-erated application performance whenthey port their multimedia applicationsonto mobile terminals. During the initialresearch stage of the relationship, TI andINRIA focused on technology develop-ment that addresses the complexity ofperformance and power-hungry Javaapplications and optimizes their compat-

ibility with the real-time constraintsimposed by several multimedia applica-tions running on the same mobile device.By combining INRIA's expertise inembedded systems software develop-ment with TI's wireless systems know-how, the two partners were able to solvethe technical challenges of the researchstage during which over 30 patents werebeing generated.The core team of thenew TI competence center is comprisedof the former INRIA engineers andresearchers who worked with TI on theoriginal joint program. The center willact as a key interface for TI's customersand partners in the Java application areaand is located in Rennes to benefit fromthe close proximity to INRIA-Rennes forfurther joint research activities.

Fraunhofer ICT GroupFriedrichstr. 6010117 Berlin, GermanyTel: +49 30 726 15 66 0, Fax: ++49 30 726 15 66 19http://www.iuk.fraunhofer.de

Institut National de Recherche en Informatique et en AutomatiqueB.P. 105, F-78153 Le Chesnay, FranceTel: +33 1 3963 5511, Fax: +33 1 3963 5330http://www.inria.fr/

Swedish Institute of Computer ScienceBox 1263, SE-164 29 Kista, SwedenTel: +46 8 633 1500, Fax: +46 8 751 72 30http://www.sics.se/

Technical Research Centre of FinlandP.O. Box 1200FIN-02044 VTT, FinlandTel:+358 9 456 6041, Fax :+358 9 456 6027http://www.vtt.fi/tte

Swiss Association for Research in Information Technologyc/o Prof. Dr Alfred Strohmeier, EPFL-IC-LGL, CH-1015 Lausanne, SwitzerlandTel +41 21 693 4231, Fax +41 21 693 5079http://www.sarit.ch/

Slovak Research Consortium for Informatics andMathematics, Comenius University, Dept.of ComputerScience, Mlynska Dolina M, SK-84248 Bratislava, SlovakiaTel: +421 2 654 266 35, Fax: 421 2 654 270 41http://www.srcim.sk

Magyar Tudományos AkadémiaSzámítástechnikai és Automatizálási Kutató IntézetP.O. Box 63, H-1518 Budapest, HungaryTel: +36 1 279 6000, Fax: + 36 1 466 7503http://www.sztaki.hu/

Trinity CollegeDepartment of Computer Science,Dublin 2, IrelandTel: +353 1 608 1765, Fax: 353 1 677 2204http://www.cs.tcd.ie/ERCIM

Austrian Association for Research in ITc/o Österreichische Computer GesellschaftWollzeile 1-3, A-1010 Wien, AustriaTel: +43 1 512 02 35 0, Fax: +43 1 512 02 35 9http://www.aarit.at/

Norwegian University of Science and Technology Faculty of Information Technology, Mathematics andElectrical Engineering, N 7491 Trondheim, NorwayTel: +47 73 59 80 35, Fax: +47 73 59 36 28http://www.ntnu.no/

ERCIM – The European Research Consortium for Informatics and Mathematics is an organisation

dedicated to the advancement of European research and development, in information technology

and applied mathematics. Its national member institutions aim to foster collaborative work within

the European research community and to increase co-operation with European industry.

Order Form

If you wish to subscribe to ERCIM News free of charge

or if you know of a colleague who would like toreceive regular copies of

ERCIM News, please fill in this form and wewill add you/them to the mailing list.

Send, fax or email this form to:ERCIM NEWS

2004 route des LuciolesBP 93

F-06902 Sophia Antipolis CedexFax: +33 4 9238 5011

E-mail: [email protected]

Data from this form will be held on a computer database. By giving your email address, you allow ERCIM to send you email

Name:

Organisation/Company:

Address:

Post Code:

City:

Country

E-mail:

✄

You can also subscribe to ERCIM News and order back copies by f i l l ing out the form at the ERCIM website at http://www.ercim.org/publication/Ercim_News/

ERCIM is the European Host of the World Wide Web Consortium.

Spanish Research Consortium for Informaticsand MathematicsEsperanza Marcos, Rey Juan Carlos University, C/ Tulipan s/n, 28933-Móstoles, Madrid, Spain, Tel: +34 91 664 74 91, Fax: 34 91 664 74 90

SpaRCIM

Consiglio Nazionale delle Ricerche, ISTI-CNRArea della Ricerca CNR di Pisa, Via G. Moruzzi 1, 56124 Pisa, ItalyTel: +39 050 315 2878, Fax: +39 050 315 2810http://www.isti.cnr.it/

Centrum voor Wiskunde en InformaticaKruislaan 413, NL-1098 SJ Amsterdam, The NetherlandsTel: +31 20 592 9333, Fax: +31 20 592 4199http://www.cwi.nl/

Council for the Central Laboratory of the ResearchCouncils, Rutherford Appleton LaboratoryChilton, Didcot, Oxfordshire OX11 0QX, United KingdomTel: +44 1235 82 1900, Fax: +44 1235 44 5385http://www.cclrc.ac.uk/

Foundation for Research and Technology – HellasInstitute of Computer ScienceP.O. Box 1385, GR-71110 Heraklion, Crete, GreeceTel: +30 2810 39 16 00, Fax: +30 2810 39 16 01http://www.ics.forth.gr/FORTH

Czech Research Consortium for Informatics and MathematicsFI MU, Botanicka 68a, CZ-602 00 Brno, Czech RepublicTel: +420 2 688 4669, Fax: +420 2 688 4903http://www.utia.cas.cz/CRCIM/home.html

Fonds National de la Recherche20, Montee de la PetrusseL-2912 LuxembourgTel. +352 26 1925 31, Fax +352 26 1925 35http:// www.fnr.lu

FWOEgmontstraat 5B-1000 Brussels, BelgiumTel: +32 2 512.9110http://www.fwo.be/

FNRSrue d'Egmont 5B-1000 Brussels, BelgiumTel: +32 2 504 92 11http://www.fnrs.be/

Documents

Analysis, Diagnosis, Planning and Simulation of Industrial ... · 3 Keynote by Stelios Orphanoudakis, ERCIM President 4 Tribute to Gerard van Oortmerssen by Stelios Orphanoudakis,