Sne.26.4.1035.OA.Pdf

SIMULATION SNE NOTES EUROPE

SNE Special Issue Modelling and Simulation in Modern Control Engineering

Volume26 N o. 4 Decem ber 2016 doi: 10.11128/sne.26.4.103 5

Journal on Developments and Trends in Modelling and Simulation ARGESIM Membership Journal for Simulation Societies and Groups in EUROSIM SNE E DITORIAL - C ONTENT - I NFORMATION

Siegfried Wassertheurer, [email protected] SNE Editorial Board AIT Austrian Inst. of Technology, Vienna, Austria SNE - Simulation Notes Europe is advised and supervised by Sigrid Wenzel, [email protected] an international scientific editorial board. This board is taking Univ. Kassel, Inst. f. Production Technique, Germany care on peer reviewing and handling of Technical Notes, Edu- cation Notes, Short Notes, Software Notes, Overview Notes, SNE Aims and Scope and of Benchmark Notes (definitions and solutions). At pre- Simulation Notes Europe publishes peer reviewed contri- sent, the board is increasing (see website): butions on developments and trends in modelling and simula- David Al-Dabass, [email protected] tion in various areas and in application and theory, with main Nottingham Trent University, UK topics being simulation aspects and interdisciplinarity. Felix Breitenecker, [email protected] Individual submissions of scientific papers are welcome, as Vienna Univ. of Technology, Austria, Editor-in-chief well as post-conference publications of contributions from Maja Atanasijevic-Kunc, [email protected] conferences of EUROSIM societies. Univ. of Ljubljana, Lab. Modelling & Control, Slovenia SNE welcomes also special issues, either dedicated to spe- Aleš Beliþ, [email protected] Sandoz / National Inst. f. Chemistry, Slovenia cial areas and / or new developments, or on occasion of vents Peter Breedveld, [email protected] as conferences and workshops with special emphasis. University of Twenty, Netherlands Furthermore SNE documents the ARGESIM Benchmarks Agostino Bruzzone, [email protected] on Modelling Approaches and Simulation Implementations Universita degli Studi di Genova, Italy with publication of definitions, solutions and discussions Francois Cellier, [email protected] (Benchmark Notes). Special Educational Notes present the use ETH Zurich, Switzerland of modelling and simulation in and for education and for e- Vlatko ýeriü, [email protected] Univ. Zagreb, Croatia learning. SNE is the official membership journal of EUROSIM, Russell Cheng, [email protected] the Federation of European Simulation Societies. A News Sec- University of Southampton, UK tion in SNE provides information for EUROSIM Simulation So- Eric Dahlquist, [email protected], Mälardalen Univ., Sweden cieties and Simulation Groups. Horst Ecker, [email protected] SNE is published in a printed version (Print ISSN 2305- Vienna Univ. of Technology, Inst. f. Mechanics, Austria 9974) and in an online version (Online ISSN 2306-0271). Vadim Engelson, [email protected] MathCore Engineering, Linköping, Sweden With Online SNE the publisher ARGESIM follows the Open Edmond Hajrizi, [email protected] Access strategy, allowing download of published contributions University for Business and Technology, Pristina, Kosovo for free – identified by a DOI (Digital Object Identifier) as- András Jávor, [email protected], signed to the publisher ARGESIM (DOI prefix 10.11128). Budapest Univ. of Technology and Economics, Hungary Esko Juuso, [email protected] Print SNE, high-resolution Online SNE, full SNE Archive, Univ. Oulu, Dept. Process/Environmental Eng., Finland and source codes of the Benchmark Notes are available for Kaj Juslin, [email protected] members of EUROSIM societies. VTT Technical Research Centre of Finland, Finland Andreas Körner, [email protected] Author’s Info. Authors are invited to submit contributions Technical Univ. Vienna, E-Learning Dpt., Vienna, Austria which have not been published and have not being considered Francesco Longo, [email protected] for publication elsewhere to the SNE Editorial Office. Fur- Univ. of Calabria, Mechanical Department, Italy thermore, SNE invites organizers of EUROSIM conferences Yuri Merkuryev, [email protected], Riga Technical Univ. to submit post-conference publication for the authors of their David Murray-Smith, [email protected] conferences. University of Glasgow, Fac. Electrical Engineering, UK SNE distinguishes different types of contributions (Notes): Gasper Music, [email protected] • Overview Note – State-of-the-Art report in a specific area, Univ. of Ljubljana, Fac. Electrical Engineering, Slovenia up to 14 pages, only upon invitation Thorsten Pawletta, [email protected] • Technical Note – scientific publication on specific topic in Univ. Wismar, Dept. Comp. Engineering, Wismar, Germany modelling and simulation, 6 – 10 pages Niki Popper, [email protected] • Education Note – modelling and simulation in / for education dwh Simulation Services, Vienna, Austria and e-learning; 6 - 8 pages Kozeta Sevrani, [email protected] • Univ. Tirana, Inst.f. Statistics, Albania Short Note – recent development on specific topic, max. 6 p. • Thomas Schriber, [email protected] Software Note – specific implementation with scientific University of Michigan, Business School, USA analysis, 4 – 6 4 pages • Yuri Senichenkov, [email protected] Benchmark Note – Solution to an ARGESIM Benchmark; St. Petersburg Technical University, Russia commented solution 4 pages, comparative solutions 4-8 pages Oliver Ullrich, [email protected] Further info and templates (doc, tex) at SNE’s website. Florida International University, USA www.sne-journal.org SNE E DITORIAL - C ONTENT - I NFORMATION

Editorial Dear Readers - This fourth SNE issue of the year 2016, SNE 26(4), the special issue ‘Modelling and Simulation in Modern Control Engineering’ was suggested and compiled by SLOSIM, the Slovenian Simulation Society – underlining the strategy of SNE to pub- lish contributions on recent trends and developments in modelling and simulation. For the front cover we have chosen figures from the eight contributions, sketching model approaches and other characteristic features, which indeed show the broad variety of modelling and simulation in modern control engineering. This issue also rounds up the emphasis of SNE on thematic – oriented issues, with special issues either of topic-oriented subjects – like this issue and SNE 26(3)with emphasis on System Dynamics – or of issues emphasizing on general developments in modelling and simulation – like SNE 26(2) – the ‘EUROSIM 2016 Congress’ special issue. And last but not least, this issue reflects the status of SNE as membership journal of EUROSIM, the Federation of European Simu- lation Societies, and the activities of the member societies as SLOSIM, which are invited to edit special issues.

I would like to thank all authors for their contributions, and especially the guest editor Vito Logar from SLOSIM for compiling this special issue. And last but not least thanks to the Editorial Office for layout, typesetting, preparations for printing, and web programming for electronic publication of this SNE issue.

Felix Breitenecker, SNE Editor-in-Chief, [email protected]; [email protected]

Contents SNE 26(4) SNE Contact & Info Special Issue ‘Modelling and Simulation SNE Print ISSN 2305-9974, SNE Online ISSN 2306-0271 in Modern Control Engineering’ ї www.sne-journal.org SNE doi: 10.11128/sne.26.4.1035 | [email protected], [email protected]

Jadex/JBdiEmo Emotional Agents in Games with SNE Editorial Office, Andreas Körner Purpose: a Feasibility Demonstration ARGESIM/Math. Modelling & Simulation Group, Š. Koreþko, B. Sobota, P. Zemianek ...... 195 Vienna Univ. of Technology /101, Wiedner Haupstrasse 8-10, 1040 Vienna , Austria Evolving Fuzzy Model (eFuMo) Method for On-line Fuzzy Model Learning with Application to Monitoring SNE S IMULATION N OTES E UROPE System. D. Dovžan, V. Logar, I. Škrjanc ...... 205 WEB: ї www.sne-journal.org, DOI prefix 10.11128/sne Model-based Prediction of the Remaining Useful Scope: Technical Notes, Short Notes and Overview Notes on devel- Life of the Machines. opments and trends in modelling and simulation in various areas and in application and theory; benchmarks and benchmark docu- P. Boškoski, B. Dolenc, B. Musizza, Ĉ. Juriþiü ...... 221 mentations of ARGESIM Benchmarks on modelling approaches and AMEBA - Evolutionary Computation Method: simulation implementations; modelling and simulation in and for Comparison and Toolbox Development. education, simulation-based e-learning; society information and M. Corn, M. Atanasijeviü-Kunc ...... 229 membership information for EUROSIM members (Federation of European Simulation Societies and Groups). Modelling and Simulation of GMA Welding Process Editor-in-Chief: Felix Breitenecker, Vienna Univ. of Technology, and Welding Power Sources. M. Golob ...... 237 Math. Modelling and Simulation Group Inverse Simulation Methods Applied to Investigations | [email protected], | [email protected] of Actuator Nonlinearities in Ship Steering. Layout / Administration: A. Körner, A. Mathe, J. Tanzler, D. J. Murray-Smith ...... 245 C. Wytrzens, et al.; | [email protected] Print SNE: Grafisches Zentrum, Vienna Univ. of Technology, Conversion of Iterative Balance Models to Directly Wiedner Hauptstrasse 8-10, 1040, Vienna, Austria Calculating Explicit Models for Real-time Process Online SNE: ARGESIM /ASIM, address below Optimization and Scheduling Publisher: ARGESIM ARBEITSGEMEINSCHAFT SIMULATION NEWS T. Björkqvist, O. Suominen, M. Vilkko, M. Korpi ...... 257 c/o Math. Modelling and Simulation Group, Modelling of Indoor Lighting Conditions in Buildings Vienna Univ. of Technology / 101, Wiedner Hauptstrasse 8-10, for Control Design Purposes. V. Logar ...... 267 1040 Vienna, Austria; www.argesim.org, | [email protected] EUROSIM Societies Short Info ...... N1 - N8 on behalf of ASIM www.asim-gi.org and EUROSIM ї www.eurosim.info

SNE 26(4) – 12/2016 i Simulation Notes Europe Scientific Board and Authors’ Info SN E Editorial SNE Special Issue ‘Modelling and Simulation in Modern Control Engineering’ The progress in computer technology and IT has made ing Process and Welding Power Sources presents a Modelling and Simulation approaches an essential tool practical problem arising from power source control in in modern industry in several aspects, especially when it welding units. The author shows that proper electrical comes to control design. The field has been given spe- model allows controller development, ensuring steady cial attention, as the pursuit of modern systems are both and pulsed direct current welding. A problem oriented state-of-the-art design and adequate process control, study from a different field is in focus of the paper In- where modelling and simulation approaches are indis- verse Simulation Methods Applied to Investigations of pensable when effective, optimal and lean operation is Actuator Nonlinearities in Ship Steering by D. J. Mur- required. ray-Smith. The paper shows that inverse simulation The issue starts with the contribution of Š. Koreþko methods can be used to predict the ship’s rudder satura- et al. - Jadex/JBdiEmo Emotional Agents in Games with tion and rate limiting effect in terms of the maneuvera- Purpose: a Feasibility Demonstration. The authors pre- bility of the vessel. It is also shown that a two-stage in- sent a 3D game engine jMonkeyEngine, combined with verse-simulation method allows direct assessment of the Jadex agent system and JBdiEmo emotional extension difference between desired and achievable maneuvers. and their use in virtual testing grounds for development Another study was performed by T. Björkqvist et al. - of software controllers of various devices, embedded to Conversion of Iterative Balance Models to Directly them. Calculating Explicit Models for Real-time Process Op- The paper by D. Dovžan et al. - Evolving Fuzzy timization and Scheduling. The authors use a method for Model (eFuMo) method for on-line fuzzy model learning direct evaluation of the model output, instead of using with application to monitoring system presents an eFu- an iterative calculation and show its implementation on Mo method – a modelling approach based on model de- modelling of the copper production line. The method is sign and adaptation according to measured data. The au- used for process optimization and scheduling and is sig- thors show that evolving fuzzy model can be used to nificantly faster than classical modelling methods. predict sensor signals in case of their failure. Similarly, The last paper Modelling of indoor lighting condi- eFuMo approach can also be used for model-based con- tions in buildings for control design purposes, by V. trol, when real-time measurements are not accessible. Logar presents a fuzzy modelling approach to describe The paper by P. Boškoski et al. - Model-based pre- indoor lighting conditions in buildings. The model can diction of the remaining useful life of the machines deals be, due to its simplicity, used for broader environments, with simulation-based life-span prediction of shot blast- such as control design or model-based control. ing machines. The authors show that simulated estima- The scientific value of this contributions will be tion of the remaining life of the machine is satisfactory used also in the preparation of the curricula and syllabi and can be used for maintanence planning of the system. for the doctoral study in the frame of European ERAS- The paper AMEBA-evolutionary computation meth- MUS+ Project 573751-EPP-1-2016-1-DE-EPPKA2- od: Comparison and toolbox development, by M. Corn CBHE-JP entitled 'InMotion - Innovative teaching and and M. Atanasijeviü-Kunc presents the AMEBA meth- learning strategies in open modelling and simulation od and the corresponding toolbox. The method is based environment for student-centered engineering education' on evolutionary algorithms and can be used for different in which the partner University of Ljubljana, Faculty of purposes, such as system identification or control de- electrical engineering also covers the area of modern sign. Comparative results between AMEBA method and control systems in computer modelling and simulation other relevant methods are shown to demonstrate its po- engineering. tential and accuracy when identifying a nonlinear multi- The editor would like to thank all authors, who have variable system. contributed to this special issue and to the SNE Editorial The next three papers deal with more problem- Office for the support in compiling this special issue. oriented challenges, arising from practical applications. Vito Logar, University of Ljubljana, Faculty of Electrical M. Golob’s - Modelling and Simulation of GMA Weld- Engineering; [email protected]

ii SNE 26(4) – 12/2016 SNE T ECHNICAL N OTE

Jadex/JBdiEmo Emotional Agents in Games with Purpose: a Feasibility Demonstration Štefan KoreÏko*, Branislav Sobota, Peter Zemianek

Department of Computers and Informatics, Faculty of Electrical Engineering and Informatics, Technical University of Košice, Letná 9, Košice, Slovakia; *[email protected] 6LPXODWLRQ1RWHV(XURSH61( %', VWDQGV IRU EHOLHI±GHVLUH±LQWHQWLRQ D PRGHO RI '2,VQHWQ KXPDQ SUDFWLFDO UHDVRQLQJ LQWURGXFHG LQ >@ $ %', 5HFHLYHG1RYHPEHU DJHQWKDVEHOLHIVH[SUHVVLQJZKDWLWNQRZVDERXWLWVHOI $FFHSWHG'HFHPEHU 6SHFLDO,VVXH5HYLHZ DQG LWV HQYLURQPHQW GHVLUHV WKDW UHSUHVHQW VWDWHV LW ZRXOGOLNHWRDFKLHYHDQGLQWHQWLRQVWKDWSURYLGHPHDQV Abstract. The jMonkeyEngine 3D game engine, com- WRDFKLHYHWKHVWDWHV bined with Jadex agent system and JBdiEmo emotional 7KHVLPSOLFLW\RI%',LVWKHVRXUFHRIERWKLWVSRSX extension may offer a suitable toolset for effective crea- ODULW\ DQG FULWLFLVP 7KH FULWLFV SRLQW RXW WKDW %', IR tion of feature-rich virtual environments, provided that FXVHV RQ UDWLRQDO UHDVRQLQJ DQG LJQRUHV RWKHU DVSHFWV an appropriate interface, allowing to use the full poten- VXFKDVHPRWLRQV7RGHDOZLWKWKLVLVVXHLQWKHFRQWH[W tial of all included components, exists. Then, such envi- RIWKH-DGH[IUDPHZRUNZHGHVLJQHGDQGLPSOHPHQWHG ronments may profit from the jMonkeyEngine ability to -%GL(PR >@ >@ HPRWLRQDO HQJLQH ZKLFK XVHV D model and simulate the physical world and capability of PRGLILHGYHUVLRQRIWKH2&&PRGHORIKXPDQHPRWLRQV Jadex and JBdiEmo to express both rational and emo- 7KH2&&PRGHO>@FRQVLGHUVHPRWLRQVWREHUHVXOWVRI tional aspects of characters inhabiting it. One of the FRJQLWLYHSURFHVVHVDQGGLYLGHVWKHPLQWRWKUHHFODVVHV meaningful ways of utilization of such environments is to HPRWLRQVWKDWDUHUHDFWLRQVWRHYHQWVUHDFWLRQVWRDJHQWV use them as virtual testing grounds for software control- DQGUHDFWLRQVWRREMHFWV7KHYHUVLRQXVHGLQ-%GL(PR lers of various devices, embedded to them. To involve RULJLQDWHVIURP>@DQGKDVDIRUPRIDQLQKHULWDQFH real humans in the testing, they may have a form of a EDVHGKLHUDUFK\RIHPRWLRQV game, where the testing occurs during an interaction $SURPLVLQJXWLOL]DWLRQRI-DGH[DQG-%GL(PRLVLQ between the devices and players. In this paper we pre- FRPSXWHUJDPHVZKHUHWKH\FDQVLPXODWHERWKUDWLRQDO sent both the interface and the embedding on an emer- DQGHPRWLRQDOEHKDYLRUDODVSHFWVRIQRQSOD\DEOHFKDU gency simulation game called JFireEmSim2. The primary DFWHUV 13&V PRGHOHG DV DJHQWV 6XFK XWLOL]DWLRQ LV goal of the player in the game is to rescue a family from VXSSRUWHGE\-DGH[YLDDYLVXDOL]DWLRQLQWHUIDFHIRUWKH a house under fire and the controller embedded into it is M0RQNH\(QJLQH M0( ' JDPH HQJLQH ,Q >@ ZH of a simple autonomous cleaning robot. The paper de- XVHGWKHLQWHUIDFHWRGHYHORSDQHPHUJHQF\VLPXODWLRQ scribes the architecture of the game, focusing on the JDPHZKHUHWKHSOD\HU¶VJRDOLVWRUHVFXHSHRSOHIURPD interface, implementation of characters as Jadex and IODWXQGHUILUH7KHJDPHSURYHGWKDW-%GL(PRFDQEH JBdiEmo agents and embedding of the controller. It also XVHG ZLWK WKH LQWHUIDFH ZLWKRXW DQ\ PRGLILFDWLRQV EXW discusses suitability of the components for the given DOVR UHYHDOHG WKDW WKH LQWHUIDFH SURYLGHV RQO\ OLPLWHG task. DFFHVVWRM0( ,Q>@ZHLQWHQGHGWRXVHWKHSODWIRUPFRQVLVWLQJRI Introduction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

SNE 26(4) – 12/2016 195 KoreÏko et al. Jadex/JBdiEmo Emotional Agents in Games: a Feasibility Demonstration TN

*Z3WU\WRKLGHWKHLUWUXHSXUSRVHEHKLQGDQLQWHU 1.1 Jadex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µRSWLRQ¶RUµVXVSHQGHG¶$JRDOLQWKHµRSWLRQ¶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he Platform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

196 SNE 26(4) – 12/2016 T N KoreÏko et al. Jadex/JBdiEmo Emotional Agents in Games: a Feasibility Demonstration

1.2 JBdiEmo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jMonkeyEngine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

1.4 Jadex 3D visualization interface )RU-DGH[M0(LVRQO\RQHRIVHYHUDORSWLRQVWRYLVXDO L]H DJHQW EHKDYLRU 7KH RWKHU RQHV DUH WH[WXDO RXWSXW DQG'JUDSKLFV7KHDSSOLFDWLRQ;0/ILOHLVUHVSRQVL EOHIRUWKHYLVXDOL]DWLRQFRQILJXUDWLRQ+HUH'PRGHOV Figure 1: OCC to BDI mapping as implemented in DQGDQLPDWLRQVDUHDVVRFLDWHGZLWKDJHQWVDQGPRGHOV JBdiEmo. WH[WXUHV DQG SRVLWLRQV RI RWKHU REMHFWV DUH GHILQHG $ 7KHHQJLQHFRQVLVWVRIVHYHQFRPSRQHQWV GLUHFWDFFHVVWRM0(LVDOVRSURYLGHGEXWRQO\WRDVXE VHWRILWVIHDWXUHV -%GL(PR FRUH UHVSRQVLEOH IRU UHSUHVHQWDWLRQ RI 7KH PDVWHUVODYH UHODWLRQVKLS EHWZHHQ -DGH[ DQG DJHQW¶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¶VHYHQWVDQGWKHZKROH WKHWLPHVSDUHGE\XVLQJ-DGH[WRLPSOHPHQW13&VFDQ -DGH[ SODWIRUP IRU D SUHVHQFH RI RWKHU HPRWLRQDO EHORVWEHFDXVHRIWKHDGGLWLRQDOLPSOHPHQWDWLRQZRUN DJHQWV ,QWHUDJHQW HPRWLRQDO PHVVDJLQJ SODQ SURYLGLQJ PHVVDJHGHOLYHU\EHWZHHQHPRWLRQDODJHQWV 2 JFireEmSim2 Game /DQJXDJHH[WHQVLRQDOORZLQJWRGLVWLQJXLVKRUGLQDU\ 7KH FRQVWUDLQV WKH RULJLQDO YLVXDOL]DWLRQ LQWHUIDFH SXWV EHOLHIVJRDOVDQGSODQVIURPWKHRQHVZLWKDVVRFLDWHG RQGHYHORSHUVPD\UHQGHUWKHZKROHLGHDRIXVLQJ-DGH[ HPRWLRQV LHIURPWKHHPRWLRQDORQHV DQG -%GL(PR IRU 13&V LPSOHPHQWDWLRQ LQHIILFLHQW )RUWXQDWHO\WKH\FDQEHRYHUFRPHE\GHVLJQLQJDQHZ LQWHUIDFHWKDWSXW-DGH[DQGM0(LQPRUHHTXDOSRVLWLRQ

SNE 26(4) – 12/2016 197 KoreÏko et al. Jadex/JBdiEmo Emotional Agents in Games: a Feasibility Demonstration TN

7KH JFireEmSim2 JDPH )LJXUH ZKLFK LPSOH 7KHFODVV AppFRQWDLQVWKHHQWU\SRLQWRIWKHJDPH PHQWVVXFKLQWHUIDFHLVVLWXDWHGLQDYLOODJHZKHUHWKH PHWKRGmain DQGPHWKRGVUHTXLUHGE\SimpleAppli- SOD\HUKDVWZRJRDOV)LUVWKHKDVWRUHVFXHDIDPLO\RI cationVXFKDVsimpleInitAppWRLQLWLDOL]HDQGsim- IRXUIURPDKRXVHXQGHUILUHDQGVHFRQGKHKDVWRVDYH pleUpdateWRXSGDWHWKHJDPHLQHDFKJDPHORRSF\FOH D GHSUHVVHG SHUVRQ RQ D QHDUE\ FHPHWHU\ EHIRUH KH ,WVSURSHUWLHVDPRQJRWKHUVSURYLGHDFFHVVWRWKHSK\V FRPPLWVVXLFLGH7RVXFFHVVIXOO\VDYHWKHIDPLO\DOOLWV LFVHQJLQH bulletAppState KDQGOHYDULRXVREMHFWVLQ PHPEHUV KDYH WR EH WDNHQ RXW RI WKH KRXVH $SHUVRQ WKHHQYLURQPHQWVXFKDVLQGLYLGXDOILUHVLWHV SURSHUWLHV PD\UHIXVHWRIROORZWKHSOD\HUEHFDXVHRIWKHIHDURI firefirePositionsDQGfireNodes DQGLPSOHPHQW JHWWLQJEXUQHGRUGXHWRWKHSRVLWLRQLQWKHIDPLO\ ' XVHU LQWHUIDFH HOHPHQWV HJ hudControl WR VKRZ 7KH DFWXDO DSSHDUDQFH RI WKH JDPH FDQ EH VHHQ LQ KHDOWK RI WKH SOD\HU DQG 13&V App DOVR LQFOXGHV DQ )LJXUHZKHUHWKHVLWXDWLRQDIWHUVDYLQJWKHILUVWPHP REMHFW FDOOHG start ZKLFK LQLWLDOL]HV -DGH[ LQ D VHSD EHURIWKHIDPLO\LVFDSWXUHG7KHUHDUHKHDOWKEDUV UHG UDWHWKUHDG IRUDOOWKHIDPLO\PHPEHUV IDWKHU-RKQPRWKHU0DULH 7KHSOD\HURIWKHJDPHLVUHSUHVHQWHGE\WKH Play- VRQ -RH DQG GDXJKWHU -DQH LQ WKH XSSHU OHIW FRUQHU erFODVVZKLFKGHILQHVLWVJUDSKLFDODSSHDUDQFHLQFOXG 3OD\HU¶VKHDOWKDQGH[WLQJXLVKHUVWDWXVDUHVKRZQLQWKH LQJ DQLPDWLRQV SURSHUWLHV HJ KHDOWK H[WLQJXLVKHU ORZHUOHIWFRUQHU7KHSOD\HU¶VFKDUDFWHULVWKHILUHPDQ FKDUJHOHYHO DQGNH\ERDUGDQGPRXVHLQSXWSURFHVVLQJ RQ WKH ULJKW VLGH VHHQ IURP EHKLQG 7KH KRXVH XQGHU ILUHLVWKHZRRGHQRQHRQWKHOHIW7KHUHDUHWZRSHUVRQV 13&VDUHGHILQHGRQWKUHHOHYHOVILUVWWZRRIWKHP LQIURQWRIWKHKRXVHDQHLJKERULQZKLWH7VKLUWRE EHORQJLQJWRWKHFRUH VHUYLQJ WKH HYHQW DQG WKH VDYHG SHUVRQ -RH LQ EOXH CharacterDFODVVZKLFKSURYLGHVEDVLFIXQFWLRQDOL VKLUW ,Q WKH EDFNJURXQG ZH FDQ VHH DQRWKHU ILUH VLWH W\IRUDOO13&VLQWKHJDPHVXFKDVPRYHPHQWFROOL DQGDILUHPDQ7KHLUUROHLVGHVFULEHGLV6HFWLRQ VLRQUHVROXWLRQDQGDQLPDWLRQV7KHFODVVDOVRGHILQHV 6WUXFWXUDOO\WKHJDPHFDQEHGLYLGHGLQWRWKUHHFRP DQ DEVWUDFW PHWKRG act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adex/jME interface FODVVHVZLWKGDUNJUD\EDFNJURXQG7KHVHEHORQJWRWKH 7KHQHZLQWHUIDFHLPSOHPHQWHGE\FODVVHVCommunica- FRQWURO SURJUDP HPEHGGHG LQWR WKH JDPH DQG DUH GH tor DQG AgentControl LV GHVLJQHG DV XQLYHUVDO LW VFULEHGLQPRUHGHWDLOLQ6HFWLRQ GRHVQ¶WHYHQUHTXLUHWKHSDUWFRQQHFWHGWR-DGH[WREH LPSOHPHQWHGLQM0(,WVSXUSRVHLVWRNHHSWKHVWDWHRI 2.1 Game core REMHFWV UHSUHVHQWLQJ 13&V RQ WKH M0( VLGH V\QFKUR 7KHFRUHRIWKHJDPHLVGHVLJQHGDVDW\SLFDOM0(DS QL]HGZLWKEHOLHIVRIFRUUHVSRQGLQJ-DGH[DJHQWV SOLFDWLRQ ZLWK LWV PDLQ FODVV App LPSOHPHQWLQJ WKH M0(EDVHFODVVSimpleApplication

198 SNE 26(4) – 12/2016 T N KoreÏko et al. Jadex/JBdiEmo Emotional Agents in Games: a Feasibility Demonstration

DOO DJHQWV LH WKH RQHV EHORQJLQJ WR M%GL(PR DQG WKH EHOLHI shared 7KH JRDOV DQG EHOLHIV RI Marrie Joe DQG Jane DUH VLPLODU WR WKRVH RI John 7KH fami- ly_savedEHOLHILVVSHFLILFWRJohnJoeDQGJaneGRQ¶W KDYHWKHchild_savedEHOLHIEXWKDYHDGGLWLRQDOEHOLHI FDOOHGreproachedJoeDOVRKDVWKHEHOLHIjane_saved DQGJaneKDVjoe_saved 7R LOOXVWUDWH KRZ WKH DJHQWV GHILQHEHKDYLRU RI WKH 13&V OHW XV KDYH D ORRN RQ WKH IDPLO\ WR EH VDYHG :KHQ WKH JDPH EHJLQV DQ DFWLYH JRDO RI DOO IDPLO\ PHPEHUVLVZDQGHUDQGWanderPlanLVFDUULHGRXW7KLV PHDQV WKDW WKH SHUVRQV UDQGRPO\ ZDQGHU DURXQG WKH KRXVH ,I D SHUVRQ JHWV WRR FORVH WR D ILUH WKH run_from_fire JRDO EHFRPHV DFWLYH DQG RunFrom- Figure 2: JFireEmSim2 screenshot. FirePlanWULHVWRJHWWKHSHUVRQWRVDIHGLVWDQFH+RZ HYHUZKHQWKHLQWHQVLW\RIWKHfearHPRWLRQIHOWWRWKH 7KH FODVV Communicator LV LPSOHPHQWHG XVLQJ VLQJOH JRDO run_from_fire EHFRPHV WRR KLJK WKH cry JRDO WRQGHVLJQSDWWHUQWRHQVXUHWKDWRQO\RQHLQVWDQFHRILW EHFRPHV DFWLYH 7KLV LQLWLDWHV WKH CryPlan ZKLFK LVDYDLODEOHLQWKHJDPH,WVSURSHUW\agentsKROGVDOLVW PDNHV WKH SHUVRQ VWRS IRU IHZ VHFRQGV DQG SHUIRUP AgentControl RI LQVWDQFHV RQH IRU HDFK -DGH[ DJHQW DSSURSULDWH DQLPDWLRQ $QRWKHU DVSHFW WKDW FDQ DERUW 7KHCommunicatorLWVHOIMXVWDOORZVWRDGGDQGUHPRYH RunFromFirePlan H[HFXWLRQ LV WKH YDOXH RI disap- DJHQWVVRWKHZKROHV\QFKURQL]DWLRQLVLQWKHKDQGVRI provingHPRWLRQIHOWWRWKHwanderJRDOEHLQJKLJKHU WKHAgentControlREMHFWVZKLFKZULWHYDOXHVWRDJHQWV¶ WKDQ 0.7 7KHQ WKH GiveUpPlan LV H[HFXWHG ZKLFK EHOLHIVZLWKWKHLUSXWPHWKRGVDQGUHDGEHOLHIYDOXHVZLWK PHDQVWKDWWKHSHUVRQVWD\VLQSODFHDQGRQO\WKHSOD\ getBooleangetIntDQGgetFloatPHWKRGV HU¶VSUHVHQFHFDQ FKDQJH LWE\ PDNLQJ WKH stay_calm :KLOHRQWKHM0(VLGHWKHLQWHUIDFHLVDFFHVVHGYLD JRDO DFWLYH DQG StayCalmPlan H[HFXWHG :KHQ WKH WKHSURSHUW\agentRIWKHFODVVHVLQKHULWHGIURPChar- SOD\HU FRPPDQGV D SHUVRQ WR IROORZ KLP WKH fol- acter RQ WKH -DGH[ VLGH LW LV GRQH WKURXJK D EHOLHI low_player JRDODQG FollowPlanDUHDFWLYDWHGSUR FDOOHGsharedDQGSODQUpdatePlan7KHVHDUHGHILQHG YLGHGWKHFRUUHVSRQGLQJFRQGLWLRQVDUHPHWWKHPRWKHU shared IRU HDFK DJHQW 7KH YDOXH RI LV DQ LQVWDQFH RI UHIXVHV WR IROORZ XQWLO WKH FKLOGUHQ DUH VDYHG DQG WKH WKH FRUUHVSRQGLQJ AgentControl REMHFW REWDLQHG YLD IDWKHU LV WKH ODVW RQH WR OHDYH 2WKHUZLVH WKH re- WKH Communicator 7KH UpdatePlan DXWRPDWLFDOO\ ject_followJRDODQGRejectPlanDUHDFWLYDWHG XSGDWHVEHOLHIVRIWKHDJHQWHYHU\WLPHWKHYDOXHVVWRUHG LQWKHFRUUHVSRQGLQJAgentControlREMHFWDUHFKDQJHG 3 Controller in Game 2.3 NPCs behavior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¶VEHOLHIVDQGWKHWKLUGRQHJRDOV3ODQVDUHVKRZQ VZLWFKHVWRDVWDQGE\PRGH DV VHSDUDWH FODVVHV FRQQHFWHG WR FRUUHVSRQGLQJ $')V 7KHGLDJUDPGRHVQ¶WVKRZJRDOVDQGEHOLHIVFRPPRQWR

SNE 26(4) – 12/2016 199 KoreÏko et al. Jadex/JBdiEmo Emotional Agents in Games: a Feasibility Demonstration TN

Figure 3: UML class diagram showing essential part of the JFireEmSim2 game structure. 200 SNE 26(4) – 12/2016 T N KoreÏko et al. Jadex/JBdiEmo Emotional Agents in Games: a Feasibility Demonstration

7KH VSHFLILHGSRVLWLRQ LV JLYHQ DV D FRPSDVV DQJOH p2gAngleInMin DQG GLVWDQFH IURP WKH FXUUHQW SRVL WLRQ p2gDst 7KH PHWKRG HYDOXDWHV WKH VLWXDWLRQ DQG LVVXHVLQVWUXFWLRQVIRUWKHURERWWRIROORZ7KHVHLQFOXGH FRPPDQGVWRWXUQWKHURERWRQRUVZLWFKLWWRVWDQGE\ WKH RXWSXW SDUDPHWHU botOn WR VWDUW RU WR VWRS WKH FOHDQLQJ SURFHVV botCleaning DQG WKH GHVWLQDWLRQ ZKHUHWKHURERWVKRXOGJR angleInMinDQGdst 1RZDVVXPHWKDWWKHFRQWUROSURJUDPSDUWKDVEHHQ GHYHORSHGXVLQJIRUPDOPHWKRGVVXFKDV%0HWKRG>@ VRZHDUHVXUHWKDWWKHVDIHW\FULWLFDOSURSHUWLHVKROGLQ LWVLPSOHPHQWDWLRQ:KDWZHDUHQRWVXUHLVZKHWKHUWKH Figure 4: Cleaning bot sensors arrangement. GLVWDQFHVsafeDstClDQGsafeDstMovDUHVHWRSWLPDO O\ 7KH\ VKRXOG EH ODUJH HQRXJK WR SUHYHQW WKH URERW 7KH URERW VKRXOG EH DEOH WR SHUIRUP LWV MRE LQ SXEOLF IURPKXUWLQJSHRSOHEXWQRWWRRODUJHDVLWZLOOFDXVH DUHDV VR IURP WKH VDIHW\ SRLQW RI YLHZ LW LV FULWLFDO WR WKHURERWWRLQWHUUXSWLWVRSHUDWLRQWRRRIWHQ7RHVWDE SUHYHQWLWIURPKXUWLQJSHRSOH7RGHWHFWWKHPLWSRV OLVKWKHGLVWDQFHVVLPXODWLRQH[SHULPHQWVFDQEHXVHG VHVVHVDFLUFXODUO\DUUDQJHGVHQVRUDUUD\ )LJXUH 7KH VHQVRUDUUD\UHWXUQVYDOXHVnneeneesessessw 3.1 Jadex/JBdiEmo/jME as simulation wsw wnwDQG nnw7KHYDOXH nne QRUWKQRUWKHDVW LV WKHGLVWDQFHWRDQHDUHVWSHUVRQGHWHFWHGLQWKHUHJLRQ platform IURP QRUWK FRPSDVV DQJOH 0 PLQXWHV WR QRUWK HDVW :HFRQVLGHU-DGH[-%GL(PRM0(WREHDVXLWDEOHSODW 2700PLQXWHV ene HDVWQRUWKHDVW WKHGLVWDQFHLQWKH IRUPIRUVXFKVLPXODWLRQEHFDXVHRIWKHIROORZLQJUHD UHJLRQIURP2700`WR5400`DQGVRRQ,IQRSHUVRQLV VRQV GHWHFWHG LQ D UHJLRQ WKHQ WKH FRUUHVSRQGLQJ YDOXH LV 6XSSRUW IRU TXLFN FRQVWUXFWLRQ RI D YLUWXDO HQYLURQ HTXDOWRWKHPD[LPXPGLVWDQFHPHDVXUDEOHE\WKHDUUD\ PHQWZKHUHWKHURERWZLOORSHUDWHWKDQNVWRWKHHGL maxMsrblDst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up- SHOLDURERWLFVFRP 7KH M%XOOHW SRUW KDV EHHQ XVHG IRU VLPXODWLRQ SXUSRVHV DV ZHOO IRU H[DPSOH LQ >@ dateAndEvaluateDVLWVLQWHUIDFH )LJXUH 7KHILUVW IRUFHOOVDQGVXUURXQGLQJIOXLG WKUHHSDUDPHWHUV RI WKH PHWKRGVSHFLI\ ZKDW WKH URERW VKRXOGGRLQWKHJLYHQVLWXDWLRQDQGWKHUHVW nneWR 3RVVLELOLW\ WR SRSXODWH WKH YLUWXDO HQYLURQPHQW ZLWK nnw DUHUHDGLQJVIURPWKHVHQVRUDUUD\7KHSDUDPHWHU FKDUDFWHUVZLWKFRPSOH[SHUVRQDOLW\WKDQNVWRWKHLQ WHJUDWLRQ RI -DGH[ DQG -%GL(PR ,Q DGGLWLRQ WKH FRPPDQGGHILQHVZKHWKHUWKHURERWVKRXOG FKDUDFWHUVFDQEHGHYHORSHGVHSDUDWHO\DQGLQWHJUDWHG • VZLWFKWRVWDQGE\PRGHLPPHGLDWHO\ YDOXH WRWKHHQYLURQPHQWDIWHUZDUGV •JRWRDVSHFLILHGSRVLWLRQDQGWKHQFOHDQ RU •JRWRDVSHFLILHGSRVLWLRQDQGVWDQGE\

SNE 26(4) – 12/2016 201 KoreÏko et al. Jadex/JBdiEmo Emotional Agents in Games: a Feasibility Demonstration TN

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

Figure 5: Cleaning bot interface.

3.2 Simulation vs. GwP ,Q SULQFLSDO DQ HQYLURQPHQW FUHDWHG ZLWK -DGH[-% GL(PRM0(FDQEHXVHGLQWZRZD\V )RUVLPXODWLRQH[SHULPHQWV $VDJDPHZLWKSXUSRVH 7KH ILUVW FDVH LV GH IDFWR D JDPH ZLWKRXW SOD\HU ZLWK Figure 6: The second fireman representing the robot ap- proaching the fire sites, i.e. places to clean. SURSHUWLHV DQG EHKDYLRURI HQWLWLHV LQKDELWLQJ WKH HQYL URQPHQWJLYHQE\WKHLUFRGHDQGPRGHOVRQO\6RGLI 7KHFRQWUROOHUFDQEHLQWHJUDWHGDVLWLVVKRZQLQ)LJ IHUHQW VLPXODWLRQ H[SHULPHQWV XQGHU HTXDO FRQGLWLRQV XUH FODVVHVZLWKGDUNJUD\EDFNJURXQG 7KHFRQWURO FDQHDVLO\EHSHUIRUPHG OHU LWVHOI FRQVLVWV RI FODVVHV CBotController ,QWKHVHFRQGFDVHDKXPDQSOD\HULVLQYROYHGDQG CBotControllerCoreDQGProximSensors ZKLFK PD\ LQWHUDFW ZLWK DQ HQWLW\ UHSUHVHQWLQJ WKH FRQWURO KDYHEHHQJHQHUDWHGIURPIRUPDOVSHFLILFDWLRQFUHDWHG SURJUDP7KH*Z3HOHPHQWLVWKDWWKHLQWHUDFWLRQSUR DQGYHULILHGXVLQJ%0HWKRG7KHupdateAndEvaluate YLGHVGDWDHVVHQWLDOIRUHYDOXDWLRQRUDGMXVWPHQWRIWKH PHWKRGFDQEHIRXQGLQCBotController FRQWURO SURJUDP HJ ZKHWKHU FHUWDLQ YDOXHV RI 7KHFRQQHFWLRQWRWKHJDPHLVLPSOHPHQWHGYLDWKH safeDstClDQGsafeDstMovPD\HQGDQJHUSHUVRQVWKDW FODVV Robot ,W VHUYHV WKH VLPLODU SXUSRVH DV WKH FODVV FRPHFORVHWRWKHFOHDQLQJERW7KLVLQWHUDFWLRQFDQEH CharacterIRU13&VLHLWSURYLGHVYLVXDOUHSUHVHQWD FRPSOHWHO\QDWXUDOIRUH[DPSOHWKHFOHDQLQJERWFDQEH WLRQRIWKHURERW DVDILUHPDQ ,WLVDOVRUHVSRQVLEOHIRU VLWXDWHG LQ DQ DUHD ZKLFK WKH SOD\HU ZLOO YLVLW HLWKHU H[HFXWLQJ WKH updateAndEvaluate PHWKRG DQG SHU ZD\ IRUPLQJDFWLRQVDFFRUGLQJWRWKHYDOXHVLWUHWXUQV7KH LQSXWSDUDPHWHUVRIupdateAndEvaluateDUHFRPSXWHG 3.3 Cleaning bot in JFireEmSim2 IURPDFWXDOSRVLWLRQVRIWKHURERWSODFHWRFOHDQSOD\HU :KLOHWKH-DGH[-%GL(PRM0(SODWIRUPDOORZVWREXLOG DQG13&VE\WKHFODVVRelativePosition DYLUWXDOHQYLURQPHQWUHODWLYHO\TXLFNO\WKHPRUHHFR QRPLFDO RSWLRQ LV WR HPEHG WKH FRQWURO V\VWHP WR EH 4 Related Work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

202 SNE 26(4) – 12/2016 T N KoreÏko et al. Jadex/JBdiEmo Emotional Agents in Games: a Feasibility Demonstration

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eferences $WRROFKDLQVLPLODUWRRXUVKDVEHHQXVHGLQ>@WR >@ $EULDO-5The B-Book: Assigning Programs to Mean- LPSOHPHQW D VHULRXV JDPH WKDW WHDFKHV SOD\HUV DERXW ingsVWHG&DPEULGJH&83S HQHUJ\FRQVXPSWLRQ7RFUHDWHWKHJDPHWKHDXWKRUVRI >@ $KQ/YRQ*DPHVZLWKD3XUSRVHComputer >@XVHGM0(0DNH+XPDQDQG%OHQGHUWRR7KHJDPH ± GRHVQ¶WLQFOXGHDQDJHQWV\VWHPEXWXVHVFRVLPXODWLRQ >@ $SSOHZKLWH*URVVR7HWDO$PXOWLVFDOHSK\VLFVHQ YLDWKH)XQFWLRQDO0RFNXS,QWHUIDFH )0, WRLQWHJUDWH JLQHEDVHGVLPXODWLRQRIFHOOXODUPLJUDWLRQ 2015 Winter Simulation Conference'HF+XQWLQJWRQ%HDFK WKHUPDODQGSK\VLFDOPRGHOVRIDEXLOGLQJDQGDSSOLDQF &$GRL:6& HV)0,VKRXOGDOVREHFRQVLGHUHGIRUDIXWXUHYHUVLRQ >@ %UDWPDQ0(Intentions, Plans, and practical rea- RIWKHM0(-DGH[LQWHUIDFHRUDQLQWHUIDFHEHWZHHQFRQ sonVWHG&DPEULGJH0$+83S WUROSURJUDPVDQGJDPHV >@ %UDXEDFK/3RNDKU$-DQGHU.7KH-DGH[3URMHFW 7KH*Z3LGHDKDVEHHQIRUPXODWHGLQ>@DQG*Z3 3URJUDPPLQJ0RGHO,Q*DQ]KD0-DLQ&/HGLWRUV XVXDOO\ FRQWDLQ JDPHSOD\ IRFXVHG RQ VROYLQJ VSHFLILF Multiagent Systems and Applications: Volume 1:Practice SUREOHPVVXFKDVSURWHLQIROGLQJLQ)ROGLW>@RUILQG and Experience%HUOLQ+HLGHOEHUJ6SULQJHUS LQJSURJUDPORRSLQYDULDQWVLQ;\OHP>@ 2WKHU LPSOHPHQWDWLRQV RI QRQHPRWLRQDO RU HPR >@ %UDXEDFK/3RNDKU$7KH-DGH[3URMHFW6LPXODWLRQ WLRQDODUWLILFLDODJHQWVLQFRPSXWHUJDPHVH[LVWDVZHOO ,Q*DQ]KD0-DLQ&/HGLWRUVMultiagent Systems and Applications: Volume 1:Practice and Experience%HUOLQ DQG D VKRUW RYHUYLHZ DQG FRPSDULVRQ WR -DGH[ DQG +HLGHOEHUJ6SULQJHUS -%GL(PRFDQEHIRXQGLQ>@ >@ (UH]77DVVD<7RGRURY(6LPXODWLRQWRROVIRUPRGHO EDVHGURERWLFV&RPSDULVRQRIEXOOHWKDYRNPXMRFR RGHDQGSK\V[ ICRA,(((SS± 5 Conclusion >@ +DEHU$0F*LOO06DPPXW&-PHVLP$QRSHQ 7KHQHZ-DGH[M0(LQWHUIDFHSUHVHQWHGKHUHDVDSDUW VRXUFHPXOWLSODWIRUPURERWLFVVLPXODWRUAustralasian RI-)LUH(P6LPJDPHDOORZVWRXVHERWK-DGH[DQG Conference on Robotics and Automation'HF :HOOLQJWRQ M0( WR WKHLU IXOO SRWHQWLDO ZKHQ EXLOGLQJ YLUWXDO HQYL >@ .DVKLI$HWDO9LUWXDO6LPXODWLRQZLWK5HDO2FFXSDQWV URQPHQWV IRU JDPHV RU VLPXODWLRQ H[SHULPHQWV 7KH 8VLQJ6HULRXV*DPHV14th International Conference of H[SHULPHQWDO LQWHJUDWLRQ RI WKH FOHDQLQJ ERW FRQWUROOHU the International Building Performance Simulation As- DOVR SURYHG IHDVLELOLW\ RI WKH LGHD RI FRPSXWHU JDPHV sociation'HF+\GHUDEDG XWLOL]DWLRQ DV WHVWLQJ HQYLURQPHQWV ZLWK DJHQWEDVHG >@ .KDWLE)HWDO&U\VWDOVWUXFWXUHRIDPRQRPHULFUHWURYL 13&V DQG DFWLYH SDUWLFLSDWLRQ RI SOD\HUV :KLOH UDOSURWHDVHVROYHGE\SURWHLQIROGLQJJDPHSOD\HUVNat -)LUH(P6LPLQWKHVWDWHSUHVHQWHGKHUHSURYLGHVRQO\ Struct Mol Biol ± EDVLF UHDOL]DWLRQ RI WKH LGHD LW FDQ EH GHYHORSHG DQG GRLQVPE H[SHULPHQWHG ZLWK LQ VHYHUDO GLIIHUHQW ZD\V 2QH RI >@ .RUHþNRâ+HULFK72Q6RPH&RQFHSWVRI(PRWLRQDO (QJLQHIRU%',$JHQW6\VWHP14th IEEE International WKHPLVWRXVH-DGH[DQGM%GL(PRWRLPSOHPHQWPRUH Symposium on Computational Intelligence and Informat- FRPSOH[ JDPHSOD\ DQG SHUVRQDOLWLHV RI 13&V 7KH ics1RY%XGDSHVWGRL JDPHSOD\VKRXOGLQFOXGHDFWLYHLQWHUDFWLRQEHWZHHQWKH &,17, SOD\HUDQGWKHURERWIRUH[DPSOHDWDVNWRDGMXVWURERW >@ .RUHþNRâ+HULFK76RERWD%-%GL(PR±2&&0RG SDUDPHWHUVIRUPD[LPXPSHUIRUPDQFH HO%DVHG(PRWLRQDO(QJLQHIRU-DGH[%',$JHQW6\VWHP 12th International Symposium on Applied Machine In- telligence and Informatics-DQ+HUĐDQ\ GRL6$0,

SNE 26(4) – 12/2016 203 KoreÏko et al. Jadex/JBdiEmo Emotional Agents in Games: a Feasibility Demonstration TN

>@ .RUHþNRâ6RERWD%ýXULOOD3(PRWLRQDO$JHQWVDV >@ 2UWKRQ\$&ORUH*&ROOLQV$The cognitive structure 1RQSOD\DEOH&KDUDFWHUVLQ*DPHV([SHULHQFHZLWK of emotions&DPEULGJH&83S -DGH[DQG-%GL(PR15th IEEE International Symposium >@ 6WHXQHEULQN%5'DVWDQL00H\HU--&K7KH2&& on Computational Intelligence and Informatics PRGHOUHYLVLWHG,Q5HLFKDUGW'HGLWRUProc. of the 4th 1RY%XGDSHVWGRL Workshop on Emotion and Computing - Current Re- &,17, search and Future Impact >@ /RJDV+HWDO6RIWZDUH9HULILFDWLRQ*DPHV'HVLJQLQJ >@ 7RVLN70DHKOH(0$56$VLPXODWLRQHQYLURQPHQW ;\OHP7KH&RGHRI3ODQWV9th Int. Conf. Foundations IRUPDULQHURERWLFVOCEANS 14 MTS/IEEE6HSW of Digital Games$SULO)W/DXGHUGDOH 6W-RKQ VGRL2&($16 >@ 2QWDQRQ6HWDO$6XUYH\RI5HDO7LPH6WUDWHJ\*DPH $,5HVHDUFKDQG&RPSHWLWLRQLQ6WDU&UDIWIEEE Trans. on Computational Intell. and AI in Games

204 SNE 26(4) – 12/2016 SNE TECHNICAL N OTE

Evolving Fuzzy Model (eFuMo) Method for on-line Fuzzy Model Learning with Application to Monitoring System

Dejan Dovžan*, Vito Logar, Igor Škrjanc

Faculty of Electrical Engineering, University of Ljubljana, Tržaška 25, 1000 Ljubljana, Slovenia; *[email protected]

SNE Simulation Notes Europe SNE 26(4), 2016, 205-220 This enables the T-S fuzzy model to approximate DOI: 10.11128/sne.26.tn10352 virtually any nonlinear system within a required accu- Received: November 15, 2016 racy, provided that enough regions are given [2]. Accepted: December 5, 2016 (Special Issue Review) The eFuMo method is an on-line learning method that is also able to adapt models during the function- Abstract. With evermore complex system the monitoring of the system. Depending on the learning abilities, ing and fault detection is becoming a crucial part of con- the on-line fuzzy-identification methods can be divided trol systems. They allow fast and effective fault diagno- into: Adaptive methods (e.g., ANFIS [3], GANFIS [4], sis and can decrease the cost of system maintenance. rFCM [5], rGK [6]), where the initial structure of the Modelling of processes plays a crucial part when design- fuzzy model must be given. The number of space par- ing a monitoring system. In this paper an on-line ap- titions/clusters does not change over time, only the pa- proach for modelling of fuzzy model is presented (Evolv- rameters of the membership functions and local mod- ing Fuzzy Model - eFuMo). As demonstrated in the paper, els are adapted; Incremental methods (e.g., RAN [7], the method can be used in the design of model based SONFIN [8], SCFNN [9], NeuroFAST [10], DENFIS fault detection system. [11], eTS [12], FLEXFIS [13], PANFIS [14]), where only adding mechanisms are implemented; Evolving methods (e.g., SAFIS [15], SOFNN [16], GAP-RBF Introduction [17], EFuNN [18, 19], D-FNN [20], GD-FNN [21], ENFM [22], eTS+ [23], ENFM [22], FLEXFIS++ Increasing demands of productivity and reliability call [24], AHLTNM [25], SOFMLS [26]) which, besides an for extending the ability of a common SCADA systems adding mechanism, implement removing and some of with the monitoring and fault detection systems. There them also merging and splitting mechanisms. More on are several approaches for designing the fault detection evolving methods can be found in [27] and [28], where system. In our paper the monitoring system is based concepts and open issues regarding these methods are on a process model. The model is based on a evolving presented. fuzzy model method (eFuMo). The presented method The paper is organized in the following order. First, is able to build Takagi-Sugeno fuzzy model (TS) from the eFuMo learning method is described, next the mon- scratch, starting with one cluster and a local model. The itoring problem is given followed by results and conclu- TS fuzzy models are a powerful practical engineering sions. tool for modelling and control of complex systems. They expand and generalize the well-known concept of gain scheduling. They utilize the idea of linearization 1 eFuMo Structure in a fuzzily defined region of the state space. Due to the fuzzy regions (clusters), the nonlinear system is The eFuMo method has two types of mechanisms for decomposed into a multi-model structure consisting of identifying the fuzzy model: the adaptation algorithm linear local models [1]. and the evolving mechanisms. The first is responsible for parameter adaptation, such as cluster centers and lo-

SNE 26(4) – 12/2016 205 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

μ can be calculated as in equation 4 (c is the number i of existing clusters), either based on rFCM [5] (equation 5), rGK [6] (equation 6) or Mahalanobis distance (equation 7). ⎧ ⎪ 1 x ( ) = v = ,..., ⎪ 2 if f k i;i 1 c ⎨⎪ ( ) η−1 c di k ∑ = ( ) μ ( )= j 1 d j k i k ⎪ ⎪ 1ifx f (k)=vi ⎩⎪ 0ifx f (k)=v j;i = j (4) Figure 1: The eFuMo top scheme. cal models’ parameters; the second is responsible for . structure update: adding, removing, merging and split- T 0 5 d (k)= x (k) − v (k) x (k) − v (k) (5) ting of clusters. A central decision logic (CDL) decides i f i f i which type of mechanism will be used at current sam- . T 1 − 0 5 ( )= x ( ) − v ( ) (F ) p F 1 x ( ) − v ( ) ple. The block scheme is presented on Figure 1. In di k f k i k det i i f k i k the following subsection, the adaptation and evolving (6) mechanisms will be presented and at the end the CDL will be described.

0.5 1.1 Adaptation mechanisms ( )= x ( ) − v ( ) T F−1 x ( ) − v ( ) di k f k i k i f k i k In order to build the TS fuzzy model clusters and local (7) linear models must be identified. Adaptation mechanisms are responsible for identifying clusters’ and local models’ parameters and for their adaptation. To parti- To get the area of cluster influence the fuzzy covari- F tion input-output data space recursive clustering algo- ance matrix i is calculated. The recursive equation for F rithm is used and for identifying the local models’ pa- i is the following: rameters the fuzzy recursive least squares is used. ( − ) μ ( )η F ( + )=γ si k 1 F ( )+ i k D ( ) i k 1 c i k Fi k si(k) si(k) Space partitioning. For data space partitioning, D (k)=(x(k) − v (k))(x(k) − v (k))T . (8) the cluster centers and fuzzy covariance matrix must be Fi i i calculated. The centers are adapted with the following where γ is the forgetting factor. To be able to calculate equation: c the Gustafson-Kessel clustering distance (equa- vi(k + 1)=vi(k)+Δvi(k) (1) tion 6) the inverse and determinant of fuzzy covariance η μi(k) x f (k) − vi(k) matrix must be calculated. The recursive equation for Δvi(k)= (2) si(k) the inverse matrix is obtained by using the Woodbury η v matrix identity lemma. The equation is following: where is fuzziness factor, i is the center position vector x is clustering vector, μ is membership degree ( ) B f i [F ( + )]−1 = 1 si k [F ( )]−1 − of the current clustering vector to the i-th cluster also i k 1 i k (9) γc si(k − 1) C called the firing degree of the i-th cluster and si(k + 1) is the sum of past membership degrees / firing levels of B =[F ( )]−1 D [F ( )]−1 i k Fi i k (10) the i-th cluster: s (k − 1) − C = γ i + dT [F (k)] 1 d (11) η c μ ( )η Fi i Fi si(k)=λcsi(k − 1)+μi(k) . (3) i k = x − v ( ). dFi f i k (12) where λc was introduced as a forgetting factor to en- able the adaptation of centers. The membership degrees

206 SNE 26(4) – 12/2016 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

The determinant is obtained using determinant lemma and the y is the process output. The firing levels are (equation 13): calculated on the input space. Usually the methods use Gaussian functions equation 19 or equation 20: − det(A + uvT)=(1 + vTA 1u) det(A). (13) (x −v )2 fk ik − η F 2 m i , The recursive equation for determinant calculation is: μi(k)=e k k k = 1,2,...p − 1 i = 1,2,...,c − ( − ) p p 1 si k 1 β = μ ( ) det(Fi(k + 1)) = γc det(Fi(k))(1 + A) i ∏ i k si(k) k=1 (14) (19) μ ( )η 1 i k T −1 A = d [Fi(k)] dF , (15) γ ( − ) Fi i − 2 c si k 1 Dik η βi = e 2 m i = 1,2,...,c, where p is the number of rows/colmuns of fuzzy co- (20) − D2 = x (k) − v T F 1 x (k) − v . variance matrix. The detailed derivations of equations ik fin iin iin fin iin are given in [6]. The eFuMo method implements the η F where m is the overlapping factor usually set to 1, i , method for stopping the cluster parameters adaptation k k is diagonal element k of fuzzy covariance matrix, x f if the clusters firing level is below a certain user defined k and vi are the k-th element of clustering vector and k-th threshold β . This prevents clusters, that are far from k cuttrh element of i-th cluster center vector, respectively. The current clustering vector, to converge to that area. The F x iin is the input fuzzy covariance matrix, fin is the clus- clusters’ firing levels 4 that are below the threshold are v tering vector containing only the input variables and iin set to zero. The rest of the firing levels are then normal- is the cluster center in an input space. The obtained fir- ized. ing levels are then normalized:

Local models’ parameters identification. β β = i i = 1,2,...,c. (21) i ∑c β Each cluster has a linear local model, that is valid in that k=1 k area. The output of the local model is calculated as: One can also use the same equation for firing level cal- ( )=θ T [ x ( )T ]T , ymi k i 1 k k (16) culation as with clustering algorithm. However, with Gussian functions the transitions between local models x ( ) θ T where k k is the regression vector and i are the local (clusters) are more smooth. model parameters. The regression vector is usually the When building the simulation model, the model pa- input part of the clustering vector: rameters can be identified more accurately using the in- T T strumental version of least squares [30]. The instrumen- x f (k)=[xk(k) y(k)] , (17) tal variable adaptation algorithm for equation 22 can be where y is the process output. However unlike many ex- written as: isting on-line fuzzy identification methods the eFuMo method allows the clustering vector to be different than x ( )=[ x ( )T ]T the regression vector. e k 1 k k T T xe (k)=[1 x (k) ] The eFuMo has different fuzzy least squares based m km identification methods included ([12], [22], [5] and 1 β P (k)x (k)xT (k)P (k) P (k + 1)= P (k) − i i em e i i λ i λ + β xT (k)P (k)x (k) [29]). Usually best results are obtained using local r r i e i em fuzzy weighted least squares presented in [12]: θ ( + )=θ ( )+P ( )β x ( ) ( ) − xT ( )θ ( ) i k 1 i k i k im em k y k e k i k T T xe(k)=[1 x (k) ] (22) k 1 β P (k)x (k)xT (k)P (k) P (k + 1)= P (k) − i i e e i i λ i λ + β xT ( )P ( )x ( ) (18) r r i e k i k e k x ( ) T where em k is the regression vector where the delayed θ (k + 1)=θ (k)+P (k)β xe(k) y(k) − x (k)θ (k) i i i i e i process outputs were replaced with model outputs and θ β x ( ) where i is the cluster index, is the vector of local im is the membership degree of vector fm k , which model’s parameters and β is the firing level of cluster is the clustering vector, where delayed process outputs

SNE 26(4) – 12/2016 207 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

! !

" Figure 3: The parameter adaptation algorithm. Adding mechanism. This is one of the most important mechanisms. It adds new clusters to the fuzzy model structure and improves the fuzzy model perfor- # mance. In the literature, there are several different conditions of adding new clusters based on model out- Figure 2: The clustering algorithm. put error, distance of current sample to existing cluster and ε-completeness which is based on current samples membership degree to existing clusters. were replaced with model outputs: In [32] (DFKNN) a cluster adding is based on Eu- clidian distance to the existing cluster centers and the x ( )=[ ( − ) ... ( ) ( − ) ... ( − )] f k u k n u k y k r y k 1 change of variance that the new sample brings to the x ( )=[ ( − ) ... ( ) ( − ) ... ( − )] fm k u k n u k ym k r ym k 1 closest cluster. The distance and variance change must (23) be greater than the predefined threshold. A new cluster is added if a certain number of sequential samples sat- where ym is the model output and y is the real output. In isfy this condition. In [11] (DENFIS) adding is based both cases the dead zone for adaptation can be consid- on an Euclidian distance. If the distance of current ered [31]. sample to closest cluster is grater than two times the The adaptation procedure can be represented by the threshold a new cluster is added. In [20] (D-FNN) and diagrams as shown on figure 2 and figure 3. In figure [21] (GD-FNN) adding is based on model error and dis- 2 the clustering procedure is represented and in figure tance of new sample vector to closest cluster. If both are 3 the local model parameters identification algorithm is grater than a user defined threshold the cluster is added. presented. The threshold is decreasing with time. In [17] (GAP- RBF) and [15] (SAFIS) a new cluster is added if the model error and distance of the current sample to ex- 1.2 Evolving mechanisms isting clusters is over a threshold. They calculate the decrease in error if current sample would be taken for a To upgrade the fuzzy model structure evolving mecha- new cluster. If the decrease is large enough new cluster nisms, such as adding and removing the clusters is im- is created. In [18, 19] (EFuNN) the adding is based on plemented in the eFuMo method. sensitivity calculated based on normalized fuzzy differ-

208 SNE 26(4) – 12/2016 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

ence distances. The eTS method [12] adds a new cluster when the potential of current sample is higher than a potential of existing clusters and if it is distanced enough from the nearest cluster. In [33] (NFCN), [22] (ENFM),

[8] (SONFIN), [9] (SCFNN), [16] (SOFNN) adding is based on ε-completeness principle. In [13] (FLEXFIS) the adding is based on distance and vicinity quotient. In practice, the distance conditions work best. Therefore, the eFuMo implements two conditions for (a) Component adding (b) Mahalanobis adding adding: the distance conditions and the consequent distance. distance. samples conditions. Both conditions must be satisfied in order for a new cluster to be added. The consequent Figure 4: Different adding distance conditions. samples condition is to prevent a new cluster being created based on outlier sample. This condition means that several consecutive samples must satisfy the distance condition before a new cluster is added. The condition dimensional space. When a cluster is added, the param- is explained in [32]. eters of the cluster must be initialized. The center of The distance adding condition is based on a normal- a new cluster is set at the position of current clustering ized distance. There is an option of choosing the com- vector. The fuzzy covariance matrix is initialized as di- ponent distances or Mahanalobis distance. The normal- agonal matrix where the distances to closest cluster are ized component distances are calculated as: considered. The diagonal elements are defined as:

|x (k) − v | 2 f j i j dij dij = , j = 1,...,pi= 1,...,c (24) f = − , (27) new jj η (ε ) kn fi jj 2 m ln β ( ) ε where x f j k is the j-th element of clustering vector, where β is a user defined constant, normally set to . vi j is the j-th component of i-th cluster center, p is the 0 15. If the distance dij is smaller than standard de- length of clustering vector, c is the number of clusters, viation ( fi jj), then this diagonal element is equal to fi jj is the j-th diagonal element of i-th cluster’s fuzzy a diagonal element of the closest cluster’s fuzzy covari- = covariance matrix and kn is the user defined constant, ance matrix ( fnew jj fi jj). usually set to 2. When using normalized Mahanalobis distance the normalization vector is formed from diag- The first cluster is added at the position of the first onal elements of fuzzy matrix: clustering vector. Its fuzzy covariance is initialized in the similar manner considering the input-output space s =[ ... ]T , boundary and expected number of clusters: inorm fi11 fi22 fipp (25) d = max(x ) − min(x ), j = 1,...,p (28) The normalized distance is then calculated as: max j j j

((x ( ) − v )T F−1(x ( ) − v ))0.5 where dmax is an expected range of j-th element of = f k i i f k i j di − (26) clustering vector. The influence zone of the j-th com- norm k (sT F 1s )0.5 n inorm i inorm ponent is then calculated as: With the first condition, a cluster can be added is any dmax of the component distance equation 24 is larger than 1. d = j , j = 1,..., p (29) in f luence j 2 c The same component distance must be larger than 1 for all existing clusters. With the second condition a cluster where c is the expected number of clusters. The di- can be added if the distances equation 26 to all clusters agonal j-th element of fuzzy covariance matrix is then are larger than 1. Figures 4(a) show the possible adding calculated as: space for the component distance conditions and figure d2 4(b) for the Mahanalobis distance condition. Both fig- σ 2 = − in f luence j = ,..., j j 1 p (30) ures show the possible adding space (orange) for two 2ηm ln(εβ )

SNE 26(4) – 12/2016 209 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

σ 2 The fuzzy covariance is built with j as: [21] (GD-FNN), where sensitivity index is introduced. ⎡ ⎤ In [15] (SAFIS) an equation is introduced to estimate σ 2 0 ··· 0 the error change if a certain cluster is removed from ⎢ 1 ⎥ ⎢ 0 σ 2 ··· 0 ⎥ the structure. If this change is small, the cluster is re- F = ⎢ 2 ⎥ i . . . . (31) moved. In [16] (SOFNN) removing is based on optimal ⎣ . . .. . ⎦ brain surgeon approach [35, 36]. In [37] (Neural gas) 00··· σ 2 p the clusters are removed based on their age. All clusters The parameters of new local model can be initialized that are older than an user defined age are removed from using weighted mean: the structure. In [38, 39] (exTS) the removing is based on cluster’s support and cluster’s age. The clusters are ∑c ω θ i=1 i j i j removed based on support-age ratio. Similar conditions θ + = (32) i 1 j ∑c ω are introduced in +eTS [23], where also the utility con- i=1 i j dition is added. This condition is based on the ratio of where i is the index of cluster and j is the parameter sum of firing levels and age of cluster. The threshold ω index. Weights i j can be equal to normalized firing values are defined with standard deviation and mean levels of clusters, or equal to normalized firing levels of values of the ratios. In general, this is not adequate, clusters combined with parameters variances: since there is usually small number of clusters; therefore, using standard deviation and mean value are not 1 really representative. In +eTS also minimal existence ωi = βi , (33) j σ 2 condition is introduced. With this condition, a newly Pi jj created cluster must gather a certain amount of support where σ 2 is the j-th diagonal element of least squares samples in a certain time period after creation. If the Pi jj covariance matrix of i-th cluster. gathered support is lower than a predefined threshold, the cluster is removed from the structure. In [18, 19] (EFuNN) the removing is based on cluster’s age and Removing mechanism. It is meant to remove old sum of cluster’s firing levels. In [32] (DFKNN) remov- clusters and clusters created based on outliers. In ing is based on minimal support and time period. If the eFuMo method, this mechanism is not so important as cluster has lower support than an user defined threshold the method incorporates the forgetting factors that en- the cluster is deleted. The cluster is also removed if in sure the adaptation of the structure to the new data. certain time period after cluster’s creation, no support However, it may happen that a cluster is created in a sample is assigned to it. partition of input-output space that doesn’t have much samples and is not very important for the model accuracy. This mechanism ensures that these clusters are The proposed eFuMo method has two conditions for removed from the model structure. In literature, differ- removing: A minimal existence condition and support- ent ideas are presented. In [34] the cluster is removed age ratio condition. The minimal existence condition if in a certain time the cluster doesn’t receive any sup- is the same as in [23]. It simply removes clusters that port sample. Cluster receives a support sample if it has in certain period after creation (kdelay) don’t receive greater firing level than other clusters. This might be a enough support samples (Nsi). The time period (kdelay) problem with industrial processes, where it might hap- and support threshold (Nstrh) are user defined constant pen that the process is in one working point for a longer usually set to 20 and 10, respectively. The support-age period of time. In this case, other clusters, that describe ratio condition is based on clusters’ supports Nsi nor- different working points, might be removed from the malized with clusters’ age (equation 35). Cluster with ε structure. In [20], [21], [17], [15] in [16] (D-FNN, GD- the ratio lower than a percent of mean ratio is deleted. FNN, GAP-RBF, SAFIS in SOFNN) the removing is Age ai is defined as a number of samples from the clus- based on model error. In [20] (D-FNN) the error re- ter’s creation ki and current sample k: duction ratio is introduced. The amount of error, that a a = k − k (34) certain cluster brings to the overall model error is calcu- i i lated. If this is small, the cluster is considered as redun- = Nsi . dant and is therefore deleted. Similar concept is used in Sni (35) ai

210 SNE 26(4) – 12/2016 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

Both conditions for removing can be written as: old and the decay constant. The current threshold is calculated as: IF < ε ( ) Sni mean Sn = ( (− / ), ), OR Nsi < Nstrh AND k > ki + kdelay (36) etrh max emaxexp N T emin (39) THEN remove i-th cluster. where etrh is the current threshold, emax is the maximal error threshold, emin is the minimal error threshold, N Splitting mechanism. It is in our case meant for and T are the number of samples that are used for error fine tuning the evolving fuzzy model. It can add clus- calculation and decay constant, respectively. ters in input-output space, where the output model er- The positions of the split clusters are calculated us- s ror is higher than predefined threshold. The concept of ing diagonal elements (vector inorm ) of the fuzzy covari- splitting was used in the on-line incremental learning ance matrix. of Gaussian Mixture Models in [40], where the Cher- vi1 = vi + 0.5si noff bound is used and in [41], where fidelity measure norm (40) v = v − . s is used. It is argued in [42] that these methods are slow i1 i 0 5 inorm and don’t produce good results. Therefore they propose an integrating a joint incremental on-line split-and- where i is the index of the cluster that is split. The new merge scenario, that should overcome under- and over- center positions can also be calculated using the sin- clustered partitions. The splitting is based on a BIC gular value decomposition as in [43]. The fuzzy co- value. The BIC is a combination of Gaussian density variance matrix, support and sum of past membership function function and cluster overlapping. The clusters degrees are set to half of their original value for both that are split are found using trail and error procedure. clusters. The time of cluster creation is for both clusters In [10] (NeuroFAST) clusters are split based on mean initialized as the creation time of the original cluster. squared error (MSE). The error is checked every P step. The cluster that has the highest MSE and is at least P- Merging mechanism. There are two types of times activated is split. merging algorithms implemented in eFuMo: supervised The eFuMo’s splitting mechanism is based on rel- and unsupervised. In literature different concept of ative model error, that clusters gather over time. The merging techniques can be found. In [32] (DFKNN), error is updated every time the splitting mechanism is the center positions are monitored. If the centers are called and the current sample doesn’t satisfy the dis- converging to the same area the clusters are merged. tance adding condition. First the relative model error is The used similarity measure is based on samples mem- calculated: bership degrees and is similar to the correlation between |y (k) − y(k)| e(k)= m , (37) clusters firing levels. It is presented in detail in [44]. In 3.4σy [18] (EFuNN), the merging is done based on clusters’ firing levels correlation. The method merges neigh- where y is the real output and ym is the model output. borhood clusters, where after merging the total radius The σy is calculated by CDL block and represents current standard deviation of the process output. The error does not exceed the predefined maximal radius. In [22] is then divided among the existing clusters and added to (ENFM),the clusters are merged if the membership de- the previous error: gree of the first cluster to the second and vice versa is higher than a predefined threshold. In [16] (SOFNN) ( )= ( − )+β ( ), clusters are merged if the cluster centers of the two clus- esumi k esumi k 1 ie k (38) ters are the same. The possibility of using similarity where βi is the firing level of i-th cluster. The splitting measure from [45] is mentioned. In [46] (FLEXFIS+), mechanism checks the cluster with the highest error. If the merging based on membership function intersec- its support from the last change in cluster number till tions is proposed and the overlapping index is calcu- now is higher than a threshold (usually set to 20) and lated. If this index for the two clusters is higher than its error normalized with N (number of samples used a predefined threshold and the angles between the local to calculate the error) is larger than a threshold value, models’ parameters are small the clusters are merged. the cluster is split. The error threshold is set by the The eFuMo unsupervised merging is based on most user, specifying the maximal and minimal error thresh- commonly used principle of merging. It merges clusters

SNE 26(4) – 12/2016 211 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning that are close together. The similarity and the vicinity weighted mean: of the two clusters are measured by the normalized dis- ωi θi + ωk θk tance: θ = j j j j j = 1,...p, (48) new j ω + ω i j k j 2 =(v −v )T −1(v −v ), , = ,..., = . dik i k Fi i k i k 1 ci k (41) where weights ω are the cluster supports Ns combined d2 with a variance of the parameters. d = ik (42) normik T −1 The supervised merging considers the prediction 2s F si inorm i norm model error. The supervised merging has three different The distances are calculated only for clusters that have measures to detect the clusters that could be merged to- higher support from last change in cluster number than gether. It uses angles between local models’ parameters an user defined threshold (usually set to 20 for both val- (angle merging condition), correlation between clusters ues). The clusters are considered for merging if both firing levels (correlation merging condition) and dis- normalized distances dnormik and dnormki are shorter than tance ratio (distance ratio merging condition). Only the predefined threshold εβ : clusters that gathered higher support and sum of past membership degrees than a predefined threshold can be < − (ε ) dnormik ln β (43) considered for supervised merging. The correlation coefficient is calculated based on monitoring of firing lev- If this criterion is satisfied, the distance ratio is checked: els and their products βij(k)=βij(k − 1)+βi(k)β j(k), βii(k)=βii(k − 1)+βi(k)βi(k) and is calculated as: min(d ,d ) − normik normki < 1 kdmerge (44) max(d ,d ) βij normik normki C (k)= (49) ij β 0.5β 0.5 ii jj if the ratio is above the user defined threshold kdmerge (usually 10 percent) clusters are merged. If the coefficient Cij(k) is above user-defined threshold (usually set to 0.9) the clusters i and j are considered for merging. The parameters of new cluster are initialized as a The distance ratio criterion for merging is similar weighted mean. The fuzzy covariance as proposed in than with the unsupervised merging. The distance ratio [22]: is calculated as: 1 1 T −1 F = (( 3 + 2 + 2)F + dik = (vi − vk) det(Fi) p F (vi − vk) new 3 Nsi 2Nsi Nsk NsiNsk i i (Nsi + Nsk) (50) |1 − min(dik,dki)| 3 2 2 K +(Ns + 2Ns Nsi + NskNs )Fk+ d k k i max(dik,dki) 2 2 T +(Ns Nsk + NsiNs )(vi − vk)(vi − vk) ) i k The clusters are considered for merging if the distance (45) ratio Kd is lower than an user defined threshold (usually The new center is calculated as: 0.05) and the correlation coefficient is at least half of the threshold defined for correlation merging condition. Ns v + Ns v v = i i k k The angle merging criterion is based on local mod- new + (46) Nsi Nsk els’ angles. First the parameters are normalized. The al- In the same manner a the new sum of past membership gorithm sweeps all local models’ parameters to find the degree is calculated. New support of the cluster 47 and vector of the largest absolute value of parameters. Then time of creation are calculated as weighted mean where the parameters of local models are normalized with this vector. The angles for the two clusters for all parame- weights are sum of past membership degrees (si, sk): ters are calculated: + = Nsisi Nsksk . Nsnew (47) α = |arctan(θ ) − arctan(θ )| (51) si + sk ijk ik jk The local linear model parameters are calculated as where k is the parameter index. The clusters are consid-

212 SNE 26(4) – 12/2016 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

α = ,.., ered for merging if all angles ijk , k 1 r, where r is the number of local model’s parameters, are below the user-defined threshold (usually set to 2 degrees) and the correlation coefficient is at least half of the threshold defined for correlation merging condition. After the eFuMo identifies the possible merging pairs with the correlation, angle and distance ratio merging conditions it then checks the local models for the error:

T x1 =[1,u¯1,...,u¯p−1] = |θ T x − θ T x | e1 i 1 j 1 p = |θ ( + σ ) − θ ( + σ )| e2 ∑ ir x1r 2 ur−1 jr x1r 2 ur−1 = r 1 (52) p = |θ ( − σ ) − θ ( − σ )| e3 ∑ ir x1r 2 ur−1 jr x1r 2 ur−1 r=1 3 = 1 e . σ ∑ er 10 2 y r=1 whereu ¯ is the mean value of a certain input variable Figure 5: Scheme of the CDL. σ σ ur−1 is its standard deviation, y is the standard deviation of the process output, p−1 is the number of inputs, θ j and i are the cluster indexes i is the i-th cluster’s local 1.3 Central decision logic θ model parameter vector and ir is the r-th parameter of the i-th local model. The CDL is responsible for proper flow of the opera- The pair that has the lowest error and the error is be- tions. It controls the calls to evolving mechanisms and low the threshold is merged. The center of the merged adaptation mechanisms. It also calculates the mean and cluster is positioned in the middle between maximum standard deviations of the inputs and output of the pro- and minimum border of both clusters: cess, that is identified with eFuMo. The scheme of the CDL block is shown on figure 5 and the sub-blocks are d1 = vi − v j shown on figure 6. v = v + (d )s i i sign 1 inorm The input to the eFuMo identification method are x x v = v − sign(d )s clustering vector ( f ), regression vector ( k), output of j j 1 jnorm (53) v − v the process (y) and number of current sample (i). The i j d2 = CDL block first checks the current sample number (i)to 2 the sample number when the last change in cluster num- v = v + d new j 2 ber was made and the user defined time delay. If the sum of these two values are smaller than a current sam- The fuzzy covariance matrix and support of a new ple number, the evolving mechanisms may be called. merged cluster is initialized as a sum of both clusters’ Otherwise the CDL skips the call to evolving mecha- fuzzy covariance matrices and supports, local model pa- nisms. rameters are initialized as a mean of both local models’ The CDL first calls the adding mechanism, then the parameters and the creation time is initialized to the cre- removing mechanism, follows the supervised merging ation time of the oldest cluster. The sum of past mem- mechanism and unsupervised merging mechanism and bership degrees is initialized to the max sum of past at the end the CDL calls the splitting mechanism. If one membership degrees of both clusters. of the mechanisms changes the cluster number other evolving mechanisms that follow are not called and the eFuMo continues with the adaptation algorithm.

SNE 26(4) – 12/2016 213 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

, + #

! ! " " ' ( ) ) ! ! *

!" #$% " #&

(a) Block A (b) Block C

$ $ # #

! ! ! ! " " " "

Figure 6: The CDL scheme blocks. The c is the number of clusters, c_trh is the maximal allowed number of clusters, age_trh is the age threshold for minimal existence condition and last_change is the sample number when the last change in cluster number occurred. 214 SNE 26(4) – 12/2016 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

the oxygen concentration. First order local models are used. The inputs were selected by a backward selection. The idea is to detect the error in the process output based on the inputs. The outputs of the FDS are yso ft (k) and alarm(k). The output alarm(k) indicates the pres- ence of the fault in the measured signal (alarm(k)=1: fault detected). The output yso ft (k) is the process output Figure 7: Scheme of the MBBR. with the removed fault. If there is no fault detected the output yso ft (k) is equal to the process output y(k). If the fault is detected, the output yso ft (k) is calculated based The CDL algorithm is also responsible for calculating on a fuzzy model that describes the proper relations be- the variance and mean of the input variables and output. tween the input signals and the monitored signal. σ 2 The variance is calculated on line by the following The FDS determines the fault based on the internal equation: fuzzy model of the signal relations. For monitored sig- 1 nal, three models are kept in the FDS’s memory: a full σ 2(k)= (k − 1)(σ 2(k − 1)+x¯(k − 1)2)+x(k)2 − evolving fuzzy model, an adaptive fuzzy model (param- x k eters of clusters and local models are adapted) and a 1 2 − ((k − 1)x¯(k − 1)+x(k)) fuzzy model with fixed parameters that holds the in- k2 (54) formation about the last good known parameters. The learning of the fuzzy models is delayed for 200 samples. where x is the variable andx ¯ is the mean of it, calculated The delay was introduced for future research to cope as: with slower faults. The data sample is used for learning 1 x¯(k)= ((k − 1)x¯(k − 1)+x) (55) if there was no fault detected. For each sample and each k model the relative prediction error is calculated. The If the splitting is enabled, the CDL also calls the error calculated error (its absolute value) is assigned to the update algorithm. The algorithm updates a cluster error model. The prediction error assigned to the fuzzy model equation 38. This algorithm is only called if the current is combined with the simulation error, which is calcu- data sample doesn’t satisfy the distance adding condi- lated periodically on every 200-th sample using the 200 tion. The CDL also calculates the clusters’ firing levels samples in the buffer. The prediction error is also used products used to calculate the correlation coefficient 49. for learning the prediction-error fuzzy model. Namely, each model that describes the signal relations is accom- 2 Monitoring Example panied by the error model. The error model is used to calculate the allowed difference between the estimated 2.1 Monitoring system idea and measured signals. For estimating the sensor output during the failure, the model with the lowest assigned The monitoring system that includes the evolving fuzzy error is used. model was tested on measured data from a pilot waste- The adaptive and fixed model structure and parame- water treatment plant, shown in figure 7. The pilot plant ters can be replaced when a cluster is added or removed consists of two anoxic reactors, two aerobic reactors from the evolving fuzzy model’s structure. Before the and an additional reactor, where the water is collected number of cluster changes, the error of each model is before returning as an internal recycle or passing down checked. If the evolving model has the smallest error, to the settler. To ensure the homogeneity, the waste wa- the adaptive and fixed model structure is replaced by the ter is mixed by mixers in the anoxic reactors and by air evolving model’s structure. In addition their error mod- flow in the aerobic reactors. In this example the mon- els are replaced. The simplified diagram of the proce- itoring of oxygen concentration in anoxic reactors will dure is shown in figure 8. be done. The monitoring system is based on Takagi- The variances denoted as σ_evolving, σ_adaptive Sugeno (TS) fuzzy model that estimates the relations and σ_ fixed are calculated from the error model: between the input and output variables. The oxygen c concentration is estimated from the air flow, the tem- σ = β , ∑ i Fir,r (56) perature in the reactor and the previous measurement of i=1

SNE 26(4) – 12/2016 215 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

! "" "#$ "#$ "#$ "

"#!(($ "#!(($ "#!(($%%( ,+-- "#$%%( + "#&$%%( * %

(./ ! (./ ! ! (./

"#$%&' Figure 9: Eﬀect of sensor calibration. "#$%('

"#$%%&' ) +"#$ )* sor was manually reduced by the operator, causing the + +"#$% ' "#$ FDS to report an error. It can be seen that the shapes +"#$%' of the estimated and measured outputs are practically the same. However, due to an offset of the signal the FDS detects the error. To automatically turn off such Figure 8: Scheme of the FDS for a subprocess. alarms, an additional algorithm was implemented to the FDS. This algorithm is turned on when a new alarm is detected. With this procedure the algorithm starts to where Fir,r is the last diagonal element of the error fuzzy model’s cluster i. This element represents the variance calculate the variances of the estimated output, the mea- of the error. As seen in figure 8, the alarm is raised sured output and the variance of their difference when if the difference between the estimated output and the the alarm is raised. The idea behind this solution is that measured output is higher than the maximum allowed the variance of the estimated and measured output (if difference. Note that the alarm is turned off when for at they are only shifted) should be higher than the variance least 30 consecutive samples the difference is below the of their difference, under the assumption that the model defined threshold for turning off the alarm. To ensure a used for estimating the output is not biased and the pro- smooth transition from the estimated output to the mea- cess output changes (there is an excitation present). The sured output, when the alarm is turned off a filter was variances are calculated recursively with equations 55 implemented that calculated the output of the FDS as: and 54. When the variance of the difference between the estimated and measured outputs falls under the vari- ((30 − kalarm)ymodel + kalarmy) ance of both, the estimated and the measured outputs y = , (57) so ft 30 the raised alarm is turned off. The algorithm starts to check this condition after the alarm is present for some where k is the number of samples from the sam- alarm time (in our case 300 samples). The algorithm is turned ple when the condition for turning the alarm off was off when, for at least a certain number of consecutive reached. The maximum number of k is 30 and its alarm samples (in our case 100), the variance of error is below value is reset to 0 every time a new alarm is raised. the model and process variance. The algorithm is also turned off if its maximum functioning time is reached. 2.2 Detecting the false alarms due to manual calibration 3 Results and Discussion Manual tuning and offset repairs of the oxygen concentration signal is performed every few months. This is The presented idea was tested on real data. To esti- seen on the upper graph in figure9. The drift of the sen- mate the performance of the system during a sensor’s

216 SNE 26(4) – 12/2016 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

Figure 11: Alarm signal and number of clusters over the experiment.

Figure 10: Oxygen-concentration fault detection. Estimation Error Concentration O2 NDEI 0.488 malfunction a failure was simulated on a known part of min. abs. 2.83e-5 g/m3 the data. Note that the duration of the simulated fault max. abs. 1.347 g/m3 was exaggerated in order to test the system. The sim- avg. abs. 0.189 g/m3 ulated faults lasted for about 7000 samples (around 39 hours). Usually, the faults last from about a few min- signal range 2.72 g/m3 utes up to 6 hours. The settings of the evolving method were obtained based on trail and error. The fault was min. rel. 1.04e-5 simulated between the samples 35000 and 43000. The max. rel. 0.495 whole experiment is shown in figure 10. The first 8000 avg. rel. 0.0695 data points were used for the initial learning of the fuzzy faulty samples [103] 35 − 43 model. The learning was performed using the eFuMo method. The alarm signal and the number of fuzzy model clusters are shown in figure 11. Besides the sim- Table 1: Estimation error during the simulated fault. ulated fault, the system also detected some faults that were not added to the signals. These faults were caused detects the alarm. Because the signal is shifted after by sudden spikes in the monitored signals and therefore the fault, the alarm is still present. The alarm is finally the detection of the fault seems justified. turned off at sample 11500, when the measured signal Even though the estimated signal is not entirely cov- comes into the allowed difference zone and stays there ering the measured signal, we believe that the estima- long enough for the fuzzy model to adapt itself to the tion accuracy is still good enough. The error between signal shift. On the lower graph in figure 9, the false- the measured and estimated signal during the fault is alarm detection was implemented. It can be seen that given in Table 1. This table also includes the NIDE the signal shift is successfully detected and the alarm is index, the minimum, maximum and mean absolute er- turned off more quickly than without the implemented ror, the signal range for the faulty samples, the mini- false-alarm detection algorithm. mum, maximum and mean relative error, and the sam- On figure 12, the course of variances are shown. The ples where the fault was simulated are given. variance of the difference (between the estimated and As can be seen on the upper graph in figure 9, the measured signal) falls under the measured signal’s vari- manual tuning creates an offset of the measured signal, ance very quickly. This is partly because the initial fault resulting in the detection of a fault. At around sam- of the measured signal is included in the variance cal- ple 8400 a real fault occurs, which then quickly van- culation. The variance of the difference falls under the ishes. Later on the measured signal is shifted. The FDS variance of the estimated signal at sample 9475. With

SNE 26(4) – 12/2016 217 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

References

[1] Johanson TA, Murray-Smith R. Operating regime

approach to nonlinear modeling and control. UK: Taylor Francis. 1981. [2] Hwang CL, Chang LJ. Fuzzy neural-based control for nonlinear time-varying delay systems. IEEE Trans Syst Man Cyber part B. 2007;37(6):1471–1485. [3] Shing J, Jang R. ANFIS : Adaptive-Network-Based Fuzzy Inference System. IEEE Trans Syst Man Cyber. Figure 12: The course of variances. 1993;23(3):665–685. [4] Azeem MF, Hanmandlu H, Ahmad N. Structure this, the conditions for overriding the original alarm are identification of generalized adaptive neuro-fuzzy met. The last alarm based on the output differences is inference systems. IEEE Trans on Fuzzy Syst. 2003; raised at sample 9740. Therefore, the variance proce- 11(5):666–681. dure is switched off at sample 9840. The procedure suc- [5] Dovzan D, Skrjanc I. Recursive fuzzy c-means cessfully detected the signal offset caused by the man- clustering for recursive fuzzy identification of ual calibration. time-varying processes. ISA Transactions. 2011; 50(2):159 – 169. [6] Dovzan D, Skrjanc I. Recursive clustering based on a 4 Conclusion Gustafson-Kessel algorithm. Evolving Systems. 2011; 2:15–24. In this paper an evolving fuzzy model method for on- line learning of fuzzy models was presented. The [7] Patt J. A resource allocating network for function method is useful when dealing with nonlinear time- interpolation. Neural Computat. 1991;3(2):213–225. varying processes. The method was used in an exam- [8] Juang CF, Lin CT. An on-line self-constructing neural ple of fault detection system. The presented results fuzzy inference network and its applications. IEEE show that the approach can be successfully used for Trans on Fuzzy Syst. 1998;6(1):12–32. such tasks. The only issue of the method and all such [9] Lin FJ, Lin CH, Shen PH. Self-constructing fuzzy methods is in its tuning. There are a number of parame- neural network speed controller for permanent-magnet ters that need to be tuned. Their tuning highly depends synchronous motor drive. IEEE Trans on Fuzzy Syst. on a problem and require an expert to tune them. Fur- 2001;9(5):751–759. ther research will be focused on lowering the number of [10] Tzafestas SG, Zikidis KC. NeuroFAST: On-line tuning parameters and on self tuning of the method. neuro-fuzzy ART-based structure and parameter learning TSK model. IEEE Trans Syst Man Cyber Part B. 2001;31(5):797–802. [11] Kasabov NK, Song Q. DENFIS: Dynamic Evolving Neural-Fuzzy Inference System and Its Application for Time-Series Prediction. IEEE Trans on Fuzzy Syst. 2002;10(2):144–154. [12] Angelov PP, Filev DP. An approach to on-line identification of Takagi-Sugeno fuzzy models. IEEE Trans Syst Man Cyber Part B. 2004;34(1):484–497. [13] Lughofer E, Klement EP. FLEXFIS: A Variant for Incremental Learning of Takagi-Sugeno Fuzzy Systems. In: Proceedings of The 2005 IEEE International Conference on Fuzzy Systems. Reno, Nevada, USA. 2005; pp. 915 – 920. [14] M Pratama PA S G Anavatti, Lughofer E. A novel incremental learning machines. IEEE Transactions on

218 SNE 26(4) – 12/2016 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

Neural Networks and Learning Systems. 2014; concepts and open issues. Information Sciences. 2013; 25(1):55–68. 251:22 – 46. [15] Rong HJ, Sundararajan N, Huang GB, Saratchandran P. [28] Lughofer E. Evolving Fuzzy Systems – Methodologies, Sequential adaptive fuzzy inference system (SAFIS) for Advanced Concepts and Applications, vol. 266 of nonlinear system identification and prediction. Fuzzy Studies in Fuzziness and Soft Computing. Springer. Sets and Syst. 2006;157(9):1260–1275. [29] de Jesús Rubio J. EVOLVING INTELLIGENT [16] Leng G, Prasad G, McGinnity TM. An on-line SYSTEMS Methodology and Applications, chap. algorithm for creating self-organizing fuzzy neural STABILITY ANALYSIS FOR AN ONLINE networks. Neural Networks. 2004;17:1477 – 1493. EVOLVING NEURO-FUZZY RECURRENT [17] Huang GB, Saratchandran P, Sundararajan N. A NETWORK, pp. 173–199. Hoboken, New Jersey: John Recursive Growing and Pruning RBF (GAP-RBF) Wiley & Sons. 2010;. Algorithm for Function Approximations. In: [30] Blažicˇ S, Škrjanc I, Gerkšicˇ S, Dolanc G, Strmcnikˇ S, Proceedings of The Fourth International Conference on Hadjiski MB, Stathaki A. Online fuzzy identification Control and Automation (ICCA-03). Montreal, Canada. for an intelligent controller based on a simple platform. 2003; pp. 10 – 12. Engineering Applications of Artificial Intelligence. [18] Kasabov NK. Evolving fuzzy neural networks for 2009;22:628–638. supervised/unsupervised on-line knowledge-based [31] Dovžan D, Škrjanc I. Predictive functional control learning. IEEE Trans Syst Man Cyber Part B. 2001; based on an adaptive fuzzy model of a hybrid 31(6):902–918. semi-batch reactor. Control Engineering Practice. [19] Kasabov N. Evolving fuzzy neural 2010;18(8):979 – 989. networks-Algorithms, applications and biological [32] Hartert L, Mouchaweh MS, Billaudel P. A motivation. In: Methodologies for the Conceptation, semi-supervised dynamic version of Fuzzy K-Nearest Design and Application of Soft Computing. Japan. 1998; Neighbours to monitor evolving systems. Evolving pp. 271–274. Systems. 2010;1:3–15. [20] Wu S, Er MJ. Dynamic fuzzy neural networks - A novel [33] Lin CT. A neural fuzzy control system with structure approach to function approximation. IEEE Trans Syst and parameter learning. Fuzzy Sets and Syst. 1995; Man Cyber Part B. 2000;30(2):358–364. 70:183–212. [21] Wu S, Er MJ, Gao Y. A fast approach for automatic [34] Crespoa F, Weberb R. A methodology for dynamic data generation of fuzzy rules by generalized dynamicc mining based on fuzzy clustering. Fuzzy Sets and fuzzy neural networks. IEEE Trans on Fuzzy Syst. Systems. 2005;150:267–284. 2001;9(4):578–594. [35] Hassibi B, Stork DG. Second-order derivatives for [22] Soleimani-B H, Lucas C, Araabi BN. Recursive network pruning: Optimal brain surgeon. Advances in Gath-Geva Clustering as a Basis for Evolving Neural Information Processing. 1993;4:164–171. Neuro-Fuzzy Modeling. Evolving Systems. 2010; [36] Leung CS, Wong KW, Sum PF, Chan LW. A pruning 1(1):59 – 71. method for the recursive least squared algorithm. [23] Angelov P. EVOLVING INTELLIGENT SYSTEMS: Neural Networks. 2001;14:147–174. Methodology and Applications, chap. Evolving [37] Fritzke B. A growing neural gas network learns Takagi-Sugeno Fuzzy Systems From Streaming Data topologies. Adv Neural Inform Processing Syst. 1995; (eTS+), pp. 21 – 50. New Jersey: Wiley. 2010;. 7:845–865. [24] Lughofer E. Learning in Non-Stationary Environments: [38] Asif M, Angelov P, Ahmed H. An Approach to Methods and Applications. Springer. Real-time Color-based Object Tracking. In: Evolving [25] Kalhor A, Araabi B, Lucas C. An online predictor Fuzzy Systems, 2006 International Symposium on. model as adaptive habitually linear and transiently 2006; pp. 86 –91. nonlinear model. Evolving Systems. 2010;1(1):29–41. [39] Angelov P, Zhou X. Evolving Fuzzy Systems from Data [26] Rubio J. SOFMLS: Online self-organizing fuzzy Streams in Real-Time. In: Evolving Fuzzy Systems, modified least-squares network. IEEE Transactions on 2006 International Symposium on. 2006; pp. 29 –35. Fuzzy Systems. 2009;17(6):1296–1309. [40] Hall P, Hicks Y. A method to add Gaussian mixture [27] Lughofer E. On-line assurance of interpretability models. Tech. rep., Department of Computer Science, criteria in evolving fuzzy systems - achievements, new University of Bath. 2004.

SNE 26(4) – 12/2016 219 DDovžanet al. Evolving Fuzzy Model Method for on-line Fuzzy Model Learning

[41] Declercq A, Piater J. Online learning of gaussian Model Identiﬁcation. Fuzzy Systems, IEEE mixture models-atwo-level approach. In: Proceedings Transactions on. 2011;PP(99):1. of the 3rd international conference on computer vision [44] Frigui H, Krishnapuram R. A robust algorithm for theory and applications VISAPP. Funchal, Portugal. automatic extraction of an unknown number of clusters 2008; pp. 605–611. from noisy data. Pattern Recognition Letters. 1996; [42] Lughofer E. A dynamic split-and-merge approach for 17:1223–1232. evolving cluster models. Evolving Systems. 2012;DOI: [45] Wang WJ. New similarity measures on fuzzy sets and 10.1007/s12530-012-9046-5. on elements. Fuzzy Sets and Systems. 1997;85:305–309. [43] Skrjanc I, Hartmann B, Banfer O, Nelles O, Sodja A, [46] Lughofer E, Bouchot JL, Shaker A. On-line elimination Teslic L. Supervised Hierarchical Clustering in Fuzzy of local redundancies in evolving fuzzy systems. Evolving Systems. 2011;(2):165–187.

220 SNE 26(4) – 12/2016 SNE T ECHNICAL N OTE

Model-based3rediction ofthe5emaining8seful /ifeofthe 0achines PavleBoškoski1 ,BoštjanDolenc1,2, BojanMusizza1,ÐaniJuričić1,3 1DepartmentofSystemsandControl,JožefStefanInstitute,Jamova39,1000Ljubljana,Slovenia; SDYOHERVNRVNL#LMVVL 2JožefStefanInternationalPostgraduateSchool,Jamova39,1000Ljubljana,Slovenia; 3 UniversityofNovaGorica,Vipavska13,5000NovaGorica,Slovenia

4/&4JNVMBUJPO/PUFT&VSPQF4/& performance and cost efficiency. To accomplish this %0*TOFUO goal, systems for on-line and non-destructive condition 3FDFJWFE/PWFNCFS monitoring (CM) have to be employed to timely alert "DDFQUFE%FDFNCFS 4QFDJBM*TTVF3FWJFX about the onset and location of fault in the early stage [1]. Indeed, the degradation of an asset usually goes Abstract. Accurate anticipation of the remaining useful life (RUL) of a machine is becoming mandatory for through a distinct incipient phase with some noticeable efficient exploitation of the asset and avoiding the un- indicators, which provide advanced warning about on- planned downtimes. This should be achieved without set of failure. However, what the operators and main- extra investments in additional sensors and processing power. In this paper we present an approach to the RUL tenance people indeed want to know is when to stop prediction of a shot blasting machine by using record- the machine and take accommodation actions. Reliable ings from inexpensive vibrational sensors. The key idea estimate of the remaining useful life (RUL) becomes consists in (i) employing generalised Jensen-Rényi divergence (JRD) as a measure of change in the vibrational indispensable. pattern and (ii) associating JRD with the abrasive wear In spite of significant advances in condition moni- in rotor blades. It is essential to note that these two toring in the last decade in terms of methodology and show monotonic relationship. Hereupon, a simple hidden Markov model with stochastic inputs and JRD as out- key enabling technologies, yet no massive use in indus- put is proposed. The hidden states of the model are up- trial sector has been witnessed to date [2]. There are dated on-line by means of Kalman filter. Prediction of the several reasons for that, including (i) (still) relatively remaining useful life is done by executing Monte Carlo simulations on the updated model and evaluation of the high cost of the design and commissioning, especially first passage time of the JRD. The approach is success- when domain specific solutions have to be adopted and fully validated experimentally by running the machine up (ii) the fact that traditional approaches require addi- to failure, hence allowing for naturally evolving wear pro- tional instrumentation (e.g. for rotational speed) to be gression and breakdown. implemented hence rising the cost. Compared to CM, predicting RUL is by far more Introduction difficult problem. Only limited success has been achieved in special cases like in aeronautics and defence Stable and anticipative condition of process equipment, systems. The problem is notoriously demanding for high availability and reliability, along with product several reasons: (i) data about overall useful life from quality are key factors that allow companies to stay similar items of equipment are seldom available, (ii) competitive on the market. However, wear, material knowledge about degradation, i.e. wear mechanisms is stress and environmental factors cause equipment to incomplete and (iii) comprehensive knowledge of oper- fail. The problem occurs if that happens unexpectedly, ating history, disturbances and past maintenance actions since the consequence can be partial or total break- is usually unavailable. down of a production line, destroyed equipment and The objective of the design approach presented be- even catastrophes. low is to comply with the three main requirements: (i) Migrating towards more cost effective condition- to come up with signatures sufficiently robust to varia- based and predictive maintenance (instead of sticking tions in the operating conditions; (ii) to set up the alarm to the outdated concepts of reactive and periodic main- threshold the required prior knowledge should be min- tenance) has become a way to raise the overall process imal (meaning that all the required information should

SNE 26(4) – 12/2016 221 Boškoski et al. Model-based 3rediction of the 5emaining useful /ife of the 0achines TN be extracted from data in fault-free operation) and (iii) controlled by setting the wheel, whilst its shape changes to perform condition monitoring (CM) using minimal both in width and length, thus forming a range of shot number of sensors thus making the method both broadly flow that hits the surface of object under treatment. applicable and financially viable. In this paper we propose an approach to the RUL prognosis based solely on vibrational records. The idea is to exploit the relationship between the degradation phenomena in the material, the remaining life and characteristic information patterns in measured signals. The latter are obtained by statistical signal processing of signals from vibrational sensors in a way to accomplish monotonous dependance with the level of machine degradation. Evaluation of the vibrational features is based on statistical analysis of the envelope of the generated vibration [3]. State of health of the machine is determined from change in the vibrational signature by calculating the "distance" between initial and current signatures. That is achieved by evaluating the generalised Jensen-Rényi divergence of the vibrational features. Since the degradation is stochastic process, we will exploit hidden Markov models to describe the degradation phenomena. The states of the models are updated on-line and then used to simulate propagation of the future degradation and hence evaluate the probability density function of the remaining useful life. The concept of RUL estimation above is applied to a shot blasting machine. Figure 1: The shot blasting machine and illustration of the The rest of the paper is organised as follows. Section principle of operation. 2 introduces the problem related to the degradation of the machines during operation. Simple process model for RUL prediction, complemented with the health in- The problem addressed in this paper concerns abra- dex, is presented in Section 3. Experimental results are sive wear of the rotor blades. Abrasive grains transverse highlighted in Section 4. The paper ends up with con- the blade from center to the periphery and their kinetic cluding remarks. energy increases due to centrifugal forces of the rotat- ing blade. Hence the abrasive grains scrap the surface of the blade thus forming "micro-chips", i.e. small pieces of material removed from the blade surface. With in- Shot%lasting0achine creasing number of the operating cycles the wear in- Shot blasting machines are widely used in the process of creases, gradually leading to the damaged blade, which surface cleaning where contaminants from the surface can eventually break and cause downtime. of castings are removed in order to prepare the metal The outlook of a new blade at the beginning of the parts for further finishing like, for example, painting, process and near failure is given in Figure 2. The prob- coating or mechanical treatment. lem is that it is not possible to accurately judge the level In shot blasting machines (Figure 1) small shots of of wear on the basis of the number of cycles. Therefore abrasive material are fed to the turbine blades where the it is of utmost interest for the operators to have an indi- shots form a stream flowing along the blade length. De- cator on the level of wear in non-intrusive manner, i.e. pending on the actual arrangement of the separating ro- without interrupting the blasting process. tor and the sleeve, the flowing stream will be roughly Inference on the level of damage is done on the basis uniform on the blades’ width and length. As soon as of signal analysis from vibrational sensor mounted on the stream of shots leaves the blades, its direction is the housing of the machine close to the rotor bearing.

222 SNE 26(4) – 12/2016 T N Boškoski et al. Model-based 3UHGLFWLRQRIWKH5HPDLQLQJXVHIXO/LIHRIWKH0DFKLQHV

.2 Jensen-Rényidivergence The generalised Jensen-Rényi divergence (JRD), de- w noted by JRα serves to quantify the dissimilarity among n pdfs P1,...,Pn. It reads: n n w JRα (P1,...,Pn)=Hα ∑ wiPi − ∑ wiHα (Pi) Figure 2: Turbine blade at the beginning of the operation i=1 i=1 (left) and at the end of the useful life (right). (1) ∑n = where i=1 wi 1 and Hα is the Rényi entropy: 1 α Vibrationalfeaturesand Hα (P)= ln ∑ p (x). (2) 1 − α health index x∈D α ∈ [ , ] .1 Featureextractionfromvibrationalsignal with 0 1 . The selection of weights wi in (1) is in principle ar- Faults in the rotational machines affect the inner pat- bitrary. If wi are selected uniformly i.e. wi = 1/n, the terns of vibrational signals referred to as features [3]. divergence reaches maximal value [5]. JR divergence By tracking the way these features evolve over time, it quantiﬁes shared information among n random vari- is possible to perform sufﬁciently accurate RUL predic- ables. If they are identical, i.e. P1 = P2 = ...= Pn, tion. the divergence is zero. Wear in a turbine blade of the machine gradually ×10−3 results in increased imbalance of the rotor system. Vi- brations resulting thereof can be viewed as the result of P 4 1 excitation, caused by rotor movement, on the machine P2 P3 eigen-structure. The resulting spectrum contains char- 3 )

· · x acteristic components at the frequencies m nblades frot ( p 2 where m ∈ 1,2,..., nblades is the number of blades and frot is rotational speed. By applying the narrow-band 1 ﬁltering around the characteristic frequency we get a narrow-band stochastic signal whose energy (or enve- 0 012 3 4 5 678 lope) is Rice distributed. x Sampling of vibrational signal is performed at high (a) frequency during short measurement sessions with an JRw(P , P ) interval of 2 hours between two consecutive sessions. 0.09 α 1 2 JRw(P , P ) Changes in the probability distribution function (pdf) 0.08 α 1 3 0.07 JRw(P , P ) are characterised by calculating the "distance" between α 2 3 )

j 0.06

P the current pdf and the reference one obtained when the , i 0.05 P ( machine is in nominal (healthy) state. Among several w α 0.04

possible metrics that can be used to describe this dis- JR 0.03 tance, we suggest the so-called f -divergence measures, 0.02 more precisely the generalised Jensen-Rényi (JR) diver- 0.01 gence [4]. The rationale is simple. Instead of compar- 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 ing two distributions, we compare two ensembles of dis- α tributions, one from fault-free reference condition and (b) the other from current condition. The strength of this Figure 3: (a) Pdfs three random signals, and (b) pairwise JR approach lies in the fact that comparing only two distri- divergence as a function of α. butions is subjected to considerable fluctuations, which make final decision making difficult. The usability of the JR divergence concept can be described with a simple example. Figure 3(a) shows

SNE 26(4) – 12/2016 223 Boškoski et al. Model-based 3UHGLFWLRQRIWKH5HPDLQLQJXVHIXO/LIHRIWKH0DFKLQHV TN three pdfs of Rician family. By considering the pairwise JR divergence with uniform weights, the relation (1) becomes: 0.3 )

x 0.2 1 ( w p JRα (P1,P2)=Hα (P1 + P2) (3) 2 0.1 1 − (Hα (P1)+Hα (P2)), 20 Measurement session 10 2 15 8 10 6 4 P P 5 2 where 1 and 2 are pdfs of interest. As shown in 0 x Figure 3(b), the JR divergence corresponds to the dis- (a) similarity between corresponding pdfs. −3 Figure 3(b) additionally shows the effect of the val- ×10 ues of the parameter α. Low value of α ≈ 0 emphasizes 3.5 Exp. λ =2 Exp. λ =4 3 dissimilarity among pdfs in the lower part of the range Univ.

)

∈ ( , ) i of random variable (approximately x 6 8 ) where 2.5 P pdfs do not differ much, hence low divergence values. , ...,

2 2

≈ α ≈ . P In the middle region (x 4, 0 2) the pdfs differ the , 1

P 1.5 most, hence the highest values of JR divergence. Fi- ( w α nally, α ∈ (0.6,1) captures the region of the bulk prob- JR 1 ability masses and the divergence drops in a relatively 0.5 linear manner. 0 5 10 15 20 Measurement session .3 Theroleofweightswi (b) To allow tracking the changes in pdfs, the exponential Figure 4: The evolution of JR divergence after measurement weights w are suggested in this paper. The weights w i i sessions. Note that all pdfs are equal except the are calculated using the exponential function of the fol- pdf #11. (a) Pdf’s of the simulated signals lowing form: ,..., λ associated to the measurement sessions 1 21 − i wi = C · e n (4) (b) JR divergence. Up to i < 11 there is no where λ is sensitivity parameter, n is the number of pdfs dissimilarity in the distribution, w (P ,P ,...,P )= (1),i = 1,2,...,n and C is normalising constant. One hence JRα 1 2 i 0. can easily see that (4) reduces to the uniform weighting for λ → 0 and n → ∞. The influence of weights wi on JR divergence can gated level of health either of a component or machine be illustrated by a simple simulated example. The sim- as a whole. In the case of shot blasting machines the ulation consists of 21 Gaussian pdfs with one heaving health H is perfect when the machine is new, hence μ = significantly different as shown in Figure 4(a). The H 0. With evolving abrasive processes on the blades, w(P ,P ,...,P ) more and more surface material is removed, which re- JR divergence is calculated as: JRα 1 2 i , = ... sults in increased vibrations. The Jensen-Rényi diver- i 1 21. gence is viewed as an appropriate metric that reflects The rate of change in JR divergence is condi- tioned with the selection of weights as shown in Fig- the change in vibrational pattern caused by the level of ure 4b. The most notable increase is observed if uni- wear in turbine blades. To find the relationship, life- form weighting is applied, i.e. w = 1/n [5], while ex- long experiments have been run in which machine oper- i ponential weighting delays the impact. ation was periodically interrupted by operators who performed invasive measurement of the blades volume. All the time during operation, the vibrations were regularly .4 Healthindex measured. The most important result of the experiment The concept of health index is widely used in system is the finding that between JRD and the extent of dam- condition monitoring and serves to describe the aggre- age (equivalent to removed volume of blade material)

224 SNE 26(4) – 12/2016 T N Boškoski et al. Model-based 3UHGLFWLRQRIWKH5HPDLQLQJXVHIXO/LIHRIWKH0DFKLQHV

there exists a monotone relationship. This is indicated tions, one can safely assume that any particular change in Figure 5. Consequently, one can adopt the health in- in the vibration’s signature in the lower frequency band = JRDk dex to be equal to the normalised JRD, i.e. Hk JRD∗ (<2 kHz) is due to mass loss and is therefore directly where JRD∗ stands for JRD when the machine turbine related to the blades’ condition. is considered worn out. Note that health index H does not rise monoton- .2 HiddenMarkovmodel ically all the time, but in the period approximately [30,100] it slightly decreases. Such a behaviour looks The Archard’s law (5) describes mass loss due to the illogical given the fact that the machine should get more blade interaction with single shot particle. During nor- and more worn with new operating cycles. The expla- mal operation a number of particles travel along the nation lies in the fact that at the begining of the opera- blade’s surface. During the interval of time [tk−1,tk] the tion, the machine is not perfectly balanced. If we take loss of volume can be written as: into account that abrasive processes are not the same on Δ = · ˜ · ˜ , all the blades, then asymmetry in abrasion slightly cor- Vk k Ak Lk (6) rects the position of the center of gravity, hence result- where A˜ is the cumulative contact area of the shot parti- ing in lower vibrations and apparently improved condi- k cles and L˜ is the cumulative traversed distance. These tion. Such a situation changes as soon as abrasion pro- k two quantities are results of stochastic processes and, gresses. Then asymmetrical wear in the blades results consequently, also ΔVk is stochastic process. Therefore, in increased imbalance and consequently raised vibrations. the total volume loss at k + 1 would be:

Vk+1 = Vk + ΔVk. (7) Stochastic0odelof$brasive :ear Due to surface changes, the contact area and the traversed length are expected to change over time. There- .1 Abrasivewear fore, based on (7), we can assume that the volume loss Δ Vk is a process defined by the stochastic variable de- The key mechanism of deterioration of condition of the fined on the set of non-negative real numbers. To con- turbine blades is abrasive wear [6]. Each time a shot sistently model such a process, several options are at particle enters the turbine, it travels along the blade’s disposal as for example, gamma or Weibull distribu- length. Along that path it removes a small layer of the tion. The problem is that in such a case recursive up- blade material of volume δV according to the Archard’s dates can be done only by numerical techniques. A law way around is to assume that the increments ΔVk fluctu- δV = k · δA · δL, (5) ate around some mean value μ. The size of fluctuation where k is the wear coefficient, δA is the contact area can be described by a normally distributed white noise 2 δ wμ ∼ N (0,σμ ) such that σμ μ. From here it follows and L is the length of the path traversed by the shot that particle on blade’s surface. In the ideal case, when all the blades were identi- Δ − Δ = − cal, the mass removed from each blade would be the Vk Vk−1 wμ,k wμ,k−1 same. Thus the center of gravity would stay at the ro- and consequently one can write tational axis, which means negligible vibrations. How- ever, due to irregularities in the particle size, angle of ΔVk = ΔVk−1 + wΔV,k entry and variations of the blade’s microstructure, there 2 are minute variations in the mass removed from each where wΔV,k ∼ N (0,2σμ ) blade. As a result, the generated vibrations tend to in- Hence we get a state-space model with states Vk and clude amplitude modulations that depend on the num- ΔVk. The problem now is that none of the states is avail- ber of blades and the rotational speed. Therefore, the able through on-line sensor reading. This can be sorted intensity of these sidebands can be directly correlated out by replacing the volume Vk by health index Hk, with the removed volume of the blade material due to which is calculated on-line from acquired vibrational abrasive wear. Since there is no other source of vibra- records.

SNE 26(4) – 12/2016 225 Boškoski et al. Model-based 3UHGLFWLRQRIWKH5HPDLQLQJXVHIXO/LIHRIWKH0DFKLQHV TN

Figure 5: The relation between fault progression and evolution of health index (JR divergence): (a) at the beginning of the experiment, (b) in the middle of the experiment and (c) at the end of the experiment.

Hence the resulting state space model reads as fol- 2. Prediction step lows = xˆk|k−1 Axˆk|k−1 Vk+1 11 Vk 0 = T + = + (8) Pk|k−1 AAPPk−1|k−1A Q ΔV + 01 ΔV wΔ , k 1 k V k xk+1 A xk wk 3. Update step: calculate the system output vector y based on calculated JRD and then update the moments The measurement equation that relates system states of state probability distribution function and computable health index Hk reads

( ) =[ ] V k + = T ( T + )−1 Hk 10 nk (9) K k Pk|k−1C CCPPk|k−1C R ΔVk = + ( − ) C xˆk|k xˆk|k−1 K k yk Cxˆk|k−1 =( − ) ∼ N ( ,σ 2) Pk|k I K kC Pk|k−1 where nk 0 n is white noise uncorrelated with wk. 4. When the next measurement session appears set = + k k 1 and go to step 2. .3 Kalmanﬁter .4 RUL predictor The states of the discrete model (8) can be effectively estimated using the Kalman ﬁlter approach [7, 8]. The Having an updated model at a given measurement ses- unknown states are updated at each measurement session k one can simulate the possible future trajectories sion resulting in the moments of the posterior distribu- of the state space model (8) by Monte Carlo approach. ∼ N ( , ) tion of system states xk xˆk|k Pk|k as follows Using realisations of random processes of noise terms = 1. Initialisation step: set the estimates xˆ0|0 wΔ,k+s, nk+s, s > 0 is is possible to calculate the cor- , = T = σ 2 x¯0 P0|0, Q wwww , R n from data obtained through responding trajectories of the state vector xk+s and the life-long experiments on similar machines. predicted health index Hk+s. Based on that one can eas-

226 SNE 26(4) – 12/2016 T N Boškoski et al. Model-based 3UHGLFWLRQRIWKH5HPDLQLQJXVHIXO/LIHRIWKH0DFKLQHV

ily calculate the distribution of first passage time, i.e. .2 RULprediction the time s∗ at which the health index H crosses the upper bound H∗. The evolution of the calculated health index is evaluated according to the Archard’s law, as described in Section 4.1. Based on results of Kalman filtering, the Resultsofexperiments trajectories of future states, and consequently health index, are calculated from a set of noise realisations. The RUL estimation algorithm was evaluated on a shot The RUL prediction based on the first 100 measure- blasting turbine in real operating environment. The ments is shown in Figure 6. At each time moment, the blades were subjected to 400 operational hours spread Kalman filter provides estimates of the posterior prob- over a period of 4.5 months. Vibration signals were ability distribution of the state vector xk+s and the out- acquired during 10 seconds long measurement sessions putyk+s. To come to the distribution of the actual RUL every two hours while the machine was in full opera- we perform Monte Carlo simulations of the output tration. In that period, three visual inspections were per- jectoris. The distribution of the RUL can be evaluated formed after 10 hours of operation, at the 120th hour from the histogram of first passage times for each sim- and at the end of the experiment. Vibrations were mea- ulation run. As shown in Figure 6, the proposed un- scented Kalman filter (UKF) provides left skewed RUL sured on the bearing housing nearest to the turbine with estimates. The 3σ confidence interval is sufficiently sampling frequency of 10 kHz. narrow and corresponds to the actual evolution of the health index. .1 Evolutionofthehealthindex For proper assessment of the model’s accuracy, the RUL estimates should be plotted versus a theoretically The health index was calculated as JR divergence ac- expected RUL. Typically, the theoretical RUL is ex- cording to (1) with unifirm weights wi. First 20 hours pected to be a linear function with gradient -1. This of operation were used as a reference point. The evolu- is shown in Figure 7. Note that during the first 2/3 tion of the health index is shown in Figure 5. of the operational life the RUL prediction is not reli- As shown in Figure 5, in the initial phase, the health able. However, in the last third of the life, predictions index values were near zero. This is an indication that become rather accurate meaning that roughly 2 months the energy distribution of the newly observed vibra- before the blades are fully worn the operators have reli- tion is very similar to the initial ’fault-free’ distribution, able information, which could be used to plan the main- hence the minimal JR divergence. tenance actions at a convenient occasion in a way that The first significant increase of the JR diver- do not disturb regular production (for example, during gence occurred around the 30th hour of operation. Af- a weekend or night shift). ter the initial increase the JR divergence gradually de- creased. As said, this effect can be attributed to the run- in phase of the turbine blades. th The onset of fault is visible at the 80 hour of operation. At this point the degradation of the blade condition commenced. This is clearly indicated by the increase in the calculated JR divergence. The observed degradation was confirmed by the visual inspection performed at 120th hour, as shown in Figure 5. The degradation trend is kept almost constant until the last fifth of the run i.e., around the 130th hour. The calculated health index surpassed the threshold at the 180th hour. The operation was halted at the 190th hour with the blade th condition corresponding to the estimated health index, Figure 6: RUL prediction at the 100 measurement as shown in Figure 5.

SNE 26(4) – 12/2016 22 Boškoski et al. Model-based 3UHGLFWLRQRIWKH5HPDLQLQJXVHIXO/LIHRIWKH0DFKLQHV TN

Conclusion 220 200 Theoretical Estimated RUL 180 The proposed feature based on JR divergence is shown 160 to be sufficiently sensitive to perform accurate condition 140 monitoring of shot blasting machines. Furthermore, it 120 is shown that the evolution of the JR divergence can be 100 directly related to the removed mass from the turbine’s 80 blades due to abrasive wear. As a result, the evolution 60 40 profile can be described through Archard’s law of abra- RUL in measurement number sive wear. Based on this result, accurate RUL prediction 20 0 is be achieved by estimating the models’s states using 020406080 100 120 140 160 180 computationally simple Kalman filter and Monte Carlo Measurement number simulations over noise realisations. Figure 7: Performance of the RUL prediction algorithm.

Acknowledgements [6] Bisson EE. Adhesive, abrasive, corrosive, and surface fatigue wear modes. Tech. Rep. TM X- The authors acknowledge the support of the Slovenian 52426, NASA Lewis Research Center, Cleveland, Research Agency through the Research Programme OH, United States, 1968. P2-0001 and the Research Project L2-7663. [7] Verhaegen M, Verdult V. Filtering and System Identiﬁcation. Cambridge University Press, Cam- References bridge, UK, 2007.

[1] Lee J, Wu F, Zhao W, Ghaffari M, Liao L, [8] M. Gašperin M, Juricic D. Application of un- Siegel D. Prognostics and health management scented transformation in nonlinear system iden- design for rotary machinery systems—reviews, tiﬁcation,” {IFAC} Proceedings Volumes, vol. 44, methodology and applications. Mechanical Sys- no. 1, pp. 4428 – 4433, 2011. 18th {IFAC} World tems and Signal Processing, vol. 42, no. 1–2, Congress. pp. 314 – 334, 2014.

[2] Jablonski A. Condition monitoring systems read- ressed and revised. vol. Nov., p. 2, 2015. [3] Boškoski P, Juricic D. Fault detection of mechanical drives under variable operating conditions based on wavelet packet rényi entropy signatures. Mechanical Systems and Signal Processing, vol. 31, pp. 369—381, 2012.

[4] Basseville M. Divergence measures for statistical data processing - an annotated bibliography. Sig-nal Processing, vol. 93, pp. 621–633, 2013.

[5] Hamza A, Krim H. Image registration and segmentation by maximizing the jensen-rényi divergence. In Energy Minimization Methods in Computer Vision and Pattern Recognition (A. Ran- garajan, M. Figueiredo, and J. Zerubia, eds.), vol. 2683 of Lecture Notes in Computer Science, pp. 147–163, Springer Berlin Heidelberg, 2003.

22SNE 26(4) – 12/2016 SNE T ECHNICAL N OTE

AMEBA-Evolutionary Computation Method: Comparison and Toolbox Development

Marko Corn*, Maja AtanasijeviÉ-Kunc

Faculty of Electrical Engineering, University of Ljubljana, Tržaška cesta 25, 1000 Ljubljana, Slovenia; [email protected]* 6LPXODWLRQ1RWHV(XURSH61( ± ,QFRQWUDVWWRSDUDPHWULFDODOJRULWKPVVWUXFWXUDODO '2,VQHWQ JRULWKPVGRQRWUHTXLUHSUHGHILQHGIRUPRIWKHFRQWURO 5HFHLYHG1RYHPEHU OHUDVWKH\FDQHYROYHWKHZKROHFRQWUROOHUWKURXJKWKHLU $FFHSWHG'HFHPEHU 6SHFLDO,VVXH5HYLHZ HYROXWLRQDU\SURFHVV 7KHPRVWSRSXODUSDUDPHWULFDODOJRULWKPVDUHJHQHW Abstract. Evolution algorithms are optimization meth- LF DOJRULWKPV *$ >@>@ HYROXWLRQDU\ VWUDWHJLHV (6 ods that mimic a process of the natural evolution. Their >@GLIIHUHQWLDOHYROXWLRQ '( >@DQGRWKHUV>@ stochastic properties result in a huge advantage over 0RVWHVWDEOLVKHGVWUXFWXUDODOJRULWKPLVJHQHWLFSUR other optimization methods, especially regarding solving JUDPLQJ *3 WKDW KDV PXOWLSOH LPSOHPHQWDWLRQV IURP complex optimization problems. In this paper, several WKHWKUHHEDVHGLPSOHPHQWDWLRQ>@WRWKHJUDPPDWLFDOO\ types of evolutionary algorithms are tested regarding a EDVHG LPSOHPHQWDWLRQ >@ DQG WKH HYROXWLRQDU\ SUR dynamic nonlinear multivariable system modelling and JUDPPLQJWKDWLVGLUHFWHGLQWRWKHHYROYHPHQWRIILQLWH control design. We have defined three problems: the first VWDWHPDFKLQHV>@ one is the so-called grey box identification problem (YROXWLRQDU\ DOJRULWKPV FDQ EH XVHG DOVR LQ WKH where the characteristic of the system’s valve is under FRPSOH[ ILHOG RI WKH GHVLJQ RI FRQWUROOHUV RI G\QDPLF investigation, the second one is a black box identification V\VWHPVHJPXOWLYDULDEOHQRQOLQHDUWLPHYDULDQW>@ where the goal is a dynamic system’s model develop- ,QWKLVSDSHUWKHHYROXWLRQRIGLIIHUHQWPRGHOVDQG ment using system’s measurements data, while the third FRQWURO VWUDWHJLHV DUH GHVLJQHG DQG FRPSDUHG ZLWK WKH one is a system’s controller design. The efficacy of solving XVDJH RI GLIIHUHQW HYROXWLRQDU\ DOJRULWKPV )URP WKH presented problems was compared to the usage of the SDUDPHWULFDO JURXS WKH HIILFDF\ RI *$ (6 DQG '( LV following optimization methods: genetic algorithms, LOOXVWUDWHGZKLOHIURPWKHVWUXFWXUDOJURXSDQDOJRULWKP differential evolution, evolutionary strategies, genetic RIWUHHEDVHGJHQHWLFSURJUDPPLQJDQGWKH$JHQW0RG programming, and a developed approach called AMEBA HOOHG(YROXWLRQDU\%DVHG$OJRULWKP $0(%$ DUHXVHG algorithm. All methods have proven to be very useful for >@>@ 5HODWLYH DGYDQWDJHV DQG GLVDGYDQWDJHV KDYH grey box identification and design of a system’s control- EHHQHVWLPDWHGUHJDUGLQJPRGHOOLQJDQGFRQWUROGHVLJQ ler, but AMEBA algorithm has also been successfully RI QRQOLQHDU PXOWLYDULDEOH G\QDPLF V\VWHP RI WKUHH used in a black box identification, where it generated a FRXSOHGWKDQNV corresponding dynamic mathematical model. 7KHSDSHULVRUJDQL]HGLQWKHIROORZLQJZD\,QWKH ILUVW VHFWLRQ D VKRUW GHVFULSWLRQ RI WKH WKUHH FRXSOHG WDQNVV\VWHPLVJLYHQ,QWKHVHFRQGVHFWLRQDVWUXFWXUH Introduction RIWKHV\VWHP¶VPRGHODQGWKHFRUUHVSRQGLQJFRQWUROOHU ,Q JHQHUDO WKH HYROXWLRQDU\ DOJRULWKPV FDQ EH GLYLGHG DUHVSHFLILHG,QWKHWKLUGDQGIRXUWKVHFWLRQVWKHPRGHO LQWRWZRPDMRUJURXSVSDUDPHWULFDODQGVWUXFWXUDODOJR OLQJDQGWKHFRQWUROGHVLJQUHVXOWVZKLFKZHUHJHQHUDWHG ULWKPV 3DUDPHWULFDO DOJRULWKPV HYROYH SDUDPHWHUV XVLQJ GLIIHUHQW HYROXWLRQDU\ DOJRULWKPV DUH SUHVHQWHG ZKLOHVWUXFWXUDODOJRULWKPVHYROYHVWUXFWXUHVRUPDSSLQJ DQG FRPSDUHG 7KH UHVXOW VHFWLRQ LV IROORZHG E\ WKH IXQFWLRQV )RU H[DPSOH LI ZH ZRXOG KDYH WR GHVLJQ D GHVFULSWLRQRI$0(%$V\VWHPWRROER[WKDWZDVXVHGWR FRQWUROOHU IRU D G\QDPLF V\VWHP SDUDPHWULF DOJRULWKP JHQHUDWH WKH UHVXOWV RI $0(%$ PHWKRG >@ $W WKH ZRXOG GHPDQG WR GHILQH SDUDPHWHUV RI WKH FKRVHQ FRQ HQGWKHFRQFOXVLRQVDQGVRPHLGHDVIRUWKHIXWXUHZRUN WUROOHUVWUXFWXUH YHU\IUHTXHQWO\D3,'FRQWUROOHULVXVHG DUHJLYHQ

SNE 26(4) – 12/2016 229 Corn et al. AMEBA-Evolutionary Computation Method: Comparison and Toolbox TN

1 Three Coupled Tanks System ,QSXW IORZ UDWHV DUH GHWHUPLQHG E\ WKH ZDWHU SXPSV ZKLFK DUH FRQWUROOHG ZLWK WKH YROWDJH VLJQDOV X DQG X 6\VWHPRIWKUHHFRXSOHGWKDQNVLVLOOXVWUDWHGLQ)LJXUH :DWHU IORZV IURP WKH ILUVW WR WKH VHFRQG WDQN DQG IURP ,WFRQVLVWVRIWKUHHLGHQWLFDOF\OLQGULFDODVVHPEOHGZDWHU WKHVHFRQGWRWKHWKLUGWDQNDUHJLYHQZLWKWKHHTXDWLRQV WDQNVZLWKFURVVDUHD6ZKLFKDUHLQWHUFRQQHFWHGZLWK

WKHSLSHVDQGWZRYDOYHV9DQG9ZKLOHWKHYDOYHDW WKHRXWSXWSLSHLV9$FWXDWRUVRIWKLVV\VWHPDUHWZR ZDWHUSXPSVWKDWVXSSO\WKHILUVWDQGWKHWKLUGWDQNZLWK ZDWHU IORZV ĭYK W DQG ĭYK W :DWHU OHYHOV LQ HDFK

WDQNK W K W DQGK W DUHPHDVXUHGZLWKWKHFRUUH 7KHVH ZDWHU IORZV GHSHQG RQ WKH ZDWHU OHYHOV LQ WKH VSRQGLQJVHQVRUV/HYHOGLIIHUHQFHEHWZHHQWKHILUVWDQG WDQNVDQGWKHFKDUDFWHULVWLFVRIWKHYDOYHV7KHVHFKDU

WKHVHFRQGWDQNJHQHUDWHVZDWHUIORZĭ W WKURXJKWKH DFWHULVWLFV DUH H[SHFWHG WR EH RI WKH VTXDUH URRW W\SH

YDOYH 9 DQG OHYHO GLIIHUHQFH EHWZHHQ WKH VHFRQG DQG )URPWKHH[SHULPHQWDOGDWDLWZDVHVWDEOLVKHGWKDWVWDWLF

WKHWKLUGWDQNJHQHUDWHVIORZĭ W WKURXJKWKHYDOYH9 FKDUDFWHULVWLFRIWKHYDOYHV3LVQRWVTXDUHURRWIXQFWLRQ

7KHRXWSXWIORZĭL]K W GHSHQGVRQO\RQWKHZDWHUOHYHO DQGVRZHKDYHWULHGWRHVWLPDWHFRUUHVSRQGLQJGHVFULS

K W DQGYDOYH9SURSHUWLHV WLRQ E\ WKH VRFDOOHG LQGLUHFW LGHQWLILFDWLRQ PHWKRG RU µJUH\ER[LGHQWLILFDWLRQ¶>@*UH\ER[LGHQWLILFDWLRQLV D SURFHVV LQ ZKLFK ZH ILUVWO\ JDWKHU PHDVXUHPHQWV RI WKHV\VWHP¶VEHKDYLRXUVHFRQGO\ZHEXLOGDPDWKHPDWL FDOPRGHODQGLQFOXGHDOOWKHGDWDWKDWZHKDYHLQWRLW 7KLUGO\ ZH WU\ WR HVWLPDWH WKH PLVVLQJ SDUDPHWHUV RU IXQFWLRQV WR WKH FRQVWUXFWHG PRGHO %ORFN GLDJUDP RI WKHFKRVHQVWUXFWXUHLVLOOXVWUDWHGLQ)LJXUH

ĭvh 1(t) h (t) Figure 1: System of three coupled tanks. u1(t) 1 h (t) 6\VWHPRIWKUHHFRXSOHGWKDQNVUHSUHVHQWVDODERUDWRU\ Pumps Part model 2 u2(t) ĭvh 2(t) GHYLFHEXWIRUWKHWHVWLQJZHKDYHXVHGLWVPRGHO>@ h3(t)

1.1 Model structure ĭIz h(t) 'XULQJ WKH SKDVH RI GHVLJQLQJ D PRGHO RI FHUWDLQ G\ Valve V3 QDPLF V\VWHP LW LV XVXDOO\ GHVLUHG WR LQFOXGH DV PXFK NQRZOHGJHRIWKHV\VWHPDVSRVVLEOH,QVXFKDZD\ZH Figure 2: Block diagram of the three coupled tanks KDYHPRUHFKDQFHVRIEXLOGLQJDVXLWDEOHPRGHO7KHR system structure. UHWLFDO PRGHOOLQJ DSSURDFK HQDEOHV PRGHO EXLOGLQJ RQ (VWLPDWLRQ RI WKH FKDUDFWHULVWLF RI WKH YDOYH 9 LV GH WKHEDVLVRIWKHHTXLOLEULXPHTXDWLRQVZKLFKGHWHUPLQH ILQHGRSWLPL]DWLRQSUREOHPDVWKHUHVWRIWKHPRGHOKDG V\VWHP¶V EDVLF EHKDYLRXU )RU IXUWKHU PRGHO LPSURYH EHHQ FRQVWUXFWHG XVLQJ WKH HTXLOLEULXP HTXDWLRQV DQG PHQW DGGLWLRQDO QRQOLQHDU IXQFWLRQV DUH QHHGHG ZKLFK PHDVXUHGFKDUDFWHULVWLFVRIWKHRWKHUSDUWV2SWLPL]DWLRQ GHVFULEH GLIIHUHQW VSHFLILF SDUWV RI WKH V\VWHP ,Q WKH SURFHVV ZDV PLQLPL]LQJ WKH GLIIHUHQFH EHWZHHQ UH ILUVW SKDVH WKH V\VWHP¶V PRGHO FDQ EH SUHVHQWHG ZLWK VSRQVHVRIWKHPRGHODQGPHDVXUHPHQWVRIWKHV\VWHP WKUHH HTXLOLEULXP HTXDWLRQV ZKLFK DUH GHVFULEHG ZLWK E\DGDSWLQJYDOYH¶VFKDUDFWHULVWLF7KHILWQHVVIXQFWLRQ HTXDWLRQV XVHGLQWKLVRSWLPL]DWLRQSURFHVVLVSUHVHQWHGE\HTXD WLRQ ଷ כ ൌ෍නȁ݄௜ሺݐሻ െ݄௜ ሺݐሻȁ݀ݐ ௜ୀଵ )LWQHVVIXQFWLRQLVHTXDOWRWKHDEVROXWHVXPRIGLIIHU HQFHEHWZHHQUHVSRQVHVRIWKHPRGHODQGFRUUHVSRQGLQJ PHDVXUHPHQWV

230 SNE 26(4) – 12/2016 T Corn et al. AMEBA-Evolutionary Computation Method: Comparison and Toolbox N

0HDVXUHPHQWVREWDLQHGIRUWKHLGHQWLILFDWLRQSURFHVV &ORVH ORRS V\VWHP VKRXOG PDLQWDLQ ZDWHU OHYHOV LQ FRQVLVWRIHLJKWUHVSRQVHVWRWKHGLIIHUHQWLQSXWRUH[FL WKHILUVWDQGLQWKHWKLUGWDQNDWWKHFRUUHVSRQGLQJUHIHU

WDWLRQVLJQDOV6L[RIWKHPZHUHXVHGLQWKHLGHQWLILFD HQFHYDOXHVhref1DQGhref2)LWQHVVIXQFWLRQWKDWLVXVHGLQ WLRQ SURFHVV DQG WZR IRU WKH YDOLGDWLRQ RI WKH PRGHO WKHRSWLPL]DWLRQSURFHVVRIWKHFRQWUROOHUGHVLJQLVSUH 2QH SDLU RI WKH H[FLWDWLRQ VLJQDOV DQG FRUUHVSRQGLQJ VHQWHGLQHTXDWLRQ UHVSRQVHVLVLOOXVWUDWHGLQ)LJXUHDQG)LJXUH ൌݓ௢௣௧ නȁ݁ଵሺݐሻȁ ൅ ȁ݁ଶሺݐሻȁ݀ݐ ൅ ሺͳ െ ݓ௢௣௧ሻනȁݑଵሺݐሻȁ ൅ ȁݑଶሺݐሻȁ݀ݐ

)LWQHVV IXQFWLRQ UHSUHVHQWV D VXP RI WKH LQWHJUDOV RI HUURUV e1 DQG e2 WKDW UHSUHVHQWV GLIIHUHQFH EHWZHHQ DFWXDOZDWHUOHYHOVh1DQGh3DQGUHIHUHQFHGYDOXHVhref1 DQGhref2 DQGLQWHJUDOVRIWKHSXPSVDFWLYLW\u1DQGu2 %RWK FRQWULEXWLRQV DUH ZHLJKWHG ZLWK WKH ZHLJKW wopt 7KH FRQWURO V\VWHP ZDV WHVWHG ZLWK WKH XVDJH RI WKH UHIHUHQFHVLJQDOVWKDWDUHSUHVHQWHGLQ)LJXUH

Figure 3: Input signals u1(t) and u2(t).

Figure 6: Reference signals. &RQWUROOHU PXVW EH DEOH WR FRQWURO WKH V\VWHPV ZDWHU Figure 4: Responses of the system to chosen input OHYHOV LQ D ZD\ WKDW LV GHPDQGHG E\ WKH VWHS VKDSHG signals. FKDQJHVRIWKHUHIHUHQFHVLJQDOV )URP WKH SUHVHQWHG UHVSRQVHV WKH FURVV FRXSOLQJV DUH YLVLEOH HDFK LQSXW LQIOXHQFHV ERWK V\VWHPV¶ RXWSXWV 2 Modelling Results K W DQG K W 7KHVH FURVV FRXSOLQJV DOVR SURYH WKDW WKHV\VWHPLVDPXOWLYDULDEOHRQH 0RGHOOLQJUHVXOWVDUHGLYLGHGLQWRWZRJURXSV7KHILUVW JURXS FRQVLVWV RI WKH UHVXOWV REWDLQHG E\ WKH SDUDPHW 1.2 Controller design ULFDO HYROXWLRQDU\ DOJRULWKPV DQG WKH VHFRQG JURXS E\ %ORFNGLDJUDPRIV\VWHP¶VFORVHORRSRSHUDWLRQLVSUH WKHVWUXFWXUDOHYROXWLRQDU\DOJRULWKPV VHQWHGLQ)LJXUH 2.1 Parametrical evolutionary algorithms

h1ref (t) e1(t) u1(t) ĭvh1(t) h1(t) 3DUDPHWULFDOHYROXWLRQDU\DOJRULWKPVFDQRSWLPL]HRQO\ Controller Pumps Process h2(t) h2ref (t) e2(t) u2(t) ĭvh2(t) SDUDPHWHUVVRZHKDYHFRQVWUXFWHGDSRO\QRPLDOPDWK

h3(t) HPDWLFDOIXQFWLRQZLWKIRXUSDUDPHWHUVa1a2a3DQGa4 ZKLFKVKRXOGGHVFULEHDVJRRGDVSRVVLEOHWKHUHODWLRQ

EHWZHHQ WKH ZDWHU OHYHO K W DQG RXWSXW ZDWHU IORZ Figure 5: Closed-loop system operation ĭL]K W ଷ ଶ Ȱ௜௭௛ሺݐሻ ൌܽଵ݄ଷሺݐሻ ൅ܽଶ݄ଷሺݐሻ ൅ܽଷ݄ଷሺݐሻ ൅ܽସ

SNE 26(4) – 12/2016 231 Corn et al. AMEBA-Evolutionary Computation Method: Comparison and Toolbox TN

:H KDYH WHVWHG DQG FRPSDUHG WKUHH SDUDPHWULFDO PHWKRGV*$(6'(2SWLPL]DWLRQSURFHVVZDVGHILQHG IRU DOO PHWKRGV LGHQWLFDOO\ LQ RUGHU WR JHW FRPSDUDEOH UHVXOWV6ROXWLRQVKDYHEHHQHYROYHGGXULQJJHQHU DWLRQV DQG ZLWK WKH JHQHUDWLRQ VL]H RI LQGLYLGXDOV 5HVXOWVDUHSUHVHQWHGLQWZRZD\V7KHILUVWZD\LVWKH FRPSDULVRQRIWKHTXDOLW\RIWKHPRGHOWKDWZDVJHQHU DWHGE\HDFKPHWKRGDQGWKHVHFRQGLVWKHFRPSDULVRQ RI WKH FRQYHUJHQFH RI WKH XVHG PHWKRGV 4XDOLW\ RI JHQHUDWHGVROXWLRQVLVSUHVHQWHGLQ7DEOH Figure 8: Average convergence of parametrical methods Error identification Error validation Met. [%] [%] 6WDWLVWLFDODQDO\VLVRIWKHPHWKRGV¶FRQYHUJHQFHVVKRZV DE 1.77 3.27 HIILFLHQF\RIHDFKDOJRULWKP GXULQJWKHVHDUFKRIRSWL ES 1.79 3.58 PDO VROXWLRQ '( KDV WKH IDVWHVW FRQYHUJHQFH DQG LW GA 1.88 4.57 JHQHUDWHVWKHEHVWUHVXOWV

Table 1: Evaluation of modelling results of parametrical 2.2 Structural evolutionary algorithms algorithms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¶V UH PXOWLRXWSXWV\VWHPV VSRQVHVRIWKHEHVWPRGHOJHQHUDWHGE\WKH'(PHWKRG 6WUXFWXUDODOJRULWKPVDUHFDSDEOHRIEXLOGLQJV\VWHP¶V LVSUHVHQWHGLQ)LJXUH VWUXFWXUH DXWRPDWLFDOO\ 6HWWLQJV RI WKH HYROXWLRQ ZHUH WKHVDPHIRUERWKPHWKRGVZKLFKHQDEOHWKHFRPSDULVRQ RIWKHUHVXOWV)RUWKH*3ZHKDYHXVHGDGGLWLRQVXE WUDFWLRQ PXOWLSOLFDWLRQ GLYLVLRQ SRZHU DQG FRQVWDQW W\SHVRIQRGHVDQGIRUWKH$0(%$DOJRULWKPZHKDYH XVHGWKHVDPHQRGHV¶W\SHVDVIRUWKH*3ZLWKWKHXVHRI DGGLWLRQDO G\QDPLF QRGHV OLNH GHOD\ LQWHJUDO GHULYD WLYH ORZ SDVV ILOWHU DQG KLJK SDVV ILOWHU 5HVXOWV DUH HYDOXDWHGLQ7DEOH

Algorithm Error ident. [%] Error valid. [%] Figure 7: Comparison of measurements with the re- GP 1.62 3.12 sponse of the model generated by the DE AMEBA valve 3.57 4.65 method. AMEBA full model 5.63 7.23

:H KDYH FRPSDUHG DOVR WKH FRQYHUJHQFH RI WKH DOJR Table 2: Evaluation of modelling results when using ULWKPVDQGWKHUHVXOWVZKLFKUHSUHVHQWWKHDYHUDJHFRQ structural algorithms. YHUJHQFH RI RSWLPL]DWLRQ UXQV IRU HDFK PHWKRG DUH *3DOJRULWKPKDVJHQHUDWHGWKHEHVWVROXWLRQDQGLWVWUHH SUHVHQWHGLQ)LJXUH UHSUHVHQWDWLRQLVSUHVHQWHGLQ)LJXUH

232 SNE 26(4) – 12/2016 T Corn et al. AMEBA-Evolutionary Computation Method: Comparison and Toolbox N

9DOYH IXQFWLRQ WKDW ZDV JHQHUDWHG E\ WKH $0(%$ DOJRULWKPLVSUHVHQWHGLQHTXDWLRQ

଴Ǥ଺଼ Ȱ௜௭௛ሺݐሻ ൌ െͲǤͷ ή ͲǤͷͶሺെͲǤͺሺ݄ଷሺݐሻሻሻ 7KH UHVXOW RI WKH YDOYH IXQFWLRQ JHQHUDWHG ZLWK WKH $0(%$ DOJRULWKP LV D QRQOLQHDU IXQFWLRQ $0(%$ DOJRULWKPKDVVXFFHVVIXOO\JHQHUDWHGDOVRDPRGHORIWKH ZKROHV\VWHPZLWKWKHSURFHVVRIEODFNER[LGHQWLILFD WLRQ:HKDYHXVHGWKHVDPHPHDVXUHPHQWVIRUJHQHUDW LQJWKLVPRGHOWKDWZHUHLQXVHIRUWKHLGHQWLILFDWLRQRI WKH YDOYH 0RGHO LV UHSUHVHQWHG LQ )LJXUH 0RGHO JHQHUDWHG ZLWK$0(%$ DOJRULWKP LV FRPSOH[ IXOO RI QRGHV RI DOO W\SHV DQG IHHGEDFN ORRSV WKDW UHSUHVHQW V\VWHPG\QDPLFSURSHUWLHV

Figure 9: Solution generated with the GP method. 6LPSOLILHGVROXWLRQRI*3LVSUHVHQWHGLQHTXDWLRQ 7KLV LV D SRO\QRPLDO IXQFWLRQ ZLWK WZR SDUWV WKH ILUVW KDVUDWLRQDOQXPEHULQWKHH[SRQHQWDQGWKHRWKHULVD OLQHDURQH

5HVXOW JHQHUDWHG E\ WKH $0(%$ DOJRULWKP LV QRW DV JRRGDVWKHUHVXOWREWDLQHGE\*3DQGLWLVSUHVHQWHGLQ Figure 11: Graph representation of system’s model )LJXUH generated with the use of AMEBA algorithm.

3 Results of the Controller Design 5HVXOWVRIGHVLJQLQJFRQWURODOJRULWKPDUHDOVRGLYLGHG LQWR WZR JURXSV LQWR D SDUDPHWULFDO DQG D VWUXFWXUDO Figure 10: Graph representation of model of the valve JURXS generated with AMEBA algorithm. 3.1 Parametrical evolutionary algorithms ,Q7DEOHDOHJHQGLVSUHVHQWHGWKDWVKRZVFRORXUVRI 3DUDPHWULF PHWKRGV XVDJH GHPDQGV D SDUDPHWULFDOO\ GLIIHUHQWW\SHVRIQRGHVDVVHPEOLQJ$0(%$DOJRULWKP GHILQHG SUREOHP VR ZH FRQVWUXFWHG D FRQWUROOHU WKDW LV VROXWLRQV DVVHPEOHG ZLWK IRXU SURSRUWLRQDOLQWHJUDO 3, FRQWURO Color Node Color Node OHUVZLWKSDUDPHWHUVWREHRSWLPL]HG Input Amplification 7KH SURSRVHG FRQWUROOHU LV D PXOWLYDULDEOH RQH ZLWK Output Exponent WZRLQSXWV GLIIHUHQFHVEHWZHHQGHVLUHGDQGDFWXDOZD WHUOHYHOV DQGWZRRXWSXWVWRGULYHZDWHUSXPSV&RQ Low pass filter Delay WUROOHU¶VSDUDPHWHUVWREHRSWLPL]HGDUHGHVFULEHGZLWK High pass filter Derivative HTXDWLRQV Multiply Integral Divide Add

Table 3: Color-legend of different types of nodes.

SNE 26(4) – 12/2016 233 Corn et al. AMEBA-Evolutionary Computation Method: Comparison and Toolbox TN

$OOSDUDPHWHUVDUHUHSUHVHQWHGLQWZRPDWULFHVKpDQG Ki5HVXOWVFDOFXODWHGZLWKWKHSDUDPHWULFDOPHWKRGVDUH SUHVHQWHGLQ7DEOH Figure 12: Graph representation of controller generated by the AMEBA algorithm. Algorithm Error Energy used DE 2.04 % 35.9% &RQWUROOHUWKDWZDVJHQHUDWHGE\$0(%$DOJRULWKPLV GA 2.04 % 36.5% LOOXVWUDWHGE\HTXDWLRQ $0(%$DOJRULWKPJHQHUDW HGDFRQWUROOHUZLWKWKHEHVWSHUIRUPDQFH ES 2.48 % 35.3% Table 4: Evaluation of controller optimization results calculated with parametrical methods. 5HVXOWV RI DOO DOJRULWKPV DUH YHU\ VLPLODU EXW WKH '( PHWKRGKDVRQHVDJDLQSURYHQWREHWKHEHVWDVLWFDOFX ODWHGWKHFRQWUROOHUZLWKWKHORZHVWHUURUDQGPLQLPXP HVWLPDWHGXVDJHRIHQHUJ\

3.2 Structural evolutionary algorithm 4 Toolbox development 6WUXFWXUDO HYROXWLRQDU\ DOJRULWKPV GRQ¶W QHHG WKH FRQ WUROOHU¶VVWUXFWXUHWREHGHILQHGLQDGYDQFHLQFRQWUDVWWR $0(%$ DOJRULWKP LV EHLQJ GHYHORSHG DOVR DV D VRIW SDUDPHWULFDO PHWKRGV 7KLV JURXS LV FDSDEOH WR HYROYH ZDUHSDFNDJHZLWKXVHUIULHQGO\JUDSKLFDOLQWHUIDFH7KH WKHVWUXFWXUHDVZHOODVDOOWKHSDUDPHWHUVDXWRPDWLFDOO\ FRUH GHYHORSPHQW LV EHLQJ EXLOW LQ -DYD SURJUDPPLQJ 5HVXOWVRIWZRPHWKRGV*3DQ$0(%$DUHSUHVHQWHG HQYLURQPHQW WKDW FDQ EHXVHG DOVR ZLWK 0DWODEZKLFK LQ7DEOH DOORZVDYHU\HIILFLHQWVXSSRUWLQVLPXODWLRQRIG\QDPLF Algorithm Error Energy used V\VWHPVYLD6LPXOLQN*UDSKLFDOLQWHUIDFHLVDOVRGHYHO RSHGLQ0DWODEGXHWRLWVJRRGJUDSKLFDOVXSSRUW AMEBA 1.5 % 34.1 % 7RROER[HQDEOHVVHWWLQJVRIWKHVLPXODWLRQHQYLURQPHQW GP 9.3 % 35.5 % ZLWKWKHLQFOXVLRQRI6LPXOLQNPRGHODVLWLVVKRZQLQ Table 5: Results of controllers generated by structural )LJXUH evolutionary methods. 7KHVROXWLRQZKLFKZDVJHQHUDWHGE\WKH*3PHWKRGLV SUHVHQWHGE\HTXDWLRQ

*3 PHWKRG GLGQ¶W JHQHUDWH D VXLWDEOH VROXWLRQ DV WKH FRQWUROOHULVQRWFDSDEOHWRIROORZFRUUHVSRQGLQJUHIHU HQFH VLJQDOV 7KH VROXWLRQ JHQHUDWHG E\ WKH $0(%$ DOJRULWKPLVSUHVHQWHGLQ)LJXUH Figure 13: Settings of simulation environment.

234 SNE 26(4) – 12/2016 T Corn et al. AMEBA-Evolutionary Computation Method: Comparison and Toolbox N

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igure 16: Node settings. OLNH PD[LPXP QXPEHU RI JHQHUDWLRQV DQG PLQLPXP FKDQJHLQILWQHVVIXQFWLRQYDOXH )LJXUH 5HSURGXFWLRQPHFKDQLVPVFDQEHVHWZLWKWKHLUSDUDPH WHURISUREDELOLW\$VDJHQWVDUHHYDOXDWHGDQGVHOHFWHG IRU UHSURGXFWLRQ WKH UHSURGXFWLRQ PHFKDQLVP LV UDQ GRPO\ VHOHFWHG DQG WKH SUREDELOLW\ SDUDPHWHU GHWHU PLQHVWKHLUSRVVLELOLW\RIEHLQJVHOHFWHG )LJXUH

Figure 14: General setting. Figure 17: Reproduction settings. 7KH QXPEHU RI LQSXWV DQG RXWSXWV RI DQ DJHQW FDQ EH $GGLWLRQDO IXQFWLRQDOLWLHV HQDEOHEHWWHUXVDELOLW\ RI WKH GHILQHG WRJHWKHU ZLWK WKH PD[LPXP QXPEHU RI QRGHV PHWKRGVXFKDVVDYLQJDQGLPSRUWLQJRIDOOVHWWLQJLQWR WKDWFDQEHJHQHUDWHGDWWKHDJHQW¶VFUHDWLRQ )LJXUH ILOHIRUODWHUXVH:LWKWKLVRSWLRQDOVRWKHLQLWLDOSRSX ODWLRQ FDQ EH LPSRUWHG ZKLFK HQDEOHV WKH LQFOXVLRQ RI FHUWDLQNQRZOHGJHRIWKHVROXWLRQLQWRWKHRSWLPL]DWLRQ SUREOHP,WLVDOVRSRVVLEOHWRFRQYHUWDJHQWLQWRPDWK HPDWLFDO HTXDWLRQ WR REVHUYH LWV VWUXFWXUH ,W FDQ DOVR JHQHUDWH0DWODE6IXQFWLRQILOHIRUWKHHDVLHULPSOHPHQ WDWLRQLQ6LPXOLQN )LJXUH

Figure 15: Agent settings. 'LIIHUHQWW\SHVRIQRGHVFDQEHVHOHFWHGIURPZKLFKWKH Figure 18: Additions functionalities of Toolbox. DOJRULWKPZLOOFKRVHDQGEXLOGDJHQWV(DFKQRGHKDVLWV RZQ VHWWLQJV WKDW GHWHUPLQH LQLWLDO YDOXH RI WKH QRGHV SDUDPHWHUDQGVWHHSQHVVRIFKDQJHLQFDVHQRGHPXWDWHV )LJXUH

SNE 26(4) – 12/2016 235 Corn et al. AMEBA-Evolutionary Computation Method: Comparison and Toolbox TN

5 Conclusions References >@ $WDQDVLMHYLü.XQF0%HOLþ$.DUED5Optimal 7KH V\VWHP RI WKUHH FRXSOHG WKDQNV ZDV VHOHFWHG WR multivariable control design using genetic algorithms SUHVHQWWKHHIILFDF\RIWKUHHGLIIHUHQWDSSURDFKHVRIWKH 9LHQQD8QLYHUVLW\RI7HFKQRORJ\WK9LHQQD XVDJHRIWKHHYROXWLRQDU\DOJRULWKPVPHWKRGVWKHJUH\ 6\PSRVLXPRQ0DWKHPDWLFDO0RGHOLQJ ER[ LGHQWLILFDWLRQ WKH EODFN ER[ LGHQWLILFDWLRQ DQG WKH >@ *ROGEHUJ'Genetic Algorithms in Search, Optimization FRQWUROOHUGHVLJQ and Machine Learning$GGLVRQ:HVOH\VW 3DUDPHWULFDO HYROXWLRQDU\ DOJRULWKPV JHQHUDWHG >@ %H\HU+*The Theory of Evolution Strategies (Natural JRRGUHVXOWVIRUERWKPRGHOOLQJDQGFRQWURORIWKHV\V Computing Series)6SULQJHUS WHP$OVRVWUXFWXUDOPHWKRGVPDQDJHWRJHQHUDWHJRRG VROXWLRQV IRU ERWK W\SHV RI SUREOHPV ,Q JHQHUDO WKH >@ 6WRUQ53ULFH.Differential Evolution – A Simple and PRVWLPSRUWDQWDGYDQWDJHRIWKHVWUXFWXUDODOJRULWKPVLQ Efficient Heuristic for Global Optimization over FRPSDULVRQWRWKHSDUDPHWULFDOPHWKRGVLVWKHDEVHQFH Continuous Spaces-RXUQDORI*OREDO2SWLPL]DWLRQ RIWKHQHHGWRGHILQHDVXLWDEOHVWUXFWXUH7KLVSURSHUW\ LV HVSHFLDOO\ LPSRUWDQW ZKHQ GHDOLQJ ZLWK PRUH FRP >@ %URZQOHH-Clever Algorithms: Nature-Inspired SOH[V\VWHPVZLWKPXOWLSOHLQSXWVDQGRXWSXWV:LWKWKH Programming Recipes6ZLQEXUQH8QLYHUVLW\LQ XVDJHRI$0(%$DOJRULWKPZHKDYHPDQDJHGWRJHQ 0HOERXUQH$XVWUDOLD6HDUFK HUDWHDOVRDFRPSOHWHPRGHORIWKHV\VWHPDQGZHJHQ >@ .R]D-5Genetic Programming On the Programming of HUDWHGDV\VWHPFRQWUROOHUZLWKWKHEHVWSHUIRUPDQFH Computers by Means of Natural Selection0,73UHVV )XWXUH ZRUN RQ $0(%$ DOJRULWKP GHYHORSPHQW WK ZLOOEHIRFXVHGRQRSWLPL]DWLRQSURFHVVDVZHDUHJRLQJ >@ :KLJKDP3$Grammatically-based Genetic WRH[SORUHWKHLPSDFWRIYDULRXVHIIHFWVRQWKHTXDOLW\RI Programming'HSDUWHQWRI&RPSXWHU6FLHQFH WKHVROXWLRQDQGRQWKHFRQYHUJHQFHUDWHRIRSWLPL]DWLRQ 8QLYHUVLW\&ROOHJH8QLYHUVLW\RI1HZ6RXWK9DOHV SURFHVV OLNH WKH HIIHFW RI VL]H RI WKH SRSXODWLRQ VL]H :RUNVKRSRQ*HQHWLF3URJUDPPLQJ)URP7KHRU\WR VXSSUHVVLRQ RI WKH DJHQWV ZLWK ODUJH QXPEHU RI QRGHV 5HDO:RUOG$SSOLFDWLRQV XVLQJ PXOWLSOH HQYLURQPHQWV DW RQFH DQG VLPLODU RI >@ )RJHO/-2ZHQV$-:DOVK0-Artificial Intelligence FRXUVHLQFRPSDULVRQZLWKRWKHURSWLPL]DWLRQDSSURDFK through Simulated Evolution-RKQ:LOH\ HV 6SHFLDO DWWHQWLRQ ZLOO EH GHYRWHG WR WKH VR FDOOHG >@ 7RPDåLþ6/RJDU9.ULVWOä.UDLQHU$âNUMDQF, VPDUW RSWLPL]DWLRQ ZKHUH DGGLWLRQDO NQRZOHGJH IURP .RãLU0Indoor-environment simulator for control FKRVHQ DUHD FDQ EH WDNHQ LQWR DFFRXQW WR LPSURYH design purposes%XLOGLQJDQG(QYLURQPHQW VHDUFKLQJ HIILFDF\ 7KH$0(%$ PHWKRG LV D ZRUN LQ >@ &RUQ0$WDQDVLMHYLü.XQF0Cell based Genetic SURJUHVV DQG WKH PHWKRG ZLOO EH DYDLODEOH DV DQ RSHQ Programming Toolbox (CGP - Toolbox)WK(5. VRXUFHSURMHFW &RQIHUHQFH >@ &RUQ0ýHUQH*A Graph-Based Evolutionary Algorithmࣟ: Cell Based Genetic Programming 3URFHHGLQJVRIWKHWK,QWHUQDWLRQDO&RQIHUHQFHRQ %LRLQVSLUHG2SWLPL]DWLRQ0HWKRGVDQGWKHLU $SSOLFDWLRQV%,20$ >@ &RUQ0AMEBA – Structural evolutionary optimization: method and toolbox developmentWK(XURSHDQ 0RGHOLQJDQG6LPXODWLRQ6\PSRVLXP >@ $WDQDVLMHYLü.XQF0Multivariabilni sistemi, Zbirke kompleksnejših problemov=DORåED)(LQ)5, S >@ 7DQ.&/L<Grey-box model identification via evolutionary computing&RQWURO(QJLQHHULQJ3UDFWLFH

236 SNE 26(4) – 12/2016 SNE T ECHNICAL N OTE

Modeling and Simulation of GMA Welding Process and Welding Power Sources

Marjan Golob

Faculty of Electrical Engineering and Computer Science, University of Maribor, Smetanova ulica 17, 2000 Maribor, Slovenija; [email protected] 6LPXODWLRQ1RWHV(XURSH61( 3URFHVV PRGHOOLQJ SURYLGHV D PHDQV RI '2,VQHWQ LQFRUSRUDWLQJSULQFLSDODQGHPSLULFDOLQIRUPDWLRQLQWRD 5HFHLYHG1RYHPEHU FRQWURO VWUDWHJ\ 0RGHOV PD\ EH XVHG RIIOLQH WR $FFHSWHG'HFHPEHU 6SHFLDO,VVXH5HYLHZ HYDOXDWHDQGWXQHDFRQWUROOHULQDVLPXODWLRQ7KH\PD\ DOVREHXVHGWRGHYHORSWUDQVIHUIXQFWLRQVRIDSURFHVV Abstract. Simulation techniques are useful tools for IRUXVHLQIRUPDOFRQWUROOHUGHVLJQRUWRSURYLGHPDSV study and research of new welding technologies, and for EHWZHHQ LQSXW DQG RXWSXW SDUDPHWHUV 3URFHVV PRGHOV the rapid development of new control algorithms and DUHLPSRUWDQWEULGJHEHWZHHQZKDWLVNQRZQDQGZKDWLV control units such as power source circuits, and welding GHVLUHG current or voltage controllers. The objective in this re- 6HYHUDO UHVHDUFK VWXGLHV IRU H[DPSOH >@>@ search is to combine the simulation of Gas Metal Arc FDWHJRUL]HWKH*0$:SURFHVVDVDQHOHFWULFDOFLUFXLW$ Welding (GMAW) process models with the simulation PDWKHPDWLFDOPRGHORIWKH*0$:SURFHVVLVQRUPDOO\ models of inverter based power machines. The GMAW GHYHORSHGILUVW$GHVFULSWLRQRIWKHHOHFWULFDUFLVWKHQ process is considered as an electrical circuit and the SUHVHQWHGDQGDOOHTXDWLRQVDUHFRPELQHGLQWRDJHQHUDO mathematical model is based on physical descriptions of PRGHO WKDW GHVFULEHV WKH *0$: SURFHVV 6LPXODWLRQ several parts of GMAW process, as are the electric circuit PHWKRGV DUH XVHG WR LOOXVWUDWH WKH EHKDYLRXU RI WKH of power supply, the arc dynamics, and the electrode *0$:SURFHVV'XULQJWKHVHVLPXODWLRQVWKHZHOGLQJ melting process. To establish the validity of the proposed SRZHU VRXUFH V G\QDPLF EHKDYLRXU LV RIWHQ VLPSOLILHG GMAW model, a simple welding application was simulat- )RU PRVW *0$: DSSOLFDWLRQV WKH GHVLUHG ZHOGLQJ ed and welding parameters were derived from experi- FRQGLWLRQV DUH VXFK WKDW WLPH FRQVWDQW RI WKH VHOI mental conditions. Next, the simulation model of full- UHJXODWLQJSURFHVVLVVKRUWHUWKDQWKHRVFLOODWLRQUDWH>@ bridge DC-DC converter is presented and the discrete PI :LWKWKHDLPRIPDLQWDLQLQJDKLJKTXDOLW\RIZHOGLQJ controller for welding current feedback control is pro- UHVXOWVWKHRXWSXWZHOGLQJFXUUHQWDQGYROWDJHPXVWEH posed. Both models, the GMAW model and the inverter FRQWUROOHG GXULQJ WKH ZHOGLQJ SURFHVV )XUWKHUPRUH D power supply model, are combined and simulated to- UHDO WLPH FRQWURO V\VWHP LV DQ LPSRUWDQW HOHPHQW RI gether. Finaly, the simulation study of firing the PRGHUQ*0$:ZHOGLQJPDFKLQH>@>@ thyristors, which enables steady and pulsed direct 0RGHUQ*0$:HTXLSPHQWLVWKHFRPELQDWLRQRID current welding with a single fully controlled bridge VRSKLVWLFDWHG SRZHU HOHFWURQLF GHYLFH DQG KLJK converter is shown. SHUIRUPDQFHPLFURSURFHVVRUEDVHGFRQWUROV\VWHPV7KH GHYHORSPHQW SURFHVV RI LQYHUWHUEDVHG ZHOGLQJ SRZHU VRXUFH ZLWK WKH FRUUHVSRQGLQJ FRQWURO V\VWHP LV D Introduction FRPSOH[DQGH[SHQVLYHSURFHVVWKDWUHTXLUHVH[WHQVLYH &RQYHQWLRQDODSSURDFKHVWRDXWRPDWLRQRIZHOGLQJKDYH KXPDQ DQG PDWHULDO UHVRXUFHV >@ :KHQ XVLQJ EHHQUHDVRQDEO\VXFFHVVIXOEXWWKHUHDUHVWLOOVLJQLILFDQW VLPXODWLRQ WKH TXDOLW\ RI GHVLJQ SURFHVV FDQ EH RSSRUWXQLWLHV IRU DGGLWLRQDO GHYHORSPHQW 6XFFHVVIXO LPSURYHG DQG WKH GHVLJQ FRVW FDQ EH VLJQLILFDQWO\ LPSOHPHQWDWLRQ RI PXOWLYDULDEOH ZHOG SURFHVV FRQWURO UHGXFHG LQYROYHVVHQVLQJPRGHOOLQJDQGFRQWURO

SNE 26(4) – 12/2016 237 Golob Modeling and Simulation of GMA Welding Process and Welding Power Sources TN

7KHUHIRUH ZLWK WKH DLP RI LPSURYLQJ WKH GHVLJQ DQGWKHG\QDPLFRIHOHFWURGHVWLFNRXWLVJLYHQE\ SURFHVVDQDWWHPSWZDVPDGHWRFRPELQHWKHVLPXODWLRQ ݀݅ PRGHORI*0$:SURFHVVZLWKWKHVLPXODWLRQPRGHOVRI ൌݒ െݒ ൅ݒ ݐ ௘ ௠ ௖݀ ZHOGLQJ SRZHU VRXUFHV >@ >@ 7KH VLPXODWLRQ UHVXOWV DUHYHU\XVHIXOIRUWKHUDSLGGHYHORSPHQWRIQHZFRQWURO :LWKUHVSHFWWR DQG WKHDUFOHQJWKVSHHGLV DOJRULWKPVDQGWKHGHVLJQLQJRIQHZSRZHUVRXUFHV ݄݀ ൌݒ െݒ െݒ ݐ ௠ ௘ ௖݀

1 GMAW Process Dynamic WKH G\QDPLFV RI WKH PHOWLQJ VSHHG vm DQG DUF YROWDJH U QHHGWREHGHVFULEHGLQJUHDWHUGHWDLO Model arc $IXQGDPHQWDOVLJQLILFDQFHRI*0$:SURFHVVLVWKDWLW LQFRUSRUDWHV DXWRPDWLF IHHGLQJ RI D FRQVXPDEOH HOHFWURGH WKDW LV SURWHFWHG E\ DQ H[WHUQDOO\ VXSSOLHG VKLHOGLQJ JDV DV LV SUHVHQWHG LQ )LJXUH $ FRQVWDQW YROWDJH SRZHU VXSSO\ LV IHG WR WKH HOHFWURGH DQG WKH ZRUNSLHFH 7R JHW WKH WKH GHVLUHG ZHOG TXDOLW\ ZLUH

IHHGVSHHGWRUFKWUDYHOVSHHGRSHQFLUFXLWYROWDJH DQG FRQWDFW WLS WR ZRUNSLHFH GLVWDQFH FDQ EH DGMXVWHG+HUHLVWKHGLVWDQFHRIWKHFHQWHURIPDVVRI WKH GURSOHW DERYH WKH ZRUNSLHFH 7KH PDWKHPDWLFDO Figure 1: Schematic diagram of GMAW process and PRGHOGHYHORSPHQWRID*0$:SURFHVVLVSHUIRUPHG electrical circuit of the self-regulating arc E\WDNLQJLQWRFRQVLGHUDWLRQWKHPRGHORIWKHHOHFWULFDO process. (1) Power source; (2) wire feed unit; (3) FLUFXLWWKHZHOGLQJZLUHPHOWLQJUDWHWKHPRGHORIWKH shielding gas; (4) welding gun; (5) workpiece; (6) G\QDPLFV RI WKH SHQGHQW GURS DQG WKH SKHQRPHQD RI welding arc and material transfer process. WKHGURSWUDQVIHUPRGH 7KHVXPRIWKHYROWDJHVDURXQGD*0$:FLUFXLWDV 7KHWRWDODUFYROWDJHuarcLVPDGHXSRIWKUHHVHSDUDWH SUHVHQWHGLQ)LJXUHLV SDUWVWKHDQRGHDQGFDWKRGHGURSYROWDJHua+cWKHGURS YROWDJH LQ WKH DUF FROXPQ ZKLFK LV D IXQFWLRQ RI WKH ݀݅ HOHFWULF ILHOG VWUHQJWK EDQGWKHDUFOHQJWKh DQG WKH ൅ܴ ή݅൅ݑ ܮݑൌܴή݅൅ ݐ ௟ ௔௥௖݀ GURS YROWDJH WKDW GHSHQGV RQ FXUUHQW i DQG DUF ZKHUHuLVRSHQFLUFXLWYROWDJHRIWKHSRZHUVRXUFHR LV UHVLVWDQFH R ,QRXUPRGHOZHVXSSRVHWKDWu LV UHVLVWDQFH RI WKH SRZHU VRXUFH L LV LQGXFWDQFH RI WKH arc a+c FRQVWDQW &RQVLGHULQJ WKLV WKH VLPSOHVW PRGHO RI WKH SRZHUVRXUFHiLVZHOGLQJFXUUHQWRlLVHOHFWURGHVWLFN HOHFWULFDODUFLVDYROWDJHHTXDWLRQ RXW UHVLVWDQFH DQG uarc LV DUF YROWDJH (OHFWURGH VWLFN

ή݄൅݅ήܴ௔௥௖ܧRXWUHVLVWDQFHLVGHSHQGHQWRQUHVLVWLYLW\RIWKHHOHFWURGH ݑ௔௥௖ ൌݑ௔ା௖ ൅ VWLFNρFURVVVHFWLRQDODUHDRIWKHHOHFWURGHZLUHADQG :KHQWKHFXUUHQWIORZVWKURXJKWKHHOHFWURGHDQGWKH HOHFWURGHVWLFNRXWOHQJWKl,WLVDVVXPHGWKDWuRLρ DUF WKH HOHFWURGH LV KHDWHG E\ WKH FXUUHQW IORZLQJ DQG A DUH FRQVWDQW SDUDPHWHUV DQG i l u DUH arc WKURXJK LW 7KLV KHDW GHSHQGV RQ WKH UHVLVWDQFH RI WKH GHSHQGHQWYDULDEOHV ZHOGLQJ ZLUH 6HYHUDO VWXGLHV KDYH GHVFULEHG WKH 7KH HOHFWURGH UHVLVWDQFH R GHSHQGV RQ WRWDO WKH l SK\VLFDO EDFNJURXQG RI D ZHOGLQJ ZLUH PHOWLQJ OHQJWKRIWKHHOHFWURGHVWLFNRXWOHQJWKDQGGURSOHQJWK SKHQRPHQRQ,Q>@WKHUHVHDUFKUHVXOWVIURPDVWXG\RI l=l +l 7KHG\QDPLFVRIl GHSHQGRQWKHIHHGLQJVSHHG s d s DQRGH DQG FDWKRGH PHOWLQJ UDWHV DUH SUHVHQWHG DQG LQ RIHOHFWURGHv WKHPHOWLQJVSHHGv DQGRQWKHYHUWLFDO e m >@WKHFKDUDFWHULVWLFRIPHOWLQJUDWHDVDIXQFWLRQRI YHORFLW\ RI WKH FRQWDFW WLS v 7KH FRQWDFW WLS WR c FXUUHQWW\SHRIJDVDQGRWKHUSDUDPHWHUVLVUHSRUWHG,Q ZRUNSLHFHGLVWDQFH &7:' LVLQGLFDWHGE\HDQGE\ WKHVHDQGRWKHUUHODWHGZRUNV>@>@WKHH[SUHVVLRQ LJQRULQJ WKH OHQJWK RI WKH GURS l DQG l l WKH d s IRUWKHWRWDOPHOWLQJYHORFLW\v LVSURSRVHGDV OHQJWKRIWKHDUFh LV m ݒ ൌ݇ ή݅൅݇ ή݅ଶ ή݈ ݄ൌܪെ݈ ௠ ଵ ଶ

238 SNE 26(4) – 12/2016 T Golob Modeling and Simulation of GMA Welding Process and Welding Power Sources N

ZKHUH KDQGK DUH HPSLULFDO FRQVWDQWV IRU JLYHQ 2.1 Inverter based power sources ZLUH PDWHULDOV DQG VL]HV $Q HTXLYDOHQW VWDWHVSDFH 7KHLQYHUWHUEDVHGZHOGLQJSRZHUVXSSO\FRQVLVWVRID UHSUHVHQWDWLRQLVSUHVHQWHGEHORZ UHFWLILHU DQ LQYHUWHU VZLWFK FLUFXLW D KLJKIUHTXHQF\ IHUULWH WUDQVIRUPHU KLJKIUHTXHQF\ UHFWLILHU DQG DQ ͳ U ήݔ LQGXFWRUDVLVSUHVHQWHGLQ)LJXUHܧݔሶ ൌ ήቆݑ െ ήݔ ή ሺݔ െݔ ሻቇെݑ െ ଵ ܮ ଵ ܣ ଵ ଷ ଶ ௔ା௖ ଶ

െ ሺܴ௔௥௖ ൅ܴሻ ήݔଵ ଶ ݔሶଶ ൌ݇ଵ ήݔଵ ൅݇ଶ ήݔଵ ή ሺݔଷ െݔଶሻ െݑଶ െݑଷ ݔሶଷ ൌݑଷ ݕൌݔ ଵ ݔଵ ൌ݅ LV WKH ZHOGLQJ FXUUHQW Figure 2: Power supply architecture of modern-inverterZKHUH WKH VWDWHV DUH .LVWKH&7:' based welding machineܪݔଶ ൌ݄ LV WKH OHQJWKRI WKH DUFݔଷ ൌ DQG LPSXWV DUH ݑ ൌݑ LV RSHQ FLUFXLW YROWDJH RI WKH ଵ 7KH VZLWFK FLUFXLWV DUH FRQWUROOHG E\ PLFURSURFHVVRU SRZHUVRXUFHݑ ൌݒLVWKHIHHGLQJVSHHGRIHOHFWURGH ଶ ௘ EDVHG 3:0 FRQWUROOHU XQLWV 7KH VFKHPDWLF RI D DQGݑ ൌݒLVWKHYHUWLFDOYHORFLW\RIWKHFRQWDFWWLS ଷ ௖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m Figure 3: Power supply architecture of modern-inverter LQYROYHV PDQ\ FRPSOH[ SDUDPHWHUV PRUH ZKLOH XVLQJ based welding machine. &2DQGRWKHUVKLHOGLQJJDVHV 0DQ\UHVHDUFKHUVKDYH ZRUNHGRQWKHPRGHOLQJRIWKH*0$:SURFHVV$¿IWK 7KH FRQYHQWLRQDO '&'& FRQYHUWHU RSHUDWHV XVLQJ D RUGHUQRQOLQHDUPRGHORI*0$:SURFHVVKDVEHHQXVHG 3XOVH:LGWK0RGXODWLRQ 3:0 FXUUHQWFRQWUROOHU7KH E\PRVWRIWKHUHVHDUFKHUVIRUWKHFRQWUROSURFHVV>@>@ '&'& FRQYHUWHU RSHUDWHV DW FRQVWDQW VZLWFKLQJ >@ IUHTXHQF\ZKLFKLVXVXDOO\OLPLWHGWRN+]7KH DPSOLWXGHRIWKHZHOGLQJFXUUHQWGHSHQGVRQWKHFKDQJH RI WKH SKDVH VKLIW EHWZHHQ WUDQVLVWRUV 6 6 6 DQG 2 Welding Power Sources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hyristor based power sources OHVVWKDQVIRUDSXOVHWLPHRIPVDQGDSXOVH 7KH V\QHUJLF SXOVHG 0,*0$* ZHOGLQJ ZLWK ZLGWK IUHTXHQF\ RI +] 7RGD\ WKH DV\PPHWULFDO KDOI FRQWUROOHG VLQHZDYH FXUUHQW SXOVHV LV PRVWO\ UHDOLVHG EULGJHIRUZDUGFRQYHUWHUEHFDPHWKHIDYRXULWHWRSRORJ\ ZLWKWK\ULVWRUEDVHGSRZHUVRXUFHV RI DQ LQYHUWHU SRZHU VRXUFH ,*%7V RU 026)(7V DUH PRVWO\ XVHG DV SRZHU VZLWFKHV GHSHQGHQW RQ VXSSO\ YROWDJHDQGRXWSXWSRZHUUDQJH

SNE 26(4) – 12/2016 239 Golob Modeling and Simulation of GMA Welding Process and Welding Power Sources TN

6XFKFXUUHQWSXOVHZDYHIRUPFRXOGEHREWDLQHGLID Parameter Descr. of the parameter Value SRZHU VRXUFH LV FRQWUROOHG E\ WK\ULVWRUV LQWHJUDWHG R Power source resistance 0.07 Ω ZKHQ WKH SXOVH IUHTXHQF\ LV FRQVWDQW DQG WKH FXUUHQW L Power source inductance 0.02 mH SXOVHSRZHULVFRQWUROOHGE\WKHGHOD\WLPHRIWKHWK\ULV Specific electrical resistance ρ 0.1 Ω/m WRULJQLWLRQ>@>@>@ of the electrode 7KH WK\ULVWRUEDVHG ZHOGLQJ SRZHU VXSSO\ FRQVLVWV Cross-sectional area of the A 1.02 ·10-6 m2 RI D UHFWLILHU DQ LQYHUWHU VZLWFK FLUFXLW D KLJK electrode wire IUHTXHQF\ IHUULWH WUDQVIRUPHU KLJKIUHTXHQF\ UHFWLILHU E Electric field strength 675 V/m

DQGDQLQGXFWRUDVLVSUHVHQWHGLQ)LJXUH ua+c Arc voltage constant 11.55 V R Arc resistance Ω 7ULID]QL'< 7ULID]QLSROQL 0,*0$* arc 0.03 WUDQVIRUPDWRU WLULVWRUVNLPRVWLþ ,QGXNFLMVNLILOWHU YDULOQLSURFHV ve Feeding speed of electrode 0.5 m/min i 1DSDMDQMH k1 Empirical constant 0.626 m/(As) a9+] k Empirical constant 7.55 ·10-5 (A2s)-1 2 Contact tip to workpiece H 0.16 m Figure 4: Power supply architecture of modern-inverter distance (CTWD) based welding machine. Table 1: GMAW process simulation parameters. 7KH WK\ULVWRUEDVHG SRZHU VRUFHV SURGXFH KDUPRQLFV LQWKH$&SRZHUVXSSO\QHWZRUN,QWKHFDVHRISXOVHG 7KHZHOGLQJFXUUHQWURVHDQGIHOOZLWKWKHFKDQJHVRIH GLUHFW FXUUHQW ZHOGLQJ LW LV SRVVLEOH ZLWK D SURSHU DQGveDVH[SHFWHG,WFDQEHVHHQIURPWKHILUVWSORWLQ PHWKRGRIILULQJWKHWK\ULVWRUVWRJHQHUDWHERWKVWHDG\ )LJXUHWKDWWKHDUFOHQJWKh GRWWHGFXUYH GHFUHDVHG GLUHFWDVZHOODVSXOVHGGLUHFWFXUUHQWZLWKDVLQJOH DIWHU WKH H FKDQJHG IURP WR PP DQG WKHQ LQ IXOO\FRQWUROOHGEULGJHFRQWUROOHU8VLQJDQDFWLYHILOWHU FUHDVHG EDFN WR WKH SUHYLRXV OHQJWK $FFRUGLQJO\ WKH GXULQJ SXOVH ZHOGLQJ ZH FRXOG UHGXFH KDUPRQLFV HI HOHFWURGHOHQJWKlFKDQJHGIURPPPWRPP IHFWLYHO\ ZKLFKPHDQWWKDWWKHHOHFWURGHPHOWHGDWDKLJKHUVSHHG ZKHQWKHFXUUHQWLQFUHDVHG,QWKHVHFRQGSORWRI)LJXUH WKH HOHFWURGH IHHGLQJ VSHHG ZDV LQFUHDVHG IURP 3 Simulation Examples and PPV GR PPV 7KLV OHG D UHGXFWLRQ RI WKH DUF UH Results VLVWDQFHDQGDQLQFUHDVLQJRIZHOGLQJFXUUHQW

3.1 Simulation of GMAW Proces $Q DXWRPDWLF ZHOGLQJ DSSOLFDWLRQ ZDV DVVXPHG 7KH SDUDPHWHUV GHULYHG IURP H[SHULPHQWDO FRQGLWLRQV DUH VKRZQLQ7DEOH $FRQVWDQWZHOGLQJVSHHGZDVDVVXPHG7KHZHOG LQJ WRUFK ZDV SRVLWLRQHG PP H IURP WKH ZRUN GLVWDQFH 7KH VHOHFWHG ZHOGLQJ ZLUH IHHG UDWH ve PPVDQGWKHRSHQFLUFXLWYROWDJHu 9ZHUHVHW Figure 5: Simulated results of contact to workpiece voltage waveform (third plot) and welding current 7KHILUVWVLPXODWLRQZDVSHUIRUPHGWRILQGWKHZHOG waveform (fourth plot). Simulation response of LQJ FXUUHQW UHVSRQVH ZKHQ WKH &7:' ZDV FKDQJHG the GMAW model when the CTWD was changed IURPPPWRPP DWWLPHV DQGEDFN DWWLPH from 16 mm to 12 mm (first plot) and the elec- V ,Q DGGLWLRQ WKH HOHFWURGH IHHGLQJ VSHHG v ZDV e trode feeding speed ve was changed from 0.5 FKDQJHGIURPFPPLQWRFPPLQDWWLPHt V m/min to 0.7 m/min (second plot). )LJXUHVKRZVWKHFKDQJHVLQWKHZHOGLQJYROWDJH DQGFXUUHQWWLPHUHVSRQVHVDQGWKHFKDQJHVRIWKHDUF OHQJWK

240 SNE 26(4) – 12/2016 T Golob Modeling and Simulation of GMA Welding Process and Welding Power Sources N

3.2 Simulation of Dynamic Behaviour of a $IWHU PV WKH ZHOGLQJ FXUUHQW V VHW SRLQW ZDV LQ Full-Bridge DC-DC converter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arameter Descr. of the parameter Value

fs Switching frequency 40 kHz C Capacitance 1 μF Pn Transformer nominal power 5 kW

n1 : n2 : n2 Transformer turns ratio 3.5 : 1 : 1

S1 - S4 Ideal switch, IGBT

Ron Switch internal resistance 140 mΩ Ω Rs Snubber resistance 1M C Snubber capacitance 4.7 nF sd Figure 7: Simulation results of generated PWM signals, T Control sample time 0.1 ms s which depend on the duty cycle controlled with T PI controller Integral constant 2 ms I simple PI controller. The second and third plots K PI controller proportional gain 0.2 %/A P show the PWM signals for driving the full-bridge Table 2: DC-DC converter and other circuit parameters. DC-DC converter switches. The fourth and fifth 7KH VLPXODWLRQ UHVXOWV IURP WKH ZHOGLQJ XVLQJ FXUUHQW plots show the corresponding changes of primary current and secondary - welding current. FRQWURO IHHGEDFN DQG WKH 3:0 IXOOEULGJH '&'& FRQYHUWHUDUHVKRZQLQ)LJXUH ,QWKHILUVWSORWRI)LJXUHWKH3:0IUHTXHQF\JHQHUD WRU LV FRPSDUHG ZLWK FXUUHQW FRQWUROOHU RXWSXW GXW\ F\FOH ,Q VHFRQG DQG WKLUG SORWV WKH 3:0 VLJQDOV IRU GULYLQJ WKH IXOOEULGJH '&'& FRQYHUWHU VZLWFKHV DUH SUHVHQWHG 7KH SHULRGV RI SXOVHV $ DQG $ FKDQJHV GHSHQGHGRQWKHGXW\F\FOHGHWHUPLQHGE\WKHGLVFUHWH 3, FRQWUROOHU 7KH PD[LPXP VLPXODWLRQ VWHSVL]H ZDV ȝVDQGWKHGLVFUHWH3,FRQWUROOHUVDPSOHWLPHZDV ȝV Figure 6: Simulation results of the welding process with current control feedback and PWM full-bridge DC-DC converter-based welding power source. 3.3 Simulation of Dynamic Behaviour of a The upper plot shows the change of the CTWD Thyristor Based Weldin Power Source from 16 mm to 18 mm. The forth plot presents 7KH SRZHU RI WKH ZHOGLQJ SRZHU VRXUFH ZLWK D WKUHH the welding current transient response, and on SKDVHWUDQVIRUPHULQDGHOWDVWDUFRQQHFWLRQZDVVHWDW the fifth plot the time response of the primary N9$:HVHOHFWHGWXUQVUDWLREHWZHHQDQG,Q current is shown. WUDQVIRUPHUEORFNZHFDQFKDQJHDOOHVVHQWLDOSDUDPH &RQVWDQW ZHOGLQJ VSHHG ZDV VXSSRVHG 7KH ZHOGLQJ WHUV VXFK DV UHVLVWDQFH DQG LQGXFWDQFH RI WKH SULPDU\ WRUFKZDVSRVLWLRQHGDWPP H IURPZRUNGLVWDQFH DQG VHFRQGDU\ ZLQGLQJV DQG WKH ORVV UHVLVWDQFH LQ WKH &7:' DQGDIWHUPV+ZDVLQFUHDVHGWRPPDV FRUH 7KH WK\ULVWRU EULGJH FRQYHUWHU ZDV FKRVHQ DV LW LVSUHVHQWHGLQWKHILUVWJUDSKRI)LJXUH7KHZHOGLQJ UHGXFHV WKH UHDFWLYH SRZHU ZLWK ILULQJ DQJOH Į EH\RQG

ZLUHIHHGUDWHveZDVVHWDWPPV GHJ Į!GHJ

SNE 26(4) – 12/2016 241 Golob Modeling and Simulation of GMA Welding Process and Welding Power Sources TN

Figure 8: The simulation scheme of a three-phase welding source with thyristor full bridge circuits in Matlab/Simulink. The scope window shows the time graphs of signals within a time window of 100 ms.

)RUILULQJDQJOHVOHVVWKDQGHJWKH'&YROWDJHRI )LJXUHVKRZVWKHRXWSXWYROWDJHFXUUHQWDQGILULQJ

WKH FRQYHUWHU LV DOZD\V SRVLWLYH DQG WKH IUHHZKHHO GL DQJOHĮIRUWKHSXOVHIUHTXHQF\fpRI+]7KHILULQJ RGH GRHV QRW FRPH LQWR RSHUDWLRQ $V WKH ILULQJ DQJOH DQJOHĮOLHVEHWZHHQĮ GHJDQGĮ GHJ7KH

DGYDQFHV EH\RQG WKLV SRLQW WKH ORDG FXUUHQW VWDUWV WR SXOVHZLGWKtPFDQEHYDULHGDVZHOO)LJXUHVKRZVDQ IUHHZKHHOWKURXJKWKHGLRGHWKXVFXWWLQJRIIWKHLQSXW H[DPSOHRIDZLGHUSXOVH OLQHFXUUHQWDQGSUHYHQWLQJWKH'&YROWDJHIURPVZLQJ LQJLQWRWKHQHJDWLYHGLUHFWLRQ7KLVUHGXFHVWKHDPRXQW RIUHDFWLYHSRZHUGUDZQIURPWKHPDLQVWKXVLPSURYLQJ LWVSRZHUIDFWRU>@)LJXUHVKRZVWKHVLPXODWLRQ VFKHPHLQWKHSURJUDP0DWODE6LPXOLQNDQGVLPXODWHG YROWDJHDQGFXUUHQWZDYHIRUPVLQDWLPHZLQGRZRI PV7KHVLPXODWLRQZDVFDUULHGRXWZLWKDILULQJDQJOH RIGHJ

Figure 10: Voltage and current waveforms and variation

of the firing angle by pulse frequency fp = 50 Hz

and pulse width tp = 3.3 ms. 3XOVHGZHOGLQJZLWKGLIIHUHQW SXOVHIUHTXHQFLHVJHQHU DWHVGLIIHUHQWKDUPRQLFVRQWKH$&VLGH7KHKDUPRQLF DUHDFFRPSDQLHGE\WZRDGMDFHQWIUHTXHQFLHVRQWKH$& VLGH>@7KH)RXULHUDQDO\VLVRIWKH$&FXUUHQWFDQ IXUWKHU YHULI\ WKLV )RU H[DPSOH ZHOGLQJ ZLWK D SXOVH Figure 9: Voltage and current waveforms and variation of IUHTXHQF\ RI +] SURGXFHV KDUPRQLFV RI DQG the firing angle by pulse frequency fp = 100 Hz +] LQ WKH DOWHUQDWLQJ FXUUHQW )LJXUH VKRZ WKH and pulse width tp = 6.6 ms. )RXULHUDQDO\VLVRIWKH$&FXUUHQWGXULQJWKHVLPXODWLRQ RIWKHSXOVHGFXUUHQWZHOGLQJSURFHVV

242 SNE 26(4) – 12/2016 T Golob Modeling and Simulation of GMA Welding Process and Welding Power Sources N

References >@ 0RRUH./1DLGX'6@ 0RRUH./1DLGX'62]FHOLN6Modeling, Sensing and Control of Gas Metal Arc Welding.2[IRUG8.(OVH YLHU6FLHQFH/WG Figure 11: Fourier spectrum of the AC current during >@ *RORE0.RYHV$3XNODYHF$7RUYRUQLN%³0RGHO pulsed current welding with pulse frequency fp OLQJVLPXODWLRQDQGIX]]\FRQWURORIWKH*0$:SUR FHVV´LQ&RQI3URFHHGLQJVRIWKHWK,QWHUQDWLRQDO = 50 Hz and pulse width tP = 3.3 ms by changing )HGHUDWLRQRI$XWRPDWLF&RQWURO ,)$& 7ULHQQLDO the firing angle from 0 deg to 60 deg. :RUOG&RQJUHVVRQ$XWRPDWLF&RQWURO%DUFHORQD 6SDLQYROSS 4 Conclusion >@ 7KRPVHQ-6³&RQWURORI3XOVHG*DV0HWDO$UF:HOG LQJ´International Journal of Modelling, Identification $VLPXODWLRQDSSOLFDWLRQKDVEHHQSUHVHQWHGIRUVLPXODW and ControlYROQRSS LQJ WKH *0$: SURFHVV LQYHUWHUEDVHG ZHOGLQJ >@ =KDQJ-:DOFRWW%/³$GDSWLYH,QWHUYDO0RGHO&RQWURO VRXUFHVDQGWK\ULWVRUEDVHGZHOGLQJVRXUFHV7KHPDWK RI$UF:HOGLQJ3URFHVV´IEEE Trans. On Control Sys- HPDWLFDO PRGHO LV EDVHG RQ SK\VLFDO GHVFULSWLRQV RI tems TechnologyYROSS1RY VHYHUDOSDUWVRIWKH*0$:SURFHVVVXFKDVWKHHOHFWULF >@ %HUD0.%DQG\RSDGK\D\%3DXO$.5REXVWQRQOLQH DUFRQWURORI*0$:V\VWHPVDKLJKHURUGHUVOLGLQJ FLUFXLWVRIWKHSRZHUVXSSO\WKHDUFG\QDPLFVWKHHOHF PRGHDSSURDFKIEEE International Conference on In- WURGHPHOWLQJSURFHVVHWF dustrial Technology ,&,7 7KHVLPXODWLRQRILQYHUWHUSRZHUVRXUFHIRUZHOGLQJ >@ 1JR0''X\9+3KXRQJ17.LP+..LP6%³'H SRZHU VXSSO\ KDV EHHQ SURSRVHG DQG WHVWHG WRJHWKHU YHORSPHQWRIGLJLWDOJDVPHWDODUFZHOGLQJV\VWHP´ ZLWK WKH *0$: VLPXODWLRQ PRGHO 7KH VLPXODWLRQ Journal of Materials Processing TechnologyYRO UHVXOWVVKRZHGWKDWWKHFRQYHQWLRQDOIXOOEULGJH'&'& QRSS FRQYHUWHU ZLWK DSSURSULDWH FXUUHQW IHHGEDFN FRQWUROOHU >@ *RORE07RUYRUQLN%³0RGHOOLQJVLPXODWLRQDQGFRQ PDNHV WKH RXWSXW ZHOGLQJ FXUUHQW IROORZ WKH VHW UHIHU WURORIJDVPHWDODUFZHOGLQJ´LQProceedings of the 7th Congress on Modelling and Simulation EUROSIM3UD HQFHV JXH&]HFK5HSXEOLFSS 7KH SURSRVHG PRGHOV DQG VLPXODWLRQV ZKLFK DUH >@ *RORE0,QWHJUDWHG0RGHOVRID*DV0HWDO$5&:HOG FRPELQHGWRJHWKHUWRVLPXODWHWKHSRZHUVRXUFHFLUFXLWV LQJ3URFHVVDQG,QYHUWHUEDVHG3RZHU6XSSO\IRU3URFHVV XVLQJ VLPXODWLRQV RI WKH *0$: SURFHVV DUH VXLWDEOH &RQWURO6LPXODWLRQ6WXGLHVELEKTRONIKA IR EL- IRU WKH GHYHORSPHQW RI QHZ SRZHU VRXUFH FLUFXLWV LH EKTROTECHNIKA UHVRQDQWFRQYHUWHUV%\HVWDEOLVKLQJDSSURSULDWHPRGHOV >@ /HVQHZLFK$³&RQWURORIWKH0HOWLQJ5DWHDQG0HWDO RIWKH*0$:SURFHVVDQGWKHIXOOEULGJH'&'&FRQ 7UDQVIHULQ*DV6KLHOGHG0HWDO$UF:HOGLQJ3DUW´ YHUWHUPRGHOVLPXODWLRQLVDQHIIHFWLYHWRROIRULQYHVWL Welding JournalYROSSVV JDWLQJ QHZ ZHOGLQJ WHFKQRORJLHV IRU H[DPSOH WKH >@ 7XVHN-6XEDQ0³'HSHQGHQFHRI0HOWLQJ5DWHLQ 3XOVHG *0$: SURFHVV RU 6XUIDFH 7HQVLRQ 7UDQVIHU 0,*0$*:HOGLQJRQWKH7\SHRI6KLHOGLQJ*DV 8VHG´Journal of Materials Processing TechnologyYRO ZHOGLQJ SURFHVV 677 6LPXODWLRQ UHVXOWV FRXOG EH SS YHU\ XVHIXO IRU WKH UDSLG GHYHORSPHQW RI QHZ FRQWURO >@ +DOP¡\(³:LUHPHOWLQJUDWHGURSOHWWHPSHUDWXUHDQG DOJRULWKPVDQGIRUWKHGHVLJQLQJRIQHZLQYHUWHUFRQWURO HIIHFWLYHDQRGHSRWHQWLDO´LQProceedings of the Interna- XQLWV tional Conference on Arc Physics and Weld Pool Behav- iou,/RQGRQ(QJODQGSS

SNE 26(4) – 12/2016 243 Golob Modeling and Simulation of GMA Welding Process and Welding Power Sources TN

>@ .UHMFDU26SLFND,)ULVFKHU5³,PSOHPHQWDWLRQRI >@ /DQJXV'.UDOM9LQ*UXP-2SWLPLVDWLRQRIZHOGLQJ )XOO)HDWXUHG3,'5HJXODWRULQ0LFURFRQWUROOHUV´Elec- SDUDPHWHUVLQSXOVHG0,*0$*ZHOGLQJZLGWK tronics and Electrical EngineeringYROQRSS FRQWUROOHGVLQHZDYHFXUUHQWSXOVHV3DUW'HWHUPLQD ± WLRQRIDJHQHUDOV\QHUJHWLFHTXDWLRQDQGDQRUPDOLVHG >@ 3HWURYDV$/LVDXVNDV65LQNHYLFLHQH5³'LJLWDO$X SDUDPHWULFGLDJUDPZLWKDGHILQHGSDUDPHWULFZHOGLQJ WRPDWLF&RQWURO6\VWHPZLWK3,'&RQWUROOHU´Electron- UDQJH,QWMPDWHUSURGWHFKQROOHWQãW ics and Electrical EngineeringYROQRSS± VWU >@ /DQJXV'.UDOM9LQ*UXP-2SWLPLVDWLRQRIZHOGLQJ >@ 7KDPRGKDUDQ0%HFN+3LQ:ROI$6WHDG\DQG SDUDPHWHUVLQSXOVHG0,*0$*ZHOGLQJZLGWK 3XOVHG'LUHFW&XUUHQW:HOGLQJZLWKD6LQJOH&RQYHUWHU FRQWUROOHGVLQHZDYHFXUUHQWSXOVHV3DUW'HWHUPLQD Supplement to the Weldig Journal0DUFKVV WLRQRIDQRSWLPXPPDWHULDOWUDQVIHUWKURXJKWKHDUFDQG DFRQWUROPHWKRG,QWMPDWHUSURGWHFKQROOHWQ ãWVWU

244 SNE 26(4) – 12/2016 SNE T ECHNICAL N OTE

Inverse Simulation Methods Applied to Investigations of Actuator Nonlinearities in Ship Steering

David J. Murray-Smith

Emeritus Professor and Honorary Senior Research Fellow, School of Engineering, Rankine Building, University of Glasgow, Glasgow G12 8QQ, Scotland, United Kingdom; [email protected]

6LPXODWLRQ1RWHV(XURSH61( ± $OWKRXJK DQDO\WLFDO DSSURDFKHV WR PRGHO LQYHUVLRQ '2,VQHWQ DUH RI JUHDW YDOXH WKH\ FDQ SUHVHQW GLIILFXOWLHV ZLWK 5HFHLYHG2FWREHU PDQ\IRUPVRIQRQOLQHDUPRGHO,QUHFHQW\HDUVH[WHQ $FFHSWHG'HFHPEHU 6SHFLDO,VVXH5HYLHZ VLYH XVH KDV EHHQ PDGH RI VLPXODWLRQ WHFKQLTXHV IRU ILQGLQJLQYHUVHVROXWLRQVUDWKHUWKDQGHSHQGLQJHQWLUHO\ Abstract. Actuators associated with control surfaces in RQDQDO\WLFDOPHWKRGVRILQYHUVLRQ([DPSOHVRIDSSOL aircraft, ships and underwater vehicles often introduce FDWLRQV RI WKLV NLQG LQFOXGH DLUFUDIW KDQGOLQJ TXDOLWLHV problems in terms of the control characteristics of the LQYHVWLJDWLRQV DQG DJLOLW\ VWXGLHV ERWK IRU IL[HGZLQJ vehicle if significant saturation and rate limiting effects DLUFUDIWDQGKHOLFRSWHUV VHHHJ>@>@ ,QVXFKFDVHV are present. Rate limits, in particular, have been linked to WKHLQYHUVHVROXWLRQSURYLGHVYLWDOLQIRUPDWLRQDERXWWKH a number of well-publicised safety and handling-qualities UHODWLYH GLIILFXOW\ RI SHUIRUPLQJ GLIIHUHQW PDQRHXYUHV issues for aircraft. Such limits also present difficulties in DQG DERXW FRQWURO PDUJLQV DYDLODEOH DV DFWXDWRU DPSOL ship steering and ship autopilot systems. This paper WXGHRUUDWHOLPLWVDUHDSSURDFKHG,QUHFHQW\HDUVPXFK describes an investigation of the effects of actuator non- SURJUHVVKDVDOVREHHQPDGHLQXVLQJLQYHUVHVLPXODWLRQ linearities involving a ship steering control application. PHWKRGVLQFRQWUROV\VWHPGHVLJQDSSOLFDWLRQV VHHHJ The method of approach involves the use of inverse >@>@ simulation to detect the onset of limiting. The paper shows that inverse simulation methods allow direct prediction of situations in which rudder saturation and 1 Models of Actuators and Ship rate limiting have significant effects in terms of the ma- Steering Dynamics noeuvrability of the vessel. It is also shown that a two- stage inverse-simulation method allows direct assess- 'HWDLOHG SK\VLFDOO\EDVHG PRGHOV RI DFWXDWRUV RI ment of the difference between desired and achievable YDULRXVNLQGVDUHDYDLODEOHLQWKHOLWHUDWXUHDQGZKHWKHU manoeuvres. WKH DFWXDWRUV DUH K\GUDXOLF HOHFWURK\GUDXOLF RU HOHFWULFDO LQ IRUP WKH DFWXDWRU V\VWHPV KDYH ZHOO GHILQHG DPSOLWXGH DQG UDWH OLPLWV $ORQJ ZLWK WKH Introduction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

SNE 26(4) – 12/2016 245 Murray-Smith Inverse Simulation Methods Applied to Ship Steering TN

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igure 1: Simplified block diagram of actuator system OLQHDUO\ ,Q WKLV FDVH WKH EORFN ODEHOOHG Gr ZRXOG QR with amplitude and rate limiting. ORQJHUEHDVLPSOHJDLQIDFWRUEXWZRXOGKDYHILUVWRUGHU 7KHEDVLFPRGHOLVJLYHQE\ ODJ FKDUDFWHULVWLFV ,Q SULQFLSOH DFFHODUDWLRQ OLPLWV DV

ZHOODVUDWHOLPLWVFRXOGEHLQFRUSRUDWHGLQWRWKLVW\SHRI ܶ߰ሷ ൅ܪேሺ߰ሶሻ Kį t EORFNGLDJUDPVWUXFWXUHEXWWKLVKDVQRWEHHQFRQVLGHUHG ZKHUHWKHYDULDEOHȥLVWKH\DZDQJOH KHDGLQJ RIWKH LQ WKH SUHVHQW LQYHVWLJDWLRQ 7KH VWUXFWXUH VKRZQ LQ YHVVHOįLVWKHUXGGHUDQJOHT LVDQLQHUWLDFRQVWDQWDQG )LJXUHWKXVUHSUHVHQWVDJHQHUDOIRUPRIPRGHOZKLFK WKH IXQFWLRQ ܪ ZKLFK LV D IXQFWLRQ RI WKH UDWH RI LV FDSDEOH RI GHVFULELQJ WKH OLQHDU DQG QRQOLQHDU ே FKDQJHRIKHDGLQJሺ߰ሶሻLVJLYHQE\ FKDUDFWHULVWLFVRIDZLGHUDQJHRIDFWXDWRUVLQDVLPSOH ଷ IDVKLRQ DQG LV DSSURSULDWH IRU DSSOLFDWLRQV LQYROYLQJ ܪேሺ߰ሶሻൌ݊ଵ߰ሶ ൅݊ଷሺ߰ሶሻ PDULQHYHKLFOHVRUDLUFUDIW ZKHUH݊ ݊ DUHSRVLWLYHGDPSLQJFRQVWDQWVNQRZQ :LWKLQWKHEORFNGLDJUDPRI)LJXUHWKHVDWXUDWLRQ ଵ ଷ DV 1RUUELQ FRHIILFLHQWV )RU WKH VSHFLILF FDVH RI WKH OLPLWEORFNKDVDVLPSOHIRUPDQGZKHQWKHLQSXWGaįc R.O.V. Zeefakkel ZKLFK LV D P ORQJ WUDLQLQJ VKLS OLHVLQWKHUDQJHEHWZHHQWKHXSSHUDQGORZHUVDWXUDWLRQ EHORQJLQJ WR WKH 5R\DO 1HWKHUODQG 1DYDO &ROOHJH WKH OLPLWV įcUDQGįcL LWEHKDYHVDVDOLQHDUJDLQHOHPHQW SDUDPHWHUV݊ଵ݊ଷKDYHEHHQHVWLPDWHGIRUDQXPEHU KDYLQJXQLW\JDLQ+RZHYHUZKHQWKHLQSXWGaįFįF8 RIGLIIHUHQWIRUZDUGVSHHGV>@&RPELQLQJ(TQV DQG WKHRXWSXWYDOXHLVOLPLWHGDWįF8DQGFRUUHVSRQGLQJO\ JLYHV ZKHQGDįcįF/WKHRXWSXWLVOLPLWHGDWįF/)RUPDQ\ FDVHVRISUDFWLFDOLPSRUWDQFHWKLVOLPLWLQJEHKDYLRXULV ଷ ߜൌ݉߰ሷ ൅݀ଵ߰ሶ ൅݀ଷሺ߰ሶሻ V\PPHWULFDOIRUSRVLWLYHDQGQHJDWLYHLQSXWVDQGį F8 ் ௡ ௡ ZKHUH m ݀ ൌ భ DQG ݀ ൌ య 9DOXHV RI WKHVH įF/7KH UDWH OLPLW EORFNKDVDQ LGHQWLFDO IRUP KDYLQJ ௄ ଵ ௄ ଷ ௄ XQLW\ JDLQ ZKHQ WKH RXWSXW RI WKH EORFN WKH UDWH RI SDUDPHWHUV YDU\ VLJQLILFDQWO\ IRU W\SLFDO VSHHG YDOXHV FKDQJH ߜሶ t RI WKH DFWXDWRU RXSXW SRVLWLRQ į t KDV RYHU WKH UDQJH RI LQWHUHVW IRU WKLV YHVVHO DV VKRZQ LQ YDOXHVWKDWOLHZLWKLQWKHVSHFLILHGXSSHUDQGORZHU 7DEOH DFWXDWRU UDWH OLPLWV 7KH UDWH OLPLW EORFN JLYHV D ,Q WKLV DSSOLFDWLRQ WKH UXGGHU DQG LWV DVVRFLDWHG FRQVWDQW RXWSXW HTXDO WR WKH SRVLWLYH RU QHJDWLYH UDWH DFWXDWRU DUH PRGHOOHG XVLQJ WKH ILUVWRUGHU ODJ W\SH RI OLPLWZKHQߜሶ t KDVDYDOXHEH\RQGWKHVSHFLILHGXSSHU GHVFULSWLRQ ZLWK LQSXW VDWXUDWLRQ DQG UDWH OLPLWV DV DQGORZHUDFWXDWRUUDWHOLPLWV VKRZQ LQ )LJXUH,Q WKH OLQHDU PRGHRIRSHUDWLRQ DQ 7KH W\SH RI DFWXDWRU PRGHO RXWOLQHG DERYH FDQ EH DFWXDWRUWLPHFRQVWDQWRIVLVJLYHQE\DYDOXHRIWKH XVHG ZLWK PDQ\ GLIIHUHQW IRUPV RI VKLS PRGHO 2QH JDLQIDFWRUGrRI QRQOLQHDU IRUP RI PRGHO ZKLFK LV FRPPRQO\XVHG WR ,I WKH UHTXLUHG UXGGHU GHIOHFWLRQ įc LV WKH YDULDEOH UHSUHVHQW WKH PDQRHXYULQJ FKDUDFWHULVWLFV RI FRXUVH VXEMHFWHG WR OLPLWLQJ WKHJDLQ IDFWRU GaLQ)LJXUHLV VWDEOHVKLSVLQ\DZLVDQH[WHQGHGIRUPRI1RPRWR¶V XQLW\ )RU WKH SXUSRVHV RI WKLV LQYHVWLJDWLRQ WKH ILUVWRUGHU PRGHO >@ >@ ZKLFK UHODWHV KHDGLQJ VDWXUDWLRQOLPLWIRUWKHUXGGHULVW\SLFDOO\RIWKHRUGHURI YDULDEOHVWRWKHUXGGHUDQJOH GHJZKLOHWKHWZRGLIIHUHQWUDWHOLPLWYDOXHVXVHGLQ WKH LOOXVWUDWLYH H[DPSOHV WKDW IROORZ DUH GHJV DQG GHJV

246 SNE 26(4) – 12/2016 T N Murray-Smith Inverse Simulation Methods Applied to Ship Steering

)RUZDUG 2WKHU LWHUDWLYH WHFKQLTXHV ZHUH DOVR GHYHORSHG IRU VSHHG T K ܶ d d ݉ൌ VLPLODU DSSOLFDWLRQV LQFOXGLQJ RSWLPL]DWLRQEDVHG DS UPV ܭ SURDFKHV E\ &HOL >@ DQG E\ /HH DQG .LP >@ 7KH SDSHU E\ 7KRPVRQ DQG %UDGOH\ >@ SURYLGHV D XVHIXO

RYHUYLHZ RI D QXPEHU RI WKHVH LWHUDWLYH WHFKQLTXHV DV GHYHORSHGLQLWLDOO\IRUDHURQDXWLFDODSSOLFDWLRQV,QYHUVH VLPXODWLRQWHFKQLTXHVEDVHGRQGLVFUHWHIRUPVRIPRGHO Table 1: Parameter values used for the model of R.O.V. KDYHDOVREHHQXVHGIRUWKHGHVLJQRIPRGHOEDVHGRXW Zeefakkel [8]. SXWWUDFNLQJ FRQWURO V\VWHPV DQG D SDSHU E\ /X 0XU UD\6PLWK DQG 0F*RRNLQ >@ GHVFULEHV WKH XVH RI LQ 2 Inverse Simulation Methods YHUVH VLPXODWLRQ LQ WKH GHVLJQ RI IHHGIRUZDUG FRQWURO V\VWHPVEDVHGRQD/\Q[KHOLFRSWHUPRGHODQGDOVRIRU ,QYHUVH VLPXODWLRQ WHFKQLTXHV PD\ EH GLYLGHG FRPELQHG VWHHULQJ FRQWURO DQG UROO VWDELOLVDWLRQ LQ D FRQYHQWLHQWO\LQWRPHWKRGVWKDWDUHEDVHGRQGLVFUHWLVHG FRQWDLQHUVKLSDSSOLFDWLRQ PRGHOV DQG DUH HVVHQWLDOO\ LWHUDWLYH LQ QDWXUH DQG WHFKQLTXHV WKDW DUH EDVHG RQ FRQWLQXRXV V\VWHP 2.2 The continuous system simulation VLPXODWLRQSULQFLSOHV$OWKRXJKWKHHPSKDVLVZLWKLQWKLV SDSHULVRQXVHRIRQHRIWKHVHFRQGJURXSRIPHWKRGV approach ERWK W\SHV RI DSSURDFK DUH UHYLHZHG KHUH VLQFH VRPH $OWKRXJK WKH LWHUDWLYH W\SH RI DSSURDFK KDV EHHQ XVHG FRQWLQXRXV V\VWHP VLPXODWLRQ DSSURDFKHV KDYH RULJLQV ZLWK FRQVLGHUDEOH VXFFHVV LQ D QXPEHU RI DHURQDXWLFDO LQLWHUDWLYHPHWKRGVLQYROYLQJGLVFUHWLVHGPRGHOV DSSOLFDWLRQVDVHFRQG DQGHQWLUHO\GLIIHUHQW DSSURDFK WR WKH GHYHORSPHQW RI LQYHUVH VLPXODWLRQ PHWKRGV KDV 2.1 Iterative methods of inverse simulation HYROYHGZKLFKLVEDVHGRQWKHXVHRIFRQWLQXRXVV\VWHP VLPXODWLRQ SULQFLSOHV DQG DYRLGV WKH QHHG IRU LWHUDWLYH based on discrete models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µLQWHJUDWLRQEDVHG¶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µGLIIHUHQWLD WKHIHHGEDFNPHWKRGFKDQJHVZLWKLQWKHPRGHOFDQEH WLRQ¶PHWKRGLQZKLFKDFRQWLQXRXVV\VWHPPRGHORIWKH LQFRUSRUDWHGZLWKRXWFKDQJHVLQWKHIHHGEDFNVWUXFWXUH JLYHQ V\VWHP LV WUDQVIRUPHG LQWR D GLVFUHWHWLPH GH RWKHUWKDQSRVVLEOHDGMXVWPHQWVRIVRPHIHHGEDFNORRS VFULSWLRQWKURXJKWKHXVHRIDILQLWHGLIIHUHQFHDSSUR[L JDLQV )RUWKLVUHDVRQWKHIHHGEDFNDSSURDFKKDVEHHQ PDWLRQ7KLVDSSURDFKZDVGHYHORSHGE\7KRPVRQDQG FKRVHQIRUWKHZRUNGHVFULEHGLQWKLVSDSHU KLV FROOHDJXHV VHH HJ >@ >@ LQ WKH FRQWH[W RI KHOLFRSWHUDSSOLFDWLRQVDQGE\.DWRDQG6XJLXUD>@IRU IL[HGZLQJDLUFUDIWSUREOHPV

SNE 26(4) – 12/2016 247 Murray-Smith Inverse Simulation Methods Applied to Ship Steering TN

2.3 Principles of the feedback approach 6RPHRIWKHHDUOLHVWGHYHORSPHQWVLQLQYHUVHVLPXODWLRQ LQYROYLQJWKHXVHRIIHHGEDFNSULQFLSOHVFDQEHIRXQGLQ ZRUNFDUULHGRXWDWWKH'/5DHURQDXWLFDOUHVHDUFKLQVWL WXWHDW%UDXQVFKZHLJLQ*HUPDQ\DVRXWOLQHGE\+DPHO VHHHJ>@ DQGGLVFXVVHGLQPRUHGHWDLOE\*UD\DQG YRQ*UQKDJHQ>@DQGE\%XFKKRO]DQGYRQ*UQKD JHQ>@7KHVHPHWKRGVKDYHPRUHUHFHQWO\EHHQXVHG Figure 2: Block diagram for inverse simulation using LQ D QXPEHU RI DSSOLFDWLRQV LQYROYLQJ DLUFUDIW SURFHVV feedback principles for a given linear or V\VWHPV DQG XQGHUZDWHU YHKLFOH PRGHOV VHH HJ > nonlinear model G. For a high value of the gain @>@ K, the variable w is a close approximation to $VLPLODUW\SHRIDSSURDFKZKLFKLVOLQNHGVSHFLIL the model input required to produce an FDOO\ WR FRQWURO V\VWHP GHVLJQ KDV EHHQ GHYHORSHG E\ output that matches a given time history v(t). 7DJDZD DQG )XNXL >@ 7KHLU RYHUDOO DSSURDFK LV ,Q LWV RULJLQV WKH IHHGEDFNEDVHG DSSURDFK FDQ EH WHUPHGµLQYHUVHG\QDPLFVFRPSHQVDWLRQYLDVLPXODWLRQ OLQNHG EDFN WR WKH XVH RI IHHGEDFN SULQFLSOHV IRU GLYL RIIHHGEDFNFRQWUROV\VWHPV¶ ,'&6 DQGWKHGHUYDWLRQ VLRQDQGLQYHUVHIXQFWLRQJHQHUDWLRQRSHUDWLRQVLQHOHF RIDQLQYHUVHVLPXODWLRQWKURXJKWKHXVHRIIHHGEDFNLVD WURQLF DQDORJ FRPSXWHUV 5HFHQW ZRUN KDV VKRZQ WKDW FHQWUDO HOHPHQW RI WKLV FRQWURO GHVLJQ PHWKRGRORJ\ WKHDSSURDFKKDVYHU\ZLGHDSSOLFDELOLW\>@DQGWKDWLW 7KH\ KDYH XVHG WKH ,'&6 PHWKRG LQ FRQWURO V\VWHP DOORZV DQDO\VLV RI WKH GHSHQGHQFH RI LQYHUVH VROXWLRQV GHVLJQDSSOLFDWLRQVLQYROYLQJVHUYRK\GUDXOLFDFWXDWRUV RQSDUDPHWHUVRIWKHIRUZDUGPRGHO ZLWKRXWSDUDPHWHU DQGURERWLFVDVGHVFULEHGLQUHFHQWSDSHUV>@>@ SHUWXUEDWLRQ WKURXJKWKHXVHRIVHQVLWLYLW\PRGHOV>@ 7KH IHHGEDFN DSSURDFK WR LQYHUVH VLPXODWLRQ FDQ 7KLVFDQKDYHDGYDQWDJHVHVSHFLDOO\LQWKHOLQHDUFDVH EHVW EH XQGHUVWRRG E\ FRQVLGHULQJ WKH FDVH RI D OLQHDU LQ WHUPV RI WKH DGGLWLRQDO SK\VLFDO LQVLJKW SURYLGHG PRGHO7KHEORFNGLDJUDPRI)LJXUHLQYROYHVDVLQ ZKHQ FRPSDUHG ZLWK SDUDPHWHU SHUWXUEDWLRQ PHWKRGV JOHLQSXW VLQJOHRXWSXW OLQHDU PRGHO G s DQG D IHHG IRUVHQVLWLYLW\LQYHVWLJDWLRQ EDFN ORRSKDYLQJ D FDVFDGHG EORFNZLWK WUDQVIHU IXQF 2QH SRWHQWLDO SUREOHP LQ DSSO\LQJ WKH IHHGEDFN WLRQ K s 7KH WUDQVIHU IXQFWLRQ UHODWLQJ WKH YDULDEOH EDVHG DSSURDFK WR SUREOHPV LQYROYLQJ DFWXDWRU VDWXUD W s WRDUHIHUHQFHLQSXWV V LVJLYHQE\ WLRQ DQG UDWH OLPLWV FRQFHUQV GLIILFXOWLHV DULVLQJ IURP

ௐሺ௦ሻ ଵ SRVVLEOH OLPLW F\FOH HIIHFWV +DUG QRQOLQHDULWLHV RI WKH భ ௏ሺ௦ሻ ାீሺ௦ሻ W\SH WKDW DULVH LQ DFWXDWRUV FDQ JLYH ULVH WR OLPLW F\FOH ಼ሺೞሻ SKHQRPHQDZLWKLQDQ\IHHGEDFNORRS)RUVLQJOHLQSXW ,IWKHWHUPK s LVYHU\VPDOOFRPSDUHGZLWKWKHPDJ VLQJOHRXWSXW IHHGEDFN V\VWHPV GHVFULELQJ IXQFWLRQ QLWXGHRIG s RYHUWKHUDQJHRIIUHTXHQFLHVRILQWHUHVW DQDO\VLV PHWKRGV VHH HJ >@ > @ FDQ EH XVHG WR WKHWUDQVIHUIXQFWLRQPD\EHDSSUR[LPDWHGE\ SUHGLFW WKH H[LVWHQFH RI OLPLW F\FOHV IRU IHHGEDFN V\V WHPV ZKLFK LQYROYH RQH GRPLQDQW QRQOLQHDULW\ DQG ௐሺ௦ሻ ଵ ൎ ௏ሺ௦ሻ ீሺ௦ሻ RWKHUZLVH FDQ EH GHVFULEHG DGHTXDWHO\ E\ OLQHDU G\ QDPLF HOHPHQWV ZLWKLQ WKH IHHGEDFN ORRS 7KH FRQGL 7KXVLIK s LVODUJHWKHWUDQVIHUIXQFWLRQW s V s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

248 SNE 26(4) – 12/2016 T N Murray-Smith Inverse Simulation Methods Applied to Ship Steering

7KLVPHDQVWKDWSUREOHPVRIOLPLWF\FOHVDUHOLNHO\ W (actuator input WR EH HQFRXQWHUHG LQ DWWHPSWLQJ WR DSSO\ WKH IHHGEDFN from the inverse DSSURDFK WR LQYHUVH VLPXODWLRQ LQ WKH FDVH RI DSSOLFD simulation) WLRQV LQYROYLQJ VLJQLILFDQW UDWH OLPLWV 7KHUHIRUH LQ VXFKFDVHVVRPHPRGLILFDWLRQVWRWKHVWDQGDUGIHHGEDFN + HR Ref. K A DSSURDFKPD\EHQHFHVVDU\RUHQWLUHO\GLIIHUHQWPHWKRGV Model V RILQYHUVHVLPXODWLRQPD\KDYHWREHDSSOLHGWKDWGRQRW - - LQYROYHWKHXVHRIIHHGEDFN L

3 Inverse Simulation Applied to the Ship Model Figure 3: Block diagram of the feedback system used for 7KH ILUVW DSSURDFK FRQVLGHUHG LQYROYHV WKH DSSOLFDWLRQ inverse simulation with the actuator RIWKHVLPSOHIHHGEDFNPHWKRGRILQYHUVHVLPXODWLRQDV sub-model incorporated within the feedback RXWOLQHG LQ 6HFWLRQ DQG GLVFXVVHG LQ JUHDWHU GHWDLO loop. Here the block A represents the actuator and V represents the vehicle. The variable HR HOVHZKHUH VHHHJ>@>@>@ is the vehicle heading rate. The reference model generates the time history of the 3.1 Feedback applied to the ship model with desired manoeuvre in terms of the required the actuator sub-model included. heading-rate time history. The block shown as )LJXUHLVDEORFNGLDJUDPZKLFKVKRZVWKHVWUXFWXUH having a gain factor L is a subsidiary feedback RIWKHIHHGEDFNV\VWHPZKLFKLVDSSOLHGDURXQGWKHVKLS loop and, in the case of the application PRGHO LQFOXGLQJ WKH DFWXDWRU VXEPRGHO ZKLFK LQ WKH considered here, involves angular acceleration JHQHUDOFDVHLQFRUSRUDWHVVDWXUDWLRQDQGUDWHOLPLWV7KH feedback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îZKLOHDQ RUWKHGHVLUHGKHDGLQJ,QWKHFDVHLQYROYLQJWKHGHVLUHG DSSURSULDWHYDOXHIRUWKHJDLQIDFWRUL LQWKHVXEVLGLDU\ KHDGLQJ WKH VWUXFWXUH DQG SDUDPHWHU YDOXHV RI WKLV ORRSLQYROYLQJIHHGEDFNRIWKHDQJXODUDFFHOHUDWLRQZDV UHIHUHQFH PRGHO DUH FKRVHQ WR JLYH D UHIHUHQFH VLJQDO IRXQGWREHî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

SNE 26(4) – 12/2016 249 Murray-Smith Inverse Simulation Methods Applied to Ship Steering TN

Figure 5: Heading change corresponding to the Figure 4: Reference input applied to the feedback heading-rate reference signal of Figure 4. system for the case of the ship model with forward speed of 5 m/s and a demanded heading change of 8 deg. 7KHUHVXOWVLQ)LJXUHVDQGVKRZWKDWIRUWKHFKRVHQ PDQRHXYUHDQGIRUZDUGVSHHGFRQGLWLRQWKHUXGGHUGLG QRWDSSURDFKLWVDQJXODUVDWXUDWLRQOLPLWRIGHJRU LWV DQJXODU UDWH OLPLW RI GHJV :KHQ DSSOLHG DV LQSXW WR WKH IRUZDUG VLPXODWLRQ PRGHO WKH UXGGHU GHIOHFWLRQ IRXQG IURP LQYHUVH VLPXODWLRQ DV VKRZQ LQ )LJXUH SURGXFHG D KHDGLQJUDWH UHVSRQVH ZKLFK PDWFKHG DOPRVW H[DFWO\ WKH UHTXLUHG KHDGLQJ UDWH ZLWK KHDGLQJUDWHHUURUVOHVVWKDQîGHJVRYHUWKH VHFRQGUHVSRQVHWLPHFRQVLGHUHG DVVKRZQLQ)LJXUH 7KLV FRUUHVSRQGV WR D PD[LPXP KHDGLQJ HUURU DV VKRZQ LQ )LJXUH RI DSSUR[LPDWHO\ î GHJ Figure 6: Rudder angle time history found using the (UURUV LQ KHDGLQJ DQJOH DQG KHDGLQJ UDWH ZRXOG RI inverse simulation process for the ship model FRXUVHEHVOLJKWO\GLIIHUHQWIRURWKHUYDOXHVRIJDLQ with forward speed of 5 m/s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igure 7: Rudder angular velocity time history found using the inverse simulation process for the ship model with forward speed of 5 m/s.

250 SNE 26(4) – 12/2016 T N Murray-Smith Inverse Simulation Methods Applied to Ship Steering

Figure 11: Rudder angular-rate time history found for forward speed of 2.6 m/s. Other conditions for Figure 8: The difference between the heading-rate this simulation are the same as for the reference input and the heading-rate found previous results. from a forward simulation using the rudder deflection time history of Figure 6. ,IWKHGHPDQGHGKHDGLQJFKDQJHLVQRZPDGHODUJHUIRU WKHIRUZDUGVSHHGRIPVWKHVDWXUDWLRQDQGUDWHOLPLWV ERWKEHFRPHLPSRUWDQW)LJXUHVDQGVKRZUHVXOWV IRUDIRUZDUGVSHHGRIPVDQGDGHVLUHGPDQRHXYUH LQYROYLQJDILQDOFRXUVHFKDQJHRIGHJ&OHDUO\ WKH DFWXDWRUSRVLWLRQ UXGGHUDQJOH QRZH[FHHGVWKHGHJ VDWXUDWLRQ OLPLW DQG WKH DQJXODU YHORFLW\ DOVR H[FHHHGV WKH GHJV UDWH OLPLW 7KLV SURFHGXUH JLYHV D FOHDU LQGLFDWLRQ RI VLWXDWLRQV ZKHUH GHPDQGHG PDQRHXYUHV H[FHHG WKH KDUG OLPLWV RI WKH DFWXDWRU DQGFRXOG FDXVH SUREOHPV LQ WHUPV RI WKH FRQWURO FKDUDFWHULVWLFV RI WKH YHVVHO 7KXV LI WKH SXUSRVH RI WKH LQYHVWLJDWLRQ LV WR HVWDEOLVKZKHWKHURUQRWDVSHFLILFPDQRHXYUHJLYHVULVH WR VDWXUDWLRQ RU UDWH OLPLWLQJ WKH LQYHUVH VLPXODWLRQ PRGHO LQYROYLQJ D OLQHDU DFWXDWRU VXEPRGHO FDQ Figure 9: The error in heading corresponding to the re- SURYLGHXVHIXOLQIRUPDWLRQ sults shown in Figure 8.

Figure 12: Rudder angle time history found for forward Figure 10: Rudder angle time history found for forward speed of 5 m/s for a demanded manoeuvre speed of 2.6 m/s. Other conditions for this corresponding to a 40 deg heading change. simulation are the same as for the previous Other conditions for this inverse simulation results. are the same as for the previous results.

SNE 26(4) – 12/2016 251 Murray-Smith Inverse Simulation Methods Applied to Ship Steering TN

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igure 13: Rudder angular-rate time history found for UDWH IHHGEDFNWHQGV WR PDNH WKLV OLPLW F\FOH EHKDYLRXU forward speed of 5 m/s s for a demanded HYHQPRUHSURQRXQFHG manoeuvre corresponding to a 40 deg ,Q DSSOLFDWLRQV ZKHUH LQYHVWLJDWLRQ RI WKH HIIHFW RI heading change. DFWXDWRUOLPLWVRQWKHRYHUDOOG\QDPLFFKDUDFWHULVWLFVRI WKH FRPSOHWH YHKLFOH LV LPSRUWDQW VRPH ZD\ PXVW EH ,WFDQDOVREHFRQFOXGHGWKDWIRUOHVVGHPDQGLQJLQSXWV IRXQGRILQFRUSRUDWLQJWKHQRQOLQHDUDFWXDWRUVXEPRGHO VXFKDVWKHGHJPDQRHXYUHFRQVLGHUHGLQ)LJXUHV ZLWKLQWKHLQYHUVHVLPXODWLRQSURFHGXUH,QYLHZRIWKH LQYHUVHVLPXODWLRQPHWKRGVGRSURYLGHDGLUHFWDQGFOHDU OLPLW F\FOH SUREOHPV HQFRXQWHUHG ZKHQ WKH QRQOLQHDU LQGLFDWLRQRIWKHPDUJLQVRIFRQWURODYDLODEOH,QWHUPV DFWXDWRU VXEPRGHO LV LQFOXGHG ZLWKLQ WKH LQYHUVH RI WKH VDWXUDWLRQ OLPLW WKLV PDUJLQ LV IRXQG IURP WKH VLPXODWLRQ DV UHSRUWHG DERYH WKH VLPSOH IHHGEDFN GLIIHUHQFHEHWZHHQWKHPD[LPXPUXGGHUGHIOHFWLRQDQG DSSURDFKLVFOHDUO\LQDSSURSULDWH2QHSRVVLEOHVWUDWHJ\ WKHVDWXUDWLRQOLPLW)RUWKHUDWHOLPLWWKHFRUUHVSRQGLQJ LV RXWOLQHG LQ WKH QH[W VHFWLRQ DQG LQYROYHV D PDUJLQ LV IRXQG E\ FRPSDULQJ UXGGHU DQJXODU YHORFLW\ FRPELQDWLRQ RI LQYHUVH VLPXODWLRQ DQG FRQYHQWLRQDO YDOXHVRYHUWKHFRPSOHWHWLPHKLVWRU\ZLWKWKHUDWHOLPLW IRUZDUGVLPXODWLRQLQDWZRVWDJHSURFHGXUH>@ YDOXH ,Q )LJXUH WKH PD[LPXP UXGGHU GHIOHFWLRQ LV DERXWGHJFRPSDUHGZLWKWKHVDWXUDWLRQOLPLWRI GHJ DQG WKHUH LV WKHUHIRUH D ODUJH PDUJLQ RI FRQWURO

GHJRIUXGGHUGHIOHFWLRQ EHIRUHWKHKHOPVPDQRU DXWRSLORW V\VWHP ZRXOG HQFRXQWHU SUREOHPV 6LPLODUO\ WKHUHVXOWVRI)LJXUHVKRZWKDWWKHUDWHRIFKDQJHRI UXGGHUDQJXODUGHIOHFWLRQRIDERXWGHJVLVEHORZWKH FULWLFDO OHYHO RI GHJV DQG WKLV VXJJHVWV WKDW WKH PDQRHXYUH FRXOG EH PDGH VOLJKWO\ PRUH GHPDQGLQJ

EHIRUH GLIILFXOWLHV GXH WR UDWH OLPLWV ZRXOG EH 5XGGHUDQJOH GHJ

HQFRXQWHUHG 7KHDYDLODELOLW\RILQIRUPDWLRQRIWKLVNLQGLVFOHDUO\ XVHIXO LQ DVVHVLQJ WKH PDQHXYUDELOLW\ RI D VSHFLILF YHKLFOHRULQFRQVLGHULQJVSHFLILFGHVLJQFKDQJHV VXFK DVZLWKLQWKHDFWXDWRUDQGUXGGHUV\VWHP 7LPH V ,IVDWXUDWLRQDQGUDWHOLPLWVDUHLQFOXGHGZLWKLQWKH Figure 14: Results in terms of rudder angle obtained by DFWXDWRUPRGHOWKHIHHGEDFNVWUXFWXUHXVHGIRULQYHUVH inverse simulation using the feedback VLPXODWLRQ FKDQJHV LWV EHKDYLRXU VLJQLILFDQWO\ LQ approach for the case of the ship model with PDQRHXYUHVIRUZKLFKDFWXDWRUOLPLWVDUHH[FHHGHG forward speed of 2.6 m/s and a demanded heading change of 30 deg with a rudder saturation limit of ±35 deg and rudder rate limit of ±7 deg/s.

252 SNE 26(4) – 12/2016 T N Murray-Smith Inverse Simulation Methods Applied to Ship Steering

3.2 A two-stage feedback method )LJXUH LV D EORFN GLDJUDP LOOXVWUDWLQJ D WZRVWDJH PHWKRGZKLFKDOORZVWKHIHHGEDFNDSSURDFKWREHXVHG EXWZKLFKDYRLGVWKHOLPLWF\FOHSUREOHPVHQFRXQWHUHG ZLWK WKH WUDGLWLRQDO PHWKRG LQ ZKLFK WKH QRQOLQHDU

DFWXDWRUPRGHOLVLQFOXGHGZLWKLQWKHIHHGEDFNORRS$V EHIRUH IHHGEDFN LV DSSOLHG DURXQG WKH VKLS PRGHO WR DOORZDUXGGHULQSXWWREHIRXQGWKDWSURGXFHVDKHDGLQJ 5XGGHUDQJOH GHJ UDWH WKDW EHVW PDWFKHV WKH UHIHUHQFH KHDGLQJ UDWH 7KH DFWXDWRU VXEPRGHO LV LQFOXGHG ZLWKLQ WKH IHHGEDFN

ORRS EXW ZLWKRXW WKH VDWXUDWLRQ DQG UDWH OLPLWV ,Q WKH ILUVWVWDJHRIWKHSURFHGXUHLQYHUVHVLPXODWLRQEDVHGRQ 7LPH V WKH IHHGEDFN VWUXFWXUH LV XVHG WR ILQG DQ LQSXW WR WKLV Figure 16: Results in terms of rudder angle obtained by OLQHDUDFWXDWRUPRGHOWRDFKLHYHWKHGHVLUHGUHVSRQVHLI inverse simulation using the two-stage ap- QR OLPLWV ZHUH SUHVHQW 7KH HIIHFW RI LQFOXGLQJ WKH proach for the case of the ship model with VDWXUDWLRQ DQG UDWH OLPLWV LV WKHQ LQYHVWLJDWHG LQ WKH forward speed of 2.6 m/s and a demanded VHFRQG VWDJH E\ DSSO\LQJ WKLV LGHDOLVHG DFWXDWRU LQSXW heading change of 30 deg with a rudder satu- IRXQG IURP LQYHUVH VLPXODWLRQ WR D IRUZDUG VLPXODWLRQ ration limit of ±35 deg and rudder rate limit RI WKH VKLS LQYROYLQJ WKH IXOO QRQOLQHDU DFWXDWRU VXE of ±7 deg/s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igure 15: Block diagram of the two-stage procedure VLPXODWLRQ EXLOGV XS VWHDGLO\ RYHU WKH FRPSOHWH WLPH for inverse simulation using the feedback KLVWRU\DQGUHDFKHVDOPRVWGHJDIWHUVHFRQGVDV method. VKRZQLQ)LJXUH

5HVXOWV REWDLQHG XVLQJ WKLV DSSURDFK DUH VKRZQ LQ

)LJXUHIRUDFDVHLQYROYLQJDGHJKHDGLQJFKDQJH IRUDIRUZDUGVSHHGRIPVZLWKDUXGGHUGHIOHFWLRQ OLPLWRIGHJDVEHIRUHDQGDUDWHOLPLWRIGHJV 7KHWLPHKLVWRU\RIWKHUXGGHUUHVSRQVHLQGLFDWHVFOHDUO\

WKDWWKHUXGGHUPRYHVDWWKHSRVLWLYHUDWHOLPLWRIGHJV IRUDQLQLWLDOSHULRGRIDERXWWRVE\ZKLFKWLPHWKH

+HDGLQJUDWHHUURU GHJV UXGGHUDQJOH LV FORVH WR WKHVDWXUDWLRQ OLPLW RI GHJ 7KHUXGGHUWKHQVWDUWVWRPRYHLQWKHRSSRVLWHGLUHFWLRQ

DQGDOPRVWLPPHGLDWHO\UHDFKHVWKHQHJDWLYHUDWHOLPLW RIGHJV7KLVUDWHLVPDLQWDLQHGIRUDIXUWKHUSHULRG 7LPH V RIDERXWVDIWHUZKLFKWKHUXGGHUUHVSRQVHHQWHUVD Figure 17: Heading rate error from forward simulation OLQHDU PRGH RI RSHUDWLRQ ZLWK WKH UXGGHU DQJOH (second stage of the two-stage inverse simu- DSSURDFKLQJ]HURLQWKHILQDOVRIWKHUHVSRQVH lation procedure) using the rudder deflection time history of Figure 16.

SNE 26(4) – 12/2016 253 Murray-Smith Inverse Simulation Methods Applied to Ship Steering TN

+RZHYHULQPRVWSUDFWLFDOVLWXDWLRQVLQYROYLQJKDUG OLPLWLQJ RI DFWXDWRUV ZH DUH FRQFHUQHG SULPDULO\ ZLWK GHWHFWLQJ FRQGLWLRQV ZKHQ OLPLWLQJ RFFXUV DQG ZLWK

ILQGLQJZD\VRIDYRLGLQJWKHVHUDWKHUWKDQREWDLQLQJD FRPSOHWH WLPHKLVWRU\ RI WKH RXWSXWV IURP WKH LQYHUVH VLPXODWLRQPRGHO

,WVKRXOGEHQRWHGWKDWWKHFRQYHQWLRQDOVLQJOHVWDJH

+HDGLQJHUURU GHJ IHHGEDFNPHWKRGRILQYHUVHVLPXODWLRQVWLOOKDVSUDFWLFDO YDOXH IRU FDVHV LQ ZKLFK UDWH OLPLWLQJ GRHV QRW RFFXU

7KLVDOORZVLQYHUVHVLPXODWLRQWREHXVHGDVDJHQHUDO SXUSRVH GHVLJQ WRRO DQG FDQ DVVLVW WKH GHVLJQHU LQ LQYHVWLJDWLQJ WKH SHUIRUPDQFH RI GLIIHUHQW SODQQHG 7LPH V FRQILJXUDWLRQV DW DQ HDUO\ VWDJH LQ WKH GHVLJQ SURFHVV Figure 18: Heading error from forward simulation (sec- )RUH[DPSOHLWFDQSURYLGHDQVZHUVWRTXHVWLRQVDERXW ond stage of the two-stage inverse simulation WKH FDSDELOLW\ RI WKH YHKLFOH XQGHU LQYHVWLJDWLRQ ZLWK procedure) using the rudder deflection time NQRZQSRZHUDQGFRQWUROOLPLWVWRSHUIRUPDVSHFLILHG history of Figure 16. PDQRHXYUH,ILWLVIRXQGWKDWWKHPDQRHXYUHFDQQRWEH 4 Discussion and Conclusions FDUULHGRXWLQYHUVHVLPXODWLRQPD\KHOSWKHGHVLJQHUWR PDNH FRQILJXUDWLRQDO FKDQJHV VXFK DV D FKDQJH RI ,WFDQEHFRQFOXGHGIURPWKLVDSSOLFDWLRQWKDWLQYHUVH DFWXDWRU FKDUDFWHULVWLFV RU UXGGHU DUHD WKDW WKHQ DOORZ VLPXODWLRQ PHWKRGV EDVHG RQ IHHGEDFN SULQFLSOHV FDQ WKHGHVLJQUHTXLUHPHQWVWREHVDWLVILHG SURYLGHXVHIXOLQIRUPDWLRQDERXWWKHPDUJLQVRIFRQWURO ,Q JHQHUDO WHUPV LW FDQ EH FRQFOXGHG IURP WKLV DYDLODEOH EHIRUH OLPLWLQJ HIIHFWV LQ DFWXDWRUV OHDG WR D DSSOLFDWLRQ WKDW ORRNLQJ GLUHFWO\ DW LQSXWV UHTXLUHG WR GRZQJUDGLQJRIV\VWHPSHUIRUPDQFH7KLVLVSRWHQWLDOO\ SHUIRUPVSHFLILFPDQRHXYUHVFDQSURYLGHLQVLJKWWKDWLV YHU\ LPSRUWDQW LQ V\VWHPV LQYROYLQJ PDQXDO FRQWURO VLJQLILFDQWO\ GLIIHUHQW IURP WKDW DYDLODEOH XVLQJ ZKHUHDFWXDWRULQSXWVDWXUDWLRQDQGUDWHOLPLWVFDQJLYH FRQYHQWLRQDO IRUZDUG VLPXODWLRQ WRROV 7KH IDFW WKDW ULVH WR XQGHVLUDEOH RVFLOODWRU\ SKHQRPHQD VXFK DV WKH LQYHUVHVLPXODWLRQDOORZVWKHVHQVLWLYLW\RIWKHUHTXLUHG SLORWLQGXFHG RVFLOODWLRQV WKDW KDYH EHHQ REVHUYHG LQ LQSXWWRFKDQJHVRIPRGHOSDUDPHWHUVWREHLQYHVWLJDWHG DLUFUDIW IOLJKW WHVWLQJ .QRZOHGJH RI FRQGLWLRQV GLUHFWO\ LV DQ LPSRUWDQW EHQHILW LQ WHUPV RI WKH GHVLJQ DVVRFLDWHGZLWKWKHRQVHWRIDFWXDWRUVDWXUDWLRQDQGUDWH SURFHVV OLPLWLQJ LV DOVR LPSRUWDQW IRU WKH GHVLJQ RI DXWRPDWLF ,QWHUPVRIWKHVSHFLILFUHVXOWVREWDLQHGLQWKLV FRQWUROV\VWHPVDVKDVEHHQGLVFXVVHGSUHYLRXVO\LQWKH DSSOLFDWLRQIXUWKHUDGMXVWPHQWRIWKHJDLQIDFWRUVLQWKH FRQWH[WRIVKLSFRQWURO VHHHJ>@ DQGDLUFUDIWIOLJKW IHHGEDFN SDWKZD\V FRXOG EH FRQVLGHUHG DQG FRXOG FRQWURO VHHHJ>@ IXUWKHU LPSURYH WKH DFFXUDF\ RI WKH LQYHUVH VLPXODWLRQ ,QYHUVH VLPXODWLRQ WHFKQLTXHV DUH SDUWLFXODUO\ UHVXOWV +RZHYHU DV DOZD\V D FRPSURPLVH KDV WR EH LPSRUWDQWLQDOORIWKHDERYHDUHDVEHFDXVHWKH\DOORZ IRXQG EHWZHHQ DFFXUDF\ DQG FRPSXWDWLRQDO VSHHG DQG LQIRUPDWLRQWREHJDWKHUHGGLUHFWO\DERXWKRZWKHLQSXW FRQYHQLHQFH WKDW LV QHHGHG WR SHUIRUP D VSHFLILF PDQRHXYUH LV ,WVKRXOGEHQRWHGWKDWWKHGHVLJQRIDIHHGEDFN DIIHFWHG E\ WKH RSHUDWLQJ FRQGLWLRQ DQG SDUDPHWHUV RI V\VWHP IRU LQYHUVH VLPXODWLRQ LV VLJQLILFDQWO\ GLIIHUHQW WKHPRGHO IURPWKHGHVLJQRIDIHHGEDFNFRQWUROV\VWHPLQYROYLQJ ,Q RUGHU WR LQYHVWLJDWH WKH HIIHFWV RI DFWXDWRU D SODQW PRGHO RI HTXLYDOHQW FRPSOH[LW\ ,VVXHV RI WKH VDWXUDWLRQDQGUDWHOLPLWLQJRQWKHRYHUDOOPRGHORXWSXW UREXVWQHVVRIWKHIHHGEDFNV\VWHPLQWHUPVRILWV D WZRVWDJH LQYHUVHVLPXODWLRQ DSSURDFK KDV EHHQ UHVSRQVH WR H[WHUQDO GLVWXUEDQFHV PHDVXUHPHQW QRLVH VKRZQ WR EH XVHIXO 7KLV DYRLGV DUWHIDFWV RI WKH DQG SDUDPHWULF XQFHUWDLQWLHV GR QRW KDYH WR EH IHHGEDFNDSSURDFKZKLFKFDQOHDGWRXQGHVLUDEOHOLPLW FRQVLGHUHG7KLVPHDQVWKDWOHVVUREXVWGHVLJQPHWKRGV F\FOHRVFLOODWLRQV,QGHHGLWFRXOGHYHQEHDUJXHGWKDW WKDW PLJKW EH FRQVLGHUHG LQDSSURSULDWH IRU FRQWURO WKHIHHGEDFNPHWKRGRILQYHUVHVLPXODWLRQFHDVHVWREH V\VWHP DSSOLFDWLRQV VXFK DV KLJK JDLQ VROXWLRQV RU YDOLGZKHQKDUGOLPLWLQJRFFXUVVLQFHWKHIHHGEDFNORRS HLJHQVWUXFWXUHGHVLJQPHWKRGVFDQRIWHQSURYHXVHIXOLQ WKHQEHFRPHVWUDQVLHQWO\LQDFWLYH WKHGHYHORSPHQWRILQYHUVHVLPXODWLRQV

254 SNE 26(4) – 12/2016 T N Murray-Smith Inverse Simulation Methods Applied to Ship Steering

References >@ /X/0XUUD\6PLWK'-7KRPVRQ'*,VVXHVRIQXPHU LFDODFFXUDF\DQGVWDELOLW\LQLQYHUVHVLPXODWLRQSimula- >@ 7KRPVRQ'*%UDGOH\57KHSULQFLSOHVDQGSUDFWLFDO tion Modelling Practice and Theory DSSOLFDWLRQRIKHOLFRSWHULQYHUVHVLPXODWLRQSimulation Practice and Theory GRLKWWSG[GRLRUJMVLPSDW GRL6 >@ 7KRPVRQ'*%UDGOH\5$QDQDO\WLFDOPHWKRGIRU >@ 7KRPVRQ'%UDGOH\5,QYHUVHVLPXODWLRQDVDWRROIRU th TXDQWLI\LQJKHOLFRSWHUDJLOLW\,Q12 European Ro- IOLJKWG\QDPLFVUHVHDUFK3ULQFLSOHVDQGDSSOLFDWLRQV torcraft Forum, Garmisch-Partenkirchen, Federal Re- Progress in Aerospace Sciences public of Germany, September 19863DSHU GRLMSDHURVFL >@ 7KRPVRQ'*%UDGOH\55HFHQWGHYHORSPHQWVLQWKH >@ /X/0XUUD\6PLWK'-0F*RRNLQ(:,QYHVWLJDWLRQ FDOFXODWLRQRILQYHUVHVROXWLRQVRIWKHKHOLFRSWHUHTXD RILQYHUVHVLPXODWLRQIRUGHVLJQRIIHHGIRUZDUGFRQWURO WLRQVRIPRWLRQ,Q8.6LPXODWLRQ&RXQFLO&RQIHUHQFH OHUVMathematical and Computer Modelling of Dynam- 8QLYHUVLW\&ROOHJHRI1RUWK:DOHV6HSWHPEHU ical Systems. *KHQW%HOJLXP8.6& GRL >@ .DWR26XJLXUD,$QLQWHUSUHWDWLRQRIDLUSODQHPRWLRQ >@ 7DJDZD<7X-<6WRWHQ'3,QYHUVHG\QDPLFVFRP DQGFRQWURODVLQYHUVHSUREOHPVAIAA Journal of Guid- SHQVDWLRQYLDµVLPXODWLRQRIIHHGEDFNFRQWUROV\VWHPV¶ ance, Control and Dynamics. Proc Inst. Mech Eng, Part I: Journal of Systems and GRL Control Engineering >@ &HOL52SWLPLVDWLRQEDVHGLQYHUVHVLPXODWLRQRIDKHOL GRL-6&( FRSWHUVODORPPDQHXYHUAIAA Journal of Guidance >@ )LHOGLQJ&)OX[3.1RQOLQHDULWLHVLQIOLJKWFRQWUROV\V Control and Dynamics. WHPVAeronautical Journal GRL GRL6 >@ /HH6.LP<7LPHGRPDLQILQLWHHOHPHQWPHWKRGIRU >@ %HUQVWHLQ'60LFKHO$1$FKURQRORJLFDOELEOLRJUDSK\ LQYHUVHSUREOHPRIDLUFUDIWPDQHXYHUVAIAA Journal of RIVDWXUDWLQJDFWXDWRUVInternational Journal of Robust Guidance, Control and Dynamics. and Nonlinear Control GRL GRLUQF >@ %ODMHU:.RáRG]LHMF]N\.,PSURYHG'$(IRUPXODWLRQ >@ )RVVHQ7,Guidance and Control of Ocean Vehicles IRULQYHUVHG\QDPLFVVLPXODWLRQRIFUDQHVMultibody &KLFKHVWHU8.:LOH\SS System Dynamics >@ YDQ$PHURQJHQ-Adaptive Steering of Ships: A model- GRLV reference approach to improved manoeuvring and eco- >@ 7KPPH0/RR\H*.XU]H02WWHU0%DOV-1RQ nomical course keeping. >3K''LVVHUWDWLRQ@'HOIW8QL OLQHDULQYHUVHPRGHOVIRUFRQWURO,Q6FKPLW]*HGLWRU YHUVLW\RI7HFKQRORJ\7KH1HWKHUODQGV Proceedings of the 4th International Modelica Confer- XXLGHGIGHGIDDIIGHDH ence, Hamburg, Germany, 7-8 March 2005/LQN|SLQJ >@ .RFLMDQ-0XUUD\6PLWK'-5REXVWQRQOLQHDUFRQWURO 6ZHGHQ0RGHOLFD2UJDQLVDWLRQ IRUVKLSVWHHULQJ,Q0DVWRUDNLV1(HGLWRUProgress in >@ 0XUUD\6PLWK'-)HHGEDFNPHWKRGVIRULQYHUVHVLPXOD Simulation, Modeling, Analysis and Synthesis of Modern WLRQRIG\QDPLFPRGHOVIRUHQJLQHHULQJDSSOLFDWLRQV Electrical and Electronic Systems'DQYHUV0DVV86$ Mathematical and Computer Modelling of Dynamical :RUOG6FLHQWLILFDQG(QJLQHHULQJ6RFLHW\3UHVV Systems. GRL >@ +HVV5$*DR&:DQJ6+$JHQHUDOL]HGWHFKQLTXHIRU >@ 0XUUD\6PLWK'-Modelling and Simulation of Inte- LQYHUVHVLPXODWLRQDSSOLHGWRDLUFUDIWPDQHXYHUVAIAA grated Systems in Engineering: Issues of Methodology, Journal of Guidance Control and Dynamics Quality, Testing and Application.&DPEULGJH8. GRL :RRGKHDGSS >@ 5XWKHUIRUG67KRPVRQ'*,PSURYHGPHWKRGRORJ\IRU >@ 0XUUD\6PLWK'-$QDSSUR[LPDWHGLIIHUHQWLDWLRQPHWK LQYHUVHVLPXODWLRQAeronautical Journal RGRILQYHUVHVLPXODWLRQEDVHGRQDFRQWLQXRXVV\VWHP VLPXODWLRQDSSURDFKSimulation Notes Europe GRL6 GRLVQHWQ

SNE 26(4) – 12/2016 255 Murray-Smith Inverse Simulation Methods Applied to Ship Steering TN

>@ +DPHO3*$HURVSDFHYHKLFOHPRGHOOLQJUHTXLUHPHQWV >@ 0XUUD\6PLWK'-0F*RRNLQ(:$FDVHVWXG\LQYROY IRUKLJKEDQGZLGWKIOLJKWFRQWURO,Q&RRN095\FURIW LQJFRQWLQXRXVV\VWHPVLPXODWLRQPHWKRGVRILQYHUVH 0-HGLWRUVAerospace Vehicle Dynamics and Control VLPXODWLRQIRUDQXQPDQQHGDHULDOYHKLFOHDSSOLFDWLRQ 2[IRUG8.&ODUHQGRQ3UHVV Proc. Inst Mech Eng, Part G, Journal of Aerospace En- >@ *UD\*-YRQ*UQKDJHQ:$QLQYHVWLJDWLRQRIRSHQ gineering. ORRSDQGLQYHUVHVLPXODWLRQDVQRQOLQHDUPRGHOYDOLGD GRL WLRQWRROVIRUKHOLFRSWHUIOLJKWPHFKDQLFVMath. and >@ 7DJDZD<)XNXL.,QYHUVHG\QDPLFVFDOFXODWLRQRI Comp. Modelling of Dynamical Syst., QRQOLQHDUPRGHOXVLQJORZVHQVLWLYLW\FRPSHQVDWRU,Q GRL Proceedings of Dynamics and Design Conference.1994 >@ %XFKKRO]--YRQ*UQKDJHQ:Inversion Impossible? 7HFKQLFDO5HSRUW8QLYHUVLW\RI$SSOLHG6FLHQFHV%UH >@ 9HQWXUH*.RMLPD77DJDZD<)DVWPRWLRQFRQWURORI PHQ*HUPDQ\ URERWLFV\VWHPVXVLQJLQYHUVHG\QDPLFVFRPSHQVDWLRQ >@ 0XUUD\6PLWK'-7KHDSSOLFDWLRQRISDUDPHWHUVHQVLWLY YLDµVLPXODWLRQRIIHHGEDFNFRQWUROV\VWHPV¶ ,'&6 ,Q st LW\DQDO\VLVPHWKRGVWRLQYHUVHVLPXODWLRQPRGHOV Proc. 1 Joint Conf. on Multibody System Dynamics, Mathematical and Computer Modelling of Dynamical May 25-27 2010, Lapeenranta, Finland Systems >@ $WKHUWRQ'3Nonlinear Control Engineering1HZ GRL @ 0XUUD\6PLWK'-,QYHUVHVLPXODWLRQDQGDQDO\VLVRI XQGHUZDWHUYHKLFOHG\QDPLFVXVLQJIHHGEDFNSULQFLSOH Mathematical and Computer Modelling of Dynamical Systems GRL

256 SNE 26(4) – 12/2016 SNE T ECHNICAL N OTE

Conversion of Iterative Balance Models to Directly Calculating Explicit Models for Real-time Process Optimization and Scheduling

Tomas Björkqvist1*, Olli Suominen1, Matti Vilkko1, Mikko Korpi2

1 Department of Automation Science and Engineering, Tampere University of Technology, Tampere, Finland; * [email protected] 2 Research Center Pori, Outotec (Finland) Oy, Pori, Finland

6LPXODWLRQ1RWHV(XURSH61( 7KHVHPRGHOVRIWHQLQFOXGHLWHUDWLYHEDODQFHFDOFX '2,VQHWQ ODWLRQV WR IXOILOO HPSLULFDO DQG SK\VLFDO SURFHVV FRQ 5HFHLYHG1RYHPEHU VWUDLQWV 7KHVH PRGHOV DUH ZHOO VXLWHG IRU VWHDG\ VWDWH $FFHSWHG'HFHPEHU 6SHFLDO,VVXH5HYLHZ SURFHVVGHVLJQDQGRIWHQXVHGZKHQRIIHULQJSODQQLQJ DQG FRQVWUXFWLQJ QHZ SURFHVV OLQHV KRZHYHU WKH\ DUH Abstract. Optimal utilization of complex processes RIWHQFRPSXWDWLRQDOO\WRRFXPEHUVRPHIRUXVHLQUHDO involves real-time operational optimization and schedul- WLPHVROXWLRQVGHPDQGLQJVKRUWH[HFXWLRQWLPH ing, especially in cases where the production line consists of both continuous and batch operated unit pro- 2SWLPDO XWLOL]DWLRQ RI SURFHVVHV VKRXOG LGHDOO\ LQ cesses. This kind of real-time optimization requires pro- FOXGHUHDOWLPHRSHUDWLRQDORSWLPL]DWLRQDQGVFKHGXOLQJ cess models which can be computed significantly faster ZKHUHUHVXOWVFDQEHSUHVHQWHGWRRSHUDWRUVDQGRUSUR than real-time. Iterative balance calculation is typically FHVVFRQWUROTXLFNO\'XHWRWKHWLPHUHTXLUHPHQWVDQG far too slow for these cases. This paper presents a meth- FRPSXWDWLRQDOFRPSOH[LW\RIWKHRSWLPL]DWLRQVFKHPHV od for converting an iterative balance model to a directly WKHXQGHUO\LQJSURFHVVPRGHOVPXVWEHFDSDEOHRISUR calculating model suitable for on-line process optimiza- GXFLQJ UHVXOWVVLJQLILFDQWO\ IDVWHU WKDQUHDOWLPH 7KXV tion. The approach is demonstrated with the first unit PRGHOVUHTXLULQJLWHUDWLYHFDOFXODWLRQVDUHW\SLFDOO\WRR process in the copper smelting line, the flash smelting FXPEHUVRPH WR LQFRUSRUDWH LQWR WKH RSWLPL]DWLRQ 7KH furnace (FSF). The method consisted of formulating an KLJK GHPDQG RQ H[HFXWLRQ WLPH FDQ RIWHQ EH FRPSHQ equation group based on the constrained FSF HSC-Sim VDWHGE\ORZHULQJGHPDQGVRQPRGHOSUHFLVLRQIRUWKH model and solving the unknown parameters and static UHDOWLPH RSHUDWLRQDO RSWLPL]DWLRQ ([DPSOHV RI GH states with use of a symbolic calculation software. The PDQGLQJ UHDOWLPH RSWLPL]DWLRQ XWLOL]HG LQ SURFHVV GH solution was implemented as a function whose calcula- VLJQFDQEHIRXQGLQ>@ tion time fulfilled the requirements for scheduling use. *RRG H[DPSOHV RI WKRURXJKO\ EXLOW VWHDG\ VWDWH PRGHOVFDQEHIRXQGLQPHWDOOXUJ\0RVWPHWDOOXUJLFDO Introduction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

SNE 26(4) – 12/2016 257 Björkqvist et al. Conversion of Iterative Balance Models TN

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xamples of Industrial WLRQ IRU D FRSSHU VPHOWLQJ OLQH RSHUDWLRQ FRQVLVWV RI Process Optiomization and PRGHOOLQJRIXQLWSURFHVVHVDQGGHVLJQLQJRIRSWLPL]D Scheduling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

258 SNE 26(4) – 12/2016 T N Björkqvist et al. Conversion of Iterative Balance Models

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opper Production Line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a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a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

SNE 26(4) – 12/2016 259 Björkqvist et al. Conversion of Iterative Balance Models TN

Figure 1:Flow sheet of copper process at Boliden Harjavalta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odel Conversion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

260 SNE 26(4) – 12/2016 T N Björkqvist et al. Conversion of Iterative Balance Models

3.1 Legacy model EROLFHTXDWLRQJURXSEDVHGRQWKHIODVKVPHOWLQJIXUQDFH 7KHIODVKVPHOWLQJIXUQDFHSURFHVVKDVEHHQPRGHOOHGLQ +6&6LP PRGHO DQG WR VROYH WKLV JURXS DQDO\WLFDOO\ +6&6LP DV D VWDWLF GLYLVLRQ SURFHVV ZLWK HPSLULFDO ZLWKV\PEROLFFRPSXWDWLRQWRDFKLHYHFDXVDORXWSXWVDV NQRZOHGJHFRQWUROOLQJSDUWVRIWKHGLYLVLRQFRHIILFLHQWV GLUHFWIXQFWLRQVRILQSXWV7KHVROXWLRQLVSRVVLEOHGXH 7KHLPSOHPHQWDWLRQLVDVSUHDGVKHHWOLNHGLYLVLRQFDOFX WR HPSLULFDO NQRZOHGJH LQFOXGHG LQ WKH &RQWUROV VKHHW ODWLRQ ZLWK LWHUDWLYH FDOFXODWLRQ WR IXOILOO FRQVWUDLQWV RIWKH)6)+6&6LPPRGHO GHULYHGIURPHPSLULFDODQGSK\VLFDONQRZOHGJH 7KXV WKH WDVN ZDV WR ZULWH D IXOO\ SDUDPHWUL]HG 7KH PRGHO FRQVLVWV RI WKUHH PDLQ VSUHDGVKHHWV ,Q HTXDWLRQ JURXS EDVHG RQ WKH )6) +6&6LP PRGHO SXW'LVWULEXWLRQVDQG2XWSXWHDFKFRQWDLQLQJEHWZHHQ ZKHUHWKHHTXDWLRQVDUHEDVHGRQWKHHTXDWLRQVRIHP DQG URZV DQG FROXPQV7KH ,QSXW VKHHW LV SLULFDONQRZOHGJHLQWKH&RQWUROVVKHHW7KHPRGHOLVLQ VSDUVHO\ ILOOHG ZLWK HOHPHQW PDVV IORZV DQG GHVFULEHV WKLV DQDO\WLF DSSURDFK VLPSOLILHG 7KH LQSXW HOHPHQWV KRZLQSXWFRPSRXQGVLQGLIIHUHQWVWUHDPVDUHEURNHQXS LQFOXGHRQO\WKHPDLQHOHPHQWVFRSSHU &X LURQ )H WR HOHPHQWV DFFRUGLQJ WR FKHPLFDO PRODU FRQVLVWHQF\ QLWURJHQ 1 R[\JHQ 2 VXOSKXU 6 VLOLFRQ 6L DQG 7KH'LVWULEXWLRQVVKHHWLVVSDUVHO\ILOOHGZLWKGLVWULEX RWKHU FRQWHQW 2W 7KH GLVWULEXWLRQ RI WKH HOHPHQWV WLRQFRHIILFLHQWVGLYLGLQJHOHPHQWPDVVIORZLQWRFRP EHWZHHQWKHRXWSXWVWUHDPVZKLFKDUHVHWWOHUJDVVHWWOHU SRXQGVIRUGLIIHUHQWRXWSXWVWUHDPVSDUWO\DFFRUGLQJWR IXPH VHWWOHU GXVW VODJ DQG PDWWH LV IXOO\ LQ OLQH ZLWK FKHPLFDOUHDFWLRQV7KH2XWSXWVKHHWLVILOOHGZLWKFRU WKH)6)+6&6LPPRGHO7KHHLJKWHTXDWLRQVGHWHUPLQ UHVSRQGLQJHOHPHQWPDVVIORZVWKDWEXLOGXSWKHRXWSXW LQJ HPSLULFDO NQRZOHGJH UHJDUGLQJ WKH PDLQ HOHPHQWV FRPSRXQGV LQ GLIIHUHQW RXWSXW VWUHDPV ,Q DGGLWLRQ WR ZDVFKRVHQDVEDVHIRUWKHHTXDWLRQV7RHQDEOHDQDQD WKHWKUHHPDLQVSUHDGVKHHWVD&RQWUROVVKHHWLQFOXGHV O\WLF VROXWLRQ ZLWK WKH V\PEROLF VRIWZDUH WKH HTXDWLRQ HPSLULFDOSURFHVVREVHUYDWLRQVWKDWPXVWEHIXOILOOHG JURXSKDVWREHH[DFWO\GHWHUPLQHG LQWKHVWDWLRQDU\VWDWH 7KHHTXDWLRQJURXSIRUPXODWLRQVWDUWVZLWKGHILQLQJ ,QSULQFLSOHWKHGLVWULEXWLRQIURPLQSXWFRPSRXQGV DOOEDVLFYDULDEOHVDVV\PEROLFYDULDEOHV7KLVH[DPSOH WR RXWSXW FRPSRXQG LV EXLOW XS DURXQG KRZ WKH PDLQ LQFOXGHG HOHPHQW PDVV IORZV GLVWULEXWLRQ FRHIIL HOHPHQWV FRSSHU &X DQG LURQ )H LV GLVWULEXWHG EH FLHQWVIRUHOHPHQWGLVWULEXWLRQWRRXWSXWVWUHDPVDQG WZHHQ FRPSRXQGV LQ WKH RXWSXW VWUHDPV7KH FKHPLFDO GLVWULEXWLRQ FRHIILFLHQWV IRU HOHPHQW GLVWULEXWLRQ LQWR UHDFWLRQV UHTXLUH DVVLVWLQJ HOHPHQW DV R[\JHQ 2 DQG FRPSRXQGV LQ WKH GLIIHUHQW RXWSXW VWUHDPV 7KH PDLQ VLOLFRQ 6L ZKLFK DUH EURXJKW LQ DV IORDWLQJ HOHPHQWV IRUPXODWLRQ ZRUN LV WRGHILQH WKHUHODWLRQVKLSEHWZHHQ 6XOSKXU 6 LV SDUWO\ KDQGOHG DV D PDLQ HOHPHQW DQG WKHVHYDULDEOHVZLWKHPSKDVLVRQWKHIRUPXODWLRQRIWKH SDUWO\ DV DQ DVVLVWLQJ HOHPHQW $V D UHVXOW WKH PRGHO IORDWDQGUHVWYDULDEOHV+HUHWKLVSDUWUHTXLUHGDERXW FRQVLVWV RI VRPH IL[HG GLVWULEXWLRQ FRHIILFLHQWV PDQ\ GHILQLWLRQV $IWHU WKHVH GHILQLWLRQV WKH RXWSXW FRP FRHIILFLHQWV ZKLFK DUH LWHUDWLYHO\ DGMXVWHG WR IXOILOO WKH SRXQGV FDQ EH IRUPXODWHG$IWHUZDUGV WKH ILQDO HTXD HPSLULFDOREVHUYDWLRQVDQGQXPHURXVFRHIILFLHQWVFDOFX WLRQVEDVHGRQWKHHPSLULFDONQRZOHGJHLQWKH&RQWUROV ODWHGDVIORDWDFFRUGLQJWRFRUUHVSRQGLQJFKHPLFDOUHDF VKHHWVFDQEHZULWWHQ7RHDVHWKHGHULYDWLRQRIWKHDQD WLRQVRUDVUHVWIRUVXUSOXVHOHPHQWV7KHPRGHOLVWKXVD O\WLF VROXWLRQ RI WKH VRIWZDUH WKH QRQOLQHDULWLHV LQ WKH V\VWHPRIIRXUVSUHDGVKHHWVZLWKDODUJHQXPEHURI HPSLULFDO NQRZOHGJH ZHUH OLQHDUL]HG 7KH VDPH YDULD LQWHUFRQQHFWHGFHOOV$QLWHUDWLYHURXWLQHLVXVHGWRVROYH EOHV DV WKH PDQLSXODWHG YDULDEOHV LQ WKH LWHUDWLYH VROX WKHGLVWULEXWLRQFRHIILFLHQWVDQGWKHUHE\WKHHOHPHQWDQG WLRQRI+6&6LPPRGHOZHUHFKRVHQDVYDULDEOHVIRUWKH FRPSRXQGVWUHDPVLQWKHVWDWLRQDU\VWDWH FDOFXODWLRQWRVROYH7KH\ZHUHGLVWULEXWLRQFRHIILFLHQW 7KH FDOFXODWLRQ LV YHU\ XVHIXO IRU RIIOLQH VWUDWHJLF IRU)HWRPDWWHGLVWULEXWLRQFRHIILFLHQWIRU)HLQVODJWR SODQQLQJ RI PHWDO SURFHVVLQJ7KH FDOFXODWLRQ LVKRZ )H6GLVWULEXWLRQFRHIILFLHQWIRU&XWRVODJGLVWULEXWLRQ HYHUWRRVORZIRUUHDOWLPHSURFHVVRSWLPL]DWLRQ FRHIILFLHQW IRU )H LQ PDWWH WR )H2 2W WR PDWWH 6L LQSXWVWUHDP2LQSXWVWUHDPDQGGLVWULEXWLRQFRHIILFLHQW 3.2 Method for derivation of fast IRU)HLQVODJWR)H2 calculating model 7KLV VWXG\ XWLOL]HV WKH 6\PEROLF 0DWK 7RROER[ LQ ,Q JHQHUDO WKH REMHFWLYH IRU WKH VWXG\ ZDV WR ILQG D WKH0DWODEVRIWZDUH:LWKWKHUHODWLRQVKLSVFRQFHUQLQJ PHWKRG IRU FRQYHUWLQJ LWHUDWLYH RXWSXW FRQWUROOHG EDO XVH RI R[\JHQ VWLOO XQGHILQHG WKH VROYHU PDQDJHG WR DQFHPRGHOVWRGLUHFWO\FDOFXODWLQJPRGHOVVXLWDEOHIRU DFKLHYH D IXOO\ V\PEROLFDO VROXWLRQ LQ DURXQG ILYH SURFHVVVFKHGXOLQJ7KHEDVLFLGHDZDVWRIRUPDV\P PLQXWHVZLWKDODSWRS

SNE 26(4) – 12/2016 261 Björkqvist et al. Conversion of Iterative Balance Models TN

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

4 Model Validation and 4.2 Calculation time Discussion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

4.1 Similarity to legacy model 5 Model Utilization )LJXUH VKRZV D FRPSDULVRQ EHWZHHQ WKH DQDO\WLFDO GLUHFWVROXWLRQUHVXOWVZLWKWKHEOXHOLQHDQGLWHUDWLYHO\ 7KH GLUHFWO\ FDOFXODWLQJ PRGHO RI WKH IODVK VPHOWLQJ FDOFXODWHG +6&6LP UHVXOWV UHG OLQH DV IXQFWLRQ RI IXUQDFHSURFHVVZLOOEHXWLOL]HGLQVFKHGXOLQJRIDFRS PDWWHFRSSHUSHUFHQWDJH SHU SURGXFWLRQ OLQH WR RSWLPL]H SURGXFWLRQ DQG FRVWV :KHQVROYLQJWKHHTXDWLRQJURXSWKHVROYDEOHYDULDEOHV FDQ EH IUHHO\ FKRVHQ 7KHUH DUH WZR HYLGHQW ZD\V RI PRGHOIRUPXODWLRQWKDWFDQEHXWLOL]HG

Figure 2: A comparison between analytical solution results with blue line and iteratively calculated HSC-Sim results with red line.

262 SNE 26(4) – 12/2016 T N Björkqvist et al. Conversion of Iterative Balance Models

$ VLPLODU URXWLQH ZLOO EH FDOOHG DW KLJK IUHTXHQF\ ZKHQWKHPRGHOLVXWLOL]HGLQWKHUHDOWLPHRSHUDWLRQDO DQGVFKHGXOLQJRSWLPL]DWLRQ7KHURXWLQHLVFDOOHGDWWKH PRPHQW ZKHQ WKH QRQOLQHDU RSWLPL]DWLRQ DOJRULWKP H[HFXWHV D QHZ LWHUDWLYH VFKHGXOH 7KH H[DPSOH LV LQ OLQH ZLWK WKH XWLOL]DWLRQ RI WKH FORVHG DQDO\WLF VROXWLRQ SUHVHQWHGLQ)LJXUH

Figure 3: Direct input output model utilized in scheduling.

5.1 Direct input output $QDWXUDOVROXWLRQZRXOGEHWRIRUPDGLUHFWLQSXWRXW SXWPRGHOWRPLPLFWKHUHDOVPHOWLQJSURFHVV)LJXUH UHSUHVHQWV D VFKHGXOLQJ VWUXFWXUH WKDWXWLOL]HV WKH LQSXW Figure 4: Closed analytic solution utilized in scheduling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losed analytic solution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odel based schedule calculation DQ RSWLPL]DWLRQ DOJRULWKP LQFOXGLQJ WZR VWHSV LQ FRQ 7R GHPRQVWUDWH WKH XVDJH RI WKH GLUHFWO\ FDOFXODWLQJ FHQWUDWHIHHGUDWHRQHVWHSLQFRSSHUFRQWHQWRIIHHGWR PRGHO DQ H[DPSOH VFKHGXOH LV GHULYHG ZKHUH VSHFLDO PDWWHDQGRQHPDWWHWDSSLQJ DWWHQWLRQLVSDLGWRWKHFDOFXODWLRQWLPH

SNE 26(4) – 12/2016 263 Björkqvist et al. Conversion of Iterative Balance Models TN

6FKHGXOHGFRQFHQWUDWHVWUHDP

WLPH>K@ 6FKHGXOHGFRSSHUSHUFHQWDJHRIIHHGWRPDWWH

WLPH>K@ 6FKHGXOHGPDWWHWDSSLQJ

WLPH>K@ Figure 5: A plausible schedule of an optimization algorithm. :KHQ FKDQJLQJ WKH FRQFHQWUDWH IHHG UDWH WKH LQWHUQDO HOHPHQWPL[KDVEHHQNHSWFRQVWDQW)LJXUHVKRZVWKH RXWFRPH RI WKH H[DPSOH URXWLQH )LUVW WKH URXWLQH KDV GLUHFWO\FDOFXODWHGWKHUHTXLUHGVLOLFDDQGR[\JHQIRUWKH GLIIHUHQW VWDWLF VPHOWLQJ SURFHVV VWHSV DFFRUGLQJ WR WKH FORVHG DQDO\WLF VROXWLRQ 6HFRQG WKH URXWLQH KDV FDOFX Figure 6: Outcome of example routine including both ODWHG WKH VWHSZLVH FKDQJLQJ HOHPHQW VWUHDPV WR PDWWH needed process feeds and matte state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onclusions WR PDWWH WR WKH DFWXDO FRSSHU PDWWH SHUFHQWDJH LQ WKH PDWWH YROXPH LH WKH FRSSHU PDWWH SHUFHQWDJH RI WKH 7KHREMHFWLYHRIWKLVVWXG\ZDVWRGHYHORSDPHWKRGIRU WDSSHGFRSSHUGHOLYHUHGWRWKHQH[WXQLWSURFHVVLQWKH FRQYHUWLQJLWHUDWLYHRXWSXWFRQWUROOHGEDODQFHPRGHOVWR VPHOWHUOLQH GLUHFWO\FDOFXODWLQJPRGHOVIRUSURFHVVRSWLPL]DWLRQDQG VFKHGXOLQJ

264 SNE 26(4) – 12/2016 T N Björkqvist et al. Conversion of Iterative Balance Models

7KLVPHWKRGZDVXVHGLQWKHFDVHRIDIODVKVPHOWLQJ References IXUQDFH SUHYLRXVO\ PRGHOOHG LQ +6&6LP 7KH IDVW >@ +DUMXQNRVNL,%DXHU5&RQILJXUDEOH6FKHGXOLQJ6ROX FDOFXODWLQJ PRGHO LV WR EH XVHG LQ RSWLPL]DWLRQ RI WKH WLRQXVLQJ)OH[LEOH+HXULVWLFV3URFHHGLQJVRIWKHWK WRWDOSURGXFWLRQOLQHRSHUDWLRQ (XURSHDQ6\PSRVLXPRQ&RPSXWHU$LGHG3URFHVV(Q 7KHPHWKRGFRQVLVWHGRIIRUPXODWLQJDQHTXDWLRQJURXS JLQHHULQJ±(6&$3(SS EDVHG RQ WKH FRQVWUDLQHG )6) +6&6LP PRGHO DQG >@ 7RXUHW]N\&+DUMXQNRVNL,%DOGHD0$)UDPHZRUNIRU VROYLQJ WKH XQNQRZQ SDUDPHWHUV DQG VWDWLF VWDWHV ZLWK ,QWHJUDWHG6FKHGXOLQJDQG&RQWUROXVLQJ'LVFUHWH7LPH '\QDPLF3URFHVV0RGHOV3URFHHGLQJVRIWKHWK(XUR XVHRI D V\PEROLF FDOFXODWLRQ VRIWZDUH7KH VWXG\ ZDV SHDQ6\PSRVLXPRQ&RPSXWHU$LGHG3URFHVV(QJLQHHU VXFFHVVIXO HYHQ LI LW UHTXLUHV FDUHIXO IRUPXODWLRQ ZRUN LQJ±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|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± 7KLVSDSHULVDQH[WHQGHGYHUVLRQRIDSDSHUSUHVHQWHG '2,MFRQHQJSUDF DWWKHWK(8526,0&RQJUHVVRQ0RGHOOLQJDQG6LPX >@ %HONKLU:5DWLHU11JX\HQ''@ =DPNRWVLDQ)&LUVWHD+7RZDUGVDQDXWRPDWLFWRROIRU PXOWLVFDOHPRGHOGHULYDWLRQLOOXVWUDWHGZLWKDPLFUR PLUURUDUUD\WK,QWHUQDWLRQDO6\PSRVLXPRQ6\PEROLF Acknowledgment DQG1XPHULF$OJRULWKPVIRU6FLHQWLILF&RPSXWLQJ 7KLV ZRUN ZDV FDUULHG RXW LQ WKH 6,03 UHVHDUFK SUR SS'2,6<1$6& JUDPFRRUGLQDWHGE\WKH)LQQLVK0HWDOVDQG(QJLQHHU >@ /LQ-&KHQ&&RPSXWHUDLGHGV\PEROLFG\QDPLF LQJ&RPSHWHQFH&OXVWHU ),0(&& /WG7KHVXSSRUWLV PRGHOLQJIRU6WHZDUWSODWIRUPPDQLSXODWRU5RERWLFD JUDWHIXOO\ DFNQRZOHGJHG 7KH DXWKRUV DUHDOVR JUDWHIXO YROSS IRU WKH SURFHVV H[SHUW NQRZOHGJH DQG JRRG ZRUNLQJ >@

SNE 26(4) – 12/2016 265 Björkqvist et al. Conversion of Iterative Balance Models TN

>@ 0RRVDYLDQ6$$3DSDGRSRXORV(([SOLFLWG\QDPLFVRI >@ 6FKOHVLQJHU0(.LQJ0-6ROH.&'DYHQSRUW:* VSDFHIUHHIO\HUVZLWKPXOWLSOHPDQLSXODWRUVYLD ([WUDFWLYHPHWDOOXUJ\RIFRSSHU(OVHYLHU 63$&(0$3/($GY5REYROSS >@ KWWSZZZRXWRWHFFRPHQ3URGXFWVVHUYLFHV+6& >@ 6l\QHYLUWD6/XRWRMlUYL0,QWHJUDWHGSDSHUSURGXFWLRQ &KHPLVWU\&DOFXODWLRQPRGXOHV6LPSURFHVV DQGHQHUJ\SODQQLQJ3XO3DSHUFRQIHUHQFHV±(QHU VLPXODWLRQ J\DQGFDUERQPDQDJHPHQW +HOVLQNL)LQODQG >@ KWWSZZZEROLGHQFRP'RFXPHQWV3UHVV3XEOLFDWLRQV >@ +DUMXQNRVNL,*URVVPDQQ,( $GHFRPSRVLWLRQ 3ODFHEURVFKXUHVEROLGHQKDUMDYDOWDHQSGI DSSURDFKIRUWKHVFKHGXOLQJRIDVWHHOSODQWSURGXFWLRQ &RPSXWHUV &KHPLFDO(QJLQHHULQJ >@ %M|UNOXQG35DQNL70LHWWLQHQ(5HFHQWH[SHULHQFHV >@ +DUMXQNRVNL,)DKO0%RUFKHUV+:6FKHGXOLQJDQG ZLWKLPSOHPHQWLQJG\QDPLFSURFHVVFRQWURODQGPRQL 2SWLPL]DWLRQRID&RSSHU3URGXFWLRQ3URFHVV/RJLVWLF WRULQJLQWKHIODVKVPHOWLQJSURFHVV&RSSHU&RQ RSWLPL]DWLRQRIFKHPLFDOSURGXFWLRQSURFHVVHV IHUHQFH&KLOH9ROXPH,,,%RRN1LFNRODV7KHPHOLV >@ 1DYDUUD$$XWRPDWHGVFKHGXOLQJDQGVFLHQWLILFPDQ V\PSRVLXPRQS\URPHWDOOXUJ\DQGSURFHVVHQJLQHHULQJ DJHPHQWRIFRSSHUVPHOWHUV0LQHUDO3URFHVVLQJDQG([ 3DJHV WUDFWLYH0HWDOOXUJ\9RO,VV >@ %M|UNTYLVW76XRPLQHQ29LONNR0.RUSL0)URP,W >@ 6XRPLQHQ20|UVN\95LWDOD59LONNR0)UDPHZRUN HUDWLYH%DODQFH0RGHOVWR'LUHFWO\&DOFXODWLQJ([SOLFLW IRURSWLPL]DWLRQDQGVFKHGXOLQJRIDFRSSHUSURGXFWLRQ 0RGHOVIRU5HDOWLPH3URFHVV2SWLPL]DWLRQDQG6FKHGXO SODQW3URFHHGLQJVRIWKHWK(XURSHDQ6\PSRVLXPRQ LQJWK(8526,0&RQJUHVVRQ0RGHOOLQJDQG &RPSXWHU$LGHG&KHPLFDO(QJLQHHULQJ(6&$3( 6LPXODWLRQ'2,(8526,0 9ROXPH3DJHV±'2, %

266 SNE 26(4) – 12/2016 SNE T ECHNICAL N OTE

Modelling of Indoor Lighting Conditions in Buildings for Control Design Purposes

Vito Logar1*

1Faculty of Electrical Engineering, University of Ljubljana, Tržaška 25, 1000 Ljubljana, Slovenia; *[email protected]

6LPXODWLRQ1RWHV(XURSH61( $VXPPDU\RIWKHUHVHDUFKFDUULHGRXWWRGDWHLQWKH '2,VQHWQ DUHD RI EXLOGLQJV HQHUJ\ HIILFLHQF\ EXLOGLQJV HQHUJ\ 5HFHLYHG1RYHPEHU SHUIRUPDQFHDQGEXLOGLQJV SURFHVVHVPRGHOOLQJFDQEH $FFHSWHG'HFHPEHU 6SHFLDO,VVXH5HYLHZ IRXQGLQWKHH[FHOOHQWUHYLHZSDSHUE\)RXFTXLHUHWDO >@ )XUWKHUPRUH VXIILFLHQW GD\OLJKW FRQGLWLRQV KDYH Abstract.Lighting conditions in buildings and efficient EHHQSURYHQWRKDYHDEHQHILFLDOHIIHFWRQKXPDQKHDOWK use of solar energy are a subject of considerate attention >@ 1XPHURXV DSSURDFKHV WR FRQWUROOLQJ LQGRRU in order to provide sufficient living comfort and to re- LOOXPLQDQFH FRQGLWLRQV KDYH EHHQ SURSRVHG PRVW RI duce the energy use. For this reason numerous methods ZKLFK DWWHPSW WR HLWKHU DFKLHYH FRQVWDQW LQGRRU and techniques, practical and theoretical, have been LOOXPLQDQFH OHYHOV VR DV WR SURYLGH VXIILFLHQW OLYLQJ developed. In this paper a theoretical approach to mod- FRPIRUWRUWRPD[LPL]HWKHXVHRIVRODUHQHUJ\ZKLOH elling of the indoor lighting conditions is proposed, VWLOOSURYLGLQJDFFHSWDEOHOLJKWLQJFRQGLWLRQV> based on fuzzy black-box modelling. The presented @ 7RJHWKHU ZLWK WKH PRGHOOLQJ RI OLJKW IOX[ LQ model is able to estimate indoor illuminance levels as its GRRU OLJKW LQWHQVLWLHV DQG VXUIDFH LOOXPLQDQFHV ZKLFK outputs, by using measured external conditions as its XVXDOO\UHSUHVHQWWKHEDVLVIRUFRQWUROGHVLJQWHFKQLTXHV inputs. The model can be used to study the influence of KDYH DOVR EHHQ WKH VXEMHFW RI PXFK DWWHQWLRQ )XUWKHU both controllable and uncontrollable variables to the PRUHPDQ\PHWKRGVH[LVWWKDWDUHDEOHWRSURYLGHDS indoor lighting conditions, such as weather, time of the SUR[LPDWH LOOXPLQDQFHOHYHO SUHGLFWLRQ LQ D FHUWDLQ year, blinds position, electric lighting and others. Fur- SRVLWLRQ LQ D URRP JLYHQ LWV JHRPHWU\ JOREDO RULHQWD thermore, using the above model studies on control WLRQWKHSRVLWLRQRIWKHVXQWKHVXUIDFHFKDUDFWHULVWLFV design can be performed in order to obtain algorithms DQGRU WKH ZHDWKHU FRQGLWLRQVPHDVXUHPHQWV > for maximal use of the solar energy and to minimize the @ 0RUHRYHU D VWXG\ SHUIRUPHG E\ /LQGHO|I >@ energy consumption. The study has shown that a fuzzy SURSRVHV D IDVW GD\OLJKW PRGHO DEOH WR REWDLQ LQGRRU illuminance model can estimate the indoor illuminance LOOXPLQDQFHV DV D OLQHDU FRPELQDWLRQ RI WKH H[WHUQDO levels comparable to the measured data. Small error JOREDODQGGLIIXVHUDGLDWLRQVYDOLGDWHGXVLQJWKH5DGL measures show that similar modelling approach can be DQFHPRGHOZKLFKFDQEHXVHGDVDUHSODFHPHQWIRUWKH used in order to integrate the proposed model into other UHDOV\VWHPRIHPEHGGHGFRQWUROOHUV6LPLODUO\DYDLOD environments and can further be used for simulations EOH VRIWZDUH WRROV LH 5DGLDQFH 'D\VLP 6N\YLVLRQ on indoor lighting comfort, control design or model- >@ DQGPDQ\RWKHUVDUHDOVRDEOHWRFDOFXODWH based control. PRUHRUOHVV DFFXUDWH LOOXPLQDQFH OHYHOV IRU D JLYHQ SRVLWLRQ LQ D URRP KRZHYHU VLJQLILFDQW NQRZOHGJHRI WKH PRGHOOHG V\VWHP FRPSOHWH JHRPHWULF DQG SKRWR Introduction PHWULFFKDUDFWHULVWLFVRIWKHURRPLQYHQWRU\ZLQGRZV ,QGRRUOLJKWLQJFRQGLWLRQVDQGWKHHIILFLHQWXVHRIVRODU EOLQGVOLJKWVHWF DQGWKHVRIWZDUHLWVHOIDUHQHHGHGLQ HQHUJ\KDYHEHFRPHYHU\LPSRUWDQWLQUHFHQWGHFDGHV RUGHU WR HQVXUH DFFXUDWH UHVXOWV $ ORW RI WKH H[LVWLQJ ERWKLQWHUPVRIRYHUDOOOLYLQJFRPIRUW>@DQGHQHU DSSURDFKHV UHO\ RQ NQRZQ PDWKHPDWLFDO GD\OLJKWLQJ J\HIILFLHQWEXLOGLQJV>@ FRQFHSWVDQGWKXVWU\WRGHVFULEHWKHSK\VLFDOUHODWLRQV EHWZHHQWKHLQSXWDQGRXWSXWYDULDEOHV

SNE 26(4) – 12/2016 267 Logar et al. Modelling of Indoor Lighting Conditions in Buildings TN

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|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ndoor Environment 7KHSURSRVHGVWUXFWXUHRIWKHIX]]\PRGHOVLVYHU\ VLPSOH LQSXWV RXWSXW *DXVVLDQ PHPEHUVKLS 7KHIROORZLQJVHFWLRQJLYHVDGHVFULSWLRQRIWKHLQGRRU IXQFWLRQVSHULQSXWDQGIX]]\UXOHV DQGWKHIX]]LILFD HQYLURQPHQW ZKRVH PHDVXUHPHQWV DUH XVHG DV D EDVLV WLRQGHIX]]LILFDWLRQSURFHGXUHVDUHVLPSOHYHFWRUPXOWL IRU WKH IX]]\ EODFNER[ PRGHO V GHYHORSPHQW DQG WKH SOLFDWLRQV SDUDPHWHULVDWLRQ

268 SNE 26(4) – 12/2016 T N Logar et al. Modelling of Indoor Lighting Conditions in Buildings

7KHLQGRRUHQYLURQPHQWFRQVLVWVRIDURRPZLWKGL N PHQVLRQVRIP[P[P O[Z[K D IORRUDUHDRIPDQGDYROXPHRIPZLWK RQHRXWVLGHZDOOWKDWKDVD ZLQGRZIDFLQJVRXWKZHVW URWDWHGDSSUR[LPDWHO\FRXQWHUFORFNZLVHIURPWKH

Indoor illuminance HDVWZHVWGLUHFWLRQ ZKHUHWKHRXWVLGHZDOOLVWKHORQJ sensor WK (on a desk, horizontal HVWZDOO7KHURRPLVORFDWHGRQWKH RIIORRUVLQD 0.9 m above floor) EXLOGLQJZLWKQRH[WHUQDOREVWDFOHVWKDWZRXOGREVFXUH WKHOLJKWIORZWKURXJKWKHZLQGRZ ODWORQ Window 7KH DUHD RI WKH ZLQGRZ LV P ZLWK

Blinds External Global and diffuse LQVWDOOHG YHQHWLDQ EOLQGV 7KH WUDQVPLVVLRQ RI YLVLEOH position illuminance solar radiation sensor Blinds sensor sensors OLJKWWKURXJKWKHZLQGRZLV7KHURRPFKDUDFWHU (facade, vertical) (roof, horizontal) LVWLFVLQWHUPVRIWKHSKRWRPHWU\DUHWKHIROORZLQJJUH\ IORRU UHIOHFWDQFH ZKLWH FHLOLQJ UHIOHF Figure 1: Floor plan of the modelled room, with the WDQFH ZKLWH ZDOOV DQG EHLJH IXUQLWXUH DYHUDJH marked positions of the sensors (indoor and UHIOHFWDQFH )LJXUHVKRZVWKHIORRUSODQRIWKHSDUWLF external illuminance, global and diffuse solar radiation (placed on the roof of the building), XODU URRP ZLWK WKH PDUNHG SRVLWLRQV RI WKH VHQVRUV blinds' position), the window area and the LQGRRU DQG H[WHUQDO LOOXPLQDQFH JOREDO DQG GLIIXVH blinds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uzzy Model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ãLUHWDO>@

SNE 26(4) – 12/2016 269 Logar et al. Modelling of Indoor Lighting Conditions in Buildings TN

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tructure of the model DQG GLIIXVH UDGLDWLRQ DV WZR RI WKH LQSXWV EHVLGHV WKH 7KHSURSRVHGEODFNER[LOOXPLQDQFHPRGHOLVEDVHGRQ H[WHUQDOLOOXPLQDQFHWKHEOLQGV SRVLWLRQDQGWKHOLJKWV D 7DNDJL6XJHQR 76 IX]]\LQIHUHQFH V\VWHP ),6 VWDWXV LVRSWLRQDODQGOHDGVWREHWWHUHVWLPDWLRQUHVXOWV >@ WKDW ZDV WUDLQHG XVLQJ UHDO HQYLURQPHQW PHDVXUH LQFRPSDULVRQZLWKWKHUHVXOWVWKDWKDYHWKHJOREDODQG PHQWVDVWKHLQSXWVDQGRXWSXWV7KHPRGHOLQ76IRUP GLIIXVHUDGLDWLRQPHDVXUHPHQWVRPLWWHG DSSUR[LPDWHV D QRQOLQHDU V\VWHP E\ VPRRWKO\ LQWHUSR 7KH UHDVRQ IRU XVLQJ WKH EOLQGV SRVLWLRQ DV RQH RI ODWLQJ DIILQH ORFDO PRGHOV >@ (DFK ORFDO PRGHO FRQ WKHLQSXWVGHVSLWHWKHIDFWWKDWLWKDVDOUHDG\EHHQXVHG WULEXWHV WR WKH JOREDO PRGHO LQ D IX]]\ VXEVHW RI WKH LQ WKH PXOWLSOLFDWLRQ ZLWK H[WHUQDO LOOXPLQDQFH LV WKDW VSDFH FKDUDFWHUL]HG E\ D PHPEHUVKLS IXQFWLRQ 7KH HYHQEHWWHUHVWLPDWLRQUHVXOWVFDQEHDFKLHYHGLQFRP DIILQH76PRGHOFDQEHXVHGWRDSSUR[LPDWHDQ\DUEL SDULVRQ ZLWK WKH UHVXOWV WKDW KDYH WKLV LQSXW RPLWWHG WUDU\IXQFWLRQZLWKDQ\GHVLUHGGHJUHHRIDFFXUDF\> ZKLFKFRXOGEHDFRQVHTXHQFHRIWKHQRQOLQHDULW\RIWKH @ 7KH JHQHUDOLW\ FDQ EH SURYHQ ZLWK WKH 6WRQH SURFHVVFDXVHGE\WKHEOLQGV UHIOHFWDQFHZKLFKLVWKXV :HLHUVWUDVVWKHRUHP>@ZKLFKVXJJHVWVWKDWDQ\FRQ UHSUHVHQWHGPRUHDFFXUDWHO\)LJXUHVKRZVDVFKHPDW WLQXRXVIXQFWLRQFDQEHDSSUR[LPDWHGE\DIX]]\EDVLV LFUHSUHVHQWDWLRQRIWKHPRGHO IXQFWLRQH[SDQVLRQ>@ Inputs

6LQFH WKH OLJKWIORZ G\QDPLFV FDQ EH IURP WKH External illuminance x Black box URRPLOOXPLQDQFHSRLQWRIYLHZFRQVLGHUHGDVLQILQLWHO\ Blinds position illuminance model Output Global solar radiation IDVW RU LQVWDQWDQHRXV WKH EODFNER[ PRGHO LV UHJDUGHG Fuzzy system Indoor illuminance Takagi-Sugeno estimation DVDVWDWLFPRGHOIURPWKHPRGHOOLQJSHUVSHFWLYH7KLV Diffuse solar radiation PHDQVWKDWDFKDQJHLQHLWKHUWKHJOREDORUWKHGLIIXVH Lights status (on/off) Blinds position (1 - 5) OLJKW WKH EOLQGV SRVLWLRQ RU WKH OLJKWV VWDWXV KDV DQ LPPHGLDWHHIIHFWRQWKHFKDQJHLQWKHREVHUYHGLOOXPL Figure 2: Schematic representation of the fuzzy black-box QDQFHZLWKQRWUDQVLWLRQDOG\QDPLFVPHDQLQJWKDWWKH illuminance model. YDOXH RI WKH FXUUHQW LOOXPLQDQFH LV FRPSOHWHO\ LQGH SHQGHQWRIWKHSUHYLRXVLOOXPLQDQFHVDPSOH

270 SNE 26(4) – 12/2016 T N Logar et al. Modelling of Indoor Lighting Conditions in Buildings

4 2.2 Illuminance reconstruction x 10 9 7KLV VHFWLRQ H[SODLQV DQ RSWLRQDO PHWKRG IRU VLPSOH Measured 8 Reconstructed H[WHUQDOLOOXPLQDQFHUHFRQVWUXFWLRQZKLFKLVQHFHVVDU\ ZKHQ WKH LOOXPLQDQFH VHQVRU KDV D OLPLW LQ WKLV FDVH 7 OX[ ,IVHQVRUVZLWKKLJKHUOLPLWV OX[ 6

RU KLJKHU DUH XVHG WR REWDLQ WKH PHDVXUHPHQWV WKH 5 H[WHUQDO LOOXPLQDQFH UHFRQVWUXFWLRQ LV QRW QHHGHG DQG 4 WKH PHDVXUHG H[WHUQDO LOOXPLQDQFH FDQ EH XVHG DV DQ 3

LQSXWWRWKHIX]]\PRGHOGLUHFWO\ External illuminance [lux] .QRZLQJWKDWGXULQJVXQQ\VXPPHUGD\VWKHH[WHU 2

QDOLOOXPLQDQFHFDQJRDERYHOX[WKHPLVVLQJ 1 VDWXUDWHG GDWD QHHGV WR EH UHFRQVWUXFWHG LQ RUGHU WR 0 0 1 2 3 4 5 DFKLHYH EHWWHU HVWLPDWLRQV RI WKH FDOFXODWHG YDOXH $V Day WKH IX]]\ PRGHO DOVR XVHV WKH JOREDO DQG GLIIXVH VRODU Figure 3: Comparison between the measured and the UDGLDWLRQV DPRQJRWKHUV DVLWVLQSXWVLWLVDEOHWRVXI reconstructed external illuminance for 5 days in ILFLHQWO\ZHOOSUHGLFWWKHLQGRRULOOXPLQDQFHOHYHOVHYHQ early August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ݐ௦௔௧ ൌݐ௘௡ௗ െݐ௦௧௔௥௧ ௥௘௖௢௡௦௧௥௨௖௧ሺݐ௦௧௔௥௧ǣݐ௘௡ௗሻݍ ൌͳǤͷ 2.3 Parameterisation of the model

ଶ ଶ 7KHSDUDPHWHULVDWLRQRIWKHIX]]\PRGHODOVRNQRZQDV ݌ܽܿ݁ሺെݐ௦௔௧ǡݐ௦௔௧ሻ ൅ݐ௦௔௧ሻݏ݌ܽܿ݁ሺെݐ௦௔௧ǡݐ௦௔௧ሻ൯ ൅ݐ௦௔௧ ሻሺ݈݅݊ݏሺെ൫݈݅݊ ൈͳͲସඨ WKH WUDLQLQJ ZDV FDUULHG RXW IRU PRQWK RI GLIIHUHQW ݐ௦௔௧ʹ PHDVXUHPHQWVZLWKDVDPSOHWLPHRIV7KHGDWDZDV ݍ௜௟௟ ൌݍ௜௟௟ ൅ݍ௥௘௖௢௡௦௧௥௨௖௧௘ௗ FKRVHQLQDPDQQHUWKDWFRYHUHGDOOIRXUVHDVRQVDQGDV

ZKHUHtstartDQGtendUHSUHVHQWWKHEHJLQQLQJDQGWKHHQG PDQ\ UHDOZRUOG VLWXDWLRQV DV SRVVLEOH LH VXQQ\

RIWKHVHQVRUVDWXUDWLRQLQKRXUVtsatUHSUHVHQWVWKHRYHU FORXG\IRJJ\ZHDWKHUGLIIHUHQWVXQD]LPXWKVDQGHOH

DOOOHQJWKRIWKHVDWXUDWLRQLQKRXUVqreconstructUHSUHVHQWV YDWLRQVTXLFNLOOXPLQDQFHFKDQJHVGXHWRSDUWLDOFORXG WKHUHFRQVWUXFWHGLOOXPLQDQFHVLJQDOWREHDGGHGWRWKH LQHVV RU LQFRPLQJ WKXQGHUVWRUPV RSHUDWLRQ RI OLJKWV

PHDVXUHG LOOXPLQDQFH VLJQDO LQ OX[ qill UHSUHVHQWV WKH VKDGLQJHWF6KRUWHURUORQJHUSHULRGVRIPHDVXUHPHQW H[WHUQDO LOOXPLQDQFH VLJQDO LQ OX[ DQG WKH FRPPDQG GDWDFDQEHXVHGWRWUDLQWKHPRGHOKRZHYHUXVLQJOHVV linspace(a,b) UHSUHVHQWV D YHFWRU RI OLQHDUO\ VSDFHG WUDLQLQJGDWDUHVXOWVLQDOHVVYHUVDWLOHDQGUREXVWPRGHO YDOXHVEHWZHHQaDQGb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

SNE 26(4) – 12/2016 271 Logar et al. Modelling of Indoor Lighting Conditions in Buildings TN

$QDGDSWLYHQHXURIX]]\K\EULGOHDUQLQJDOJRULWKP 3 Results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x 10 9 GDWDFOXVWHUVHDFKLQSXWKDVWKUHHPHPEHUVKLSIXQFWLRQV Measured 8 Reconstructed ZLWKGLIIHUHQWFHQWUHVDQGNXUWRVLV

cl cl cl cl cl cl 1 ill-3 ill-1 ill-2 1 glob-3 glob-1 glob-2 6 0.8 0.8

0.6 0.6 5 0.4 0.4

0.2 0.2

4 Degree of membership Degree of membership 0 0 0.5 1 1.5 2 0 200 400 600 800 3 q *s [lux] 5 q [W/m2] ill blinds x 10 glob

External illuminance [lux]

cl cl cl cl cl 2 1 diff-3 diff-1 diff-2 1 lights-1 lights-3 0.8 0.8

1 0.6 0.6 cllights-2 0.4 0.4 0 0.2 0.2

0 1 2 3 4 5 Degree of membership Degree of membership Day 0 0 1000 0 50 100 150 200 250 0 0.2 0.4 0.6 0.8 1 q [W/m2] s [0 - OFF, 1 - ON] Global solar rad. diff lights Diffuse solar rad. 800 clblinds-3

] 1

0.8 clblinds-1 cl blinds-2 600 0.6

0.4

400 0.2

Degree of membership 0

Radiation [W/m 200 1.5 2 2.5 3 3.5 4 4.5 s [1 - 5] blinds 0 0 1 2 Day 3 4 5 Figure 5: Shape and distribution of the fuzzy membership Lights status Blinds position functions on all five inputs for all three data 5 clusters as obtained by the ANFIS training 4 algorithm. 3 )LJXUH VKRZV WKH VKDSH DQG WKH GLVWULEXWLRQ RI WKH

Lights status [0 - 1] Blinds position [1 - 5] 2 PHPEHUVKLSIXQFWLRQVZKLFKFRQWULEXWHWRWKHYDOXHRI 1 WKH RXWSXW LH WKH LQGRRU LOOXPLQDQFH OHYHO ZKLFK LV 0 REWDLQHGXVLQJWKHIROORZLQJHTXDWLRQV 0 1 2 Day 3 4 5

ௗ௜௙݈ܿௗ௜௙ିଵǡݍ௚௟௢௕݈ܿ௚௟௢௕ିଵǡݍ௕௟௜௡ௗ௦݈ܿ௜௟௟ିଵǡݏ௜௟௟ݍFigure 4: A part (5 days in early August) of the input data ݋ݑݐଵ ൌሾ for the model training, including the ݏ௟௜௚௛௧௦݈ܿ௟௜௚௛௧௦ିଵǡݏ௕௟௜௡ௗ௦݈ܿ௕௟௜௡ௗ௦ିଵሿൈ reconstructed external illuminance (upper ሾͲǤͲͲͲʹʹ͵ͺǡ െͲǤʹʹ͵Ͷǡ ʹǤͲͻͳǡ Ͷͷ͵Ǥ͵ǡ ͳͺʹǤͶሿ் െ panel), the global and diffuse solar radiation ͵͹͵Ǥͳ (middle panel), the lights' status and the blinds' ǡ ݈ܿ ݍǡ ݈ܿ ݍǡ ݈ܿ ݏ ݍ݋ݑݐ ൌሾ position (lower panel). ଶ ௜௟௟ ௕௟௜௡ௗ௦ ௜௟௟ିଶ ௚௟௢௕ ௚௟௢௕ିଶ ௗ௜௙ ௗ௜௙ିଶ ݏ ݈ܿ ǡݏ ݈ܿ ሿൈ $V LV FOHDU IURP ILJXUH IRU WKH VHOHFWHG GD\V WKH ௟௜௚௛௧௦ ௟௜௚௛௧௦ିଶ ௕௟௜௡ௗ௦ ௕௟௜௡ௗ௦ିଶ ் JOREDODQGGLIIXVHUDGLDWLRQWDNHYDOXHVEHWZHHQDQG ሾͲǤͲͲʹ͸͵͹ǡͲǤ͵ͷͳʹǡ ͳǤ͵͸ͳǡ െͻͻǤͲͳǡ െͳ͸Ǥͳ͵ሿ െ :PWKHH[WHUQDOLOOXPLQDQFHEHWZHHQDQGPRUH ͳ͸ͺǤͳ ௗ௜௙݈ܿௗ௜௙ିଷǡݍ௚௟௢௕݈ܿ௚௟௢௕ିଷǡݍ௕௟௜௡ௗ௦݈ܿ௜௟௟ିଷǡݏ௜௟௟ݍUHFRQVWUXFWHG DERYH OX[ DV ݋ݑݐଷ ൌሾ WKDQ OX[ VKRZQ LQ ILJXUH WKH EOLQGV SRVLWLRQ FDQ EH LQ WKH ݏ௟௜௚௛௧௦݈ܿ௟௜௚௛௧௦ିଷǡݏ௕௟௜௡ௗ௦݈ܿ௕௟௜௡ௗିଷሿൈ UDQJHIURPWRDQGWKHOLJKWV VWDWXVFDQEHHLWKHURU ሾͲǤͲͳ͵Ͳ͹ǡ െͲǤͺͲ͹͹ǡ ͳǤͳͶ͸ǡ ͳʹͺǤͷǡ െͳǤͶͶʹሿ் ൅ DVDOUHDG\GHVFULEHG ͳͻǤͳ͹

272 SNE 26(4) – 12/2016 T N Logar et al. Modelling of Indoor Lighting Conditions in Buildings

1800 ZKHUH qill UHSUHVHQWV WKH H[WHUQDO LOOXPLQDQFHsblinds Measured Simulated UHSUHVHQWV WKH EOLQGV SRVLWLRQ qglobDQGqdifUHSUHVHQWV 1600

WKH JOREDO DQG GLIIXVH VRODU UDGLDWLRQ slightsUHSUHVHQWV 1400 WKH OLJKWV VWDWXV DQG clx-n x FRUUHVSRQGHQW LQSXWn FOXVWHU QXPEHU UHSUHVHQWV WKH GHJUHH RI PHPEHUVKLS 1200 IURP WR RI WKH JLYHQ LQSXW WR WKH FRUUHVSRQGLQJ 1000

FOXVWHU )LQDOO\ WKH LQGRRU LOOXPLQDQFH qin LV REWDLQHG 800 XVLQJWKHIROORZLQJHTXDWLRQ 600

Indoor illuminance [lux] ݍ௜௡ ൌݍ௢௨௧ଵ ൅ݍ௢௨௧ଶ ൅ݍ௢௨௧ଷ 400

7KHPHPEHUVKLSIXQFWLRQVVKRZQLQILJXUHDQGWKH 200 IX]]\UXOHVGHVFULEHGE\HTXDWLRQVWRGHILQHWKHVR 0 FDOOHGPDSSLQJRIWKHLQSXWRXWSXWGDWDVSDFH*HQHUDO 0 2 4 6 8 10 12 4 x 10 Day O\ HDFK LQSXW FRQWULEXWHV WR WKH FRPELQHG RXWSXW DF 8 Measured FRUGLQJWRWKHFRUUHVSRQGLQJPHPEHUVKLSIXQFWLRQVLWV Reconstructed FHQWUHSRVLWLRQDQGNXUWRVLVDQGWKHHTXDWLRQVWR 6 )LJXUHVKRZVWKHYDOLGDWLRQRIWKHPRGHOE\FRP 4 SDULQJ WKH PHDVXUHG DQG PRGHOHVWLPDWHG LQGRRU LOOX

PLQDQFHVIRUGLIIHUHQWGD\V GD\V-XQHGD\V 2

6HSWHPEHUGD\V'HFHPEHUGD\V0DUFK External illuminance [lux] FRYHULQJ GLIIHUHQW VRODU SRVLWLRQV DQG FOLPDWLF FRQGL 0 0 2 4 6 8 10 12 WLRQV 7KH PRGHO ZDV YDOLGDWHG XVLQJ WKH GDWD IURP Day Lights status GLIIHUHQWGD\VWKDQZHUHXVHGIRUWKHPRGHOWUDLQLQJ Blinds position $VFDQEHVHHQLQWKHILUVWJUDSKRIILJXUHWKHLQ 5 GRRULOOXPLQDQFHOHYHOVREWDLQHGZLWKWKHIX]]\PRGHO 4 LQJHQHUDOFRUUHVSRQGWRWKHPHDVXUHGLOOXPLQDQFHOHY 3 HOV 7KH LQGRRU LOOXPLQDQFH PHDVXUHPHQWV GXULQJ WKH 2

Lights status [0 - 1] SHULRGVRIGD\OLJKWDUHDSSUR[LPDWHO\OLPLWHGWRDUDQJH Blinds position [1 - 5] 1 IURP WR OX[ DQG DUH D FRQVHTXHQFH RI WKH 0 0 2 4 6 8 10 12 H[WHUQDO LOOXPLQDQFH FRQGLWLRQV WKH SRVLWLRQ RI WKH Day 200 EOLQGVDQGWKHVWDWXVRIWKHOLJKWV/RZHULQGRRU Estimation error LOOXPLQDQFHYDOXHVRFFXUGXULQJWKHPRUQLQJDQGHDUO\ 150 DIWHUQRRQZKLOHWKHVHYDOXHVDUHKLJKHUODWHULQWKHGD\ 100 ZKLFKLVDGLUHFWFRQVHTXHQFHRIWKHURRP VRULHQWDWLRQ 50 0

GLUHFWVXQOLJKWWRWKHZLQGRZDUHDDSSHDUVLQWKHODWH Error [lux] DIWHUQRRQDURXQGSP 7KHLOOXPLQDQFHVDWQLJKWERWK -50 PHDVXUHG DQG HVWLPDWHG DUH DURXQG OX[ DQG QRW -100 -150 FORVH WR OX[ DV H[SHFWHG ZKLFK FRXOG EH D FRQVH 0 2 4 6 8 10 12 TXHQFHRI WKHVN\JORZ OLJKW SROOXWLRQ DQG WKHQHLJK Day ERXULQJVWUHHWOLJKWLQJ Figure 6: Illuminance model validation; first graph shows 7KHVHFRQGJUDSKRIILJXUHVKRZVWKDWH[WHUQDOLO the comparison of the measured (black line) and model simulated (grey line) indoor illuminances OXPLQDWLRQ PHDVXUHPHQWV UHDFK RYHU OX[ QHDU for 12 different days (days 1-3: June, days 4-6: OX[HVWLPDWHG LQWKHILUVWYDOLGDWLRQGD\V GDWD September, days 7-9: December, days 10-12: -XQHDQG6HSWHPEHU ZKLFKLVUHDVRQDEOHIRUWKLVSHUL March); second graph shows the measured RG RI WKH \HDU +RZHYHU VKRUWHU LQWHUYDOV RI VHQVRU external illuminances with reconstruction; third VDWXUDWLRQDSSHDULQGD\VWR 6HSWHPEHU GXHWRWKH graph shows the position of the blinds and the VKRUWHU LQWHUYDOV RI GD\OLJKW ZKLFK LV DOVR UHIOHFWHG LQ lights' status, respectively; fourth graph shows ORZHUHVWLPDWHGSHDNLOOXPLQDQFHV XSWROX[ the illuminance estimation error (dashed vertical lines show the mean absolute error interval).

SNE 26(4) – 12/2016 273 Logar et al. Modelling of Indoor Lighting Conditions in Buildings TN

)XUWKHUPRUH LQGLYLGXDO RFFXUUHQFHV RI LOOXPLQDQFH 4 Conclusion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

274 SNE 26(4) – 12/2016 T N Logar et al. Modelling of Indoor Lighting Conditions in Buildings

:KLOHWKHSURSRVHGDSSURDFKKDVFHUWDLQDGYDQWDJHV 7KHUHIRUHWKHXVHRIDVLPSOHLOOXPLQDQFHPRGHORI RYHU RWKHU PHWKRGV LW DOVR KDV VRPH JHQHUDO IODZV VXIILFLHQW DFFXUDF\ DOORZVTXLFN DQG UHSHDWDEOH WHVWLQJ ZKLFK PDLQO\ RULJLQDWH IURP WKH QHHG IRU PHDVXUH RI WKH SDUWLFXODU PHWKRGV 6LQFH WKH PRGHO LV UHOLDEOH PHQWV$IDLUQXPEHURIPHDVXUHPHQWVLQDVPDQ\UHDO DQGDOORZVIRUDTXLFNHVWLPDWLRQRIWKHLOOXPLQDQFHLW VLWXDWLRQV DV SRVVLEOH DUH QHHGHG LQ RUGHU WR REWDLQ D FDQDOVREHXVHGLQDVLPLODUZD\WRWKHPRGHOSUHVHQWHG UREXVWDQGYHUVDWLOHPRGHOWKDWFRYHUVDEURDGUDQJHRI E\ /LQGHO|I >@ LH DV D PRGHOEDVHG FRQWURO DS FOLPDWLF FRQGLWLRQV H[WHUQDO LOOXPLQDQFH VRODU UDGLD SURDFKXVLQJHPEHGGHGFRQWUROOHUV WLRQ DQG RWKHU LQIOXHQWLDO YDULDEOHV EOLQGV OLJKWV 0RUHRYHUWKHSURSRVHGDSSURDFKLVDEOHWRHVWLPDWHWKH References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`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ãLU0.UDLQHU$'RYMDN0.ULVWOä$XWRPDWLFDOO\ 7KH EODFNER[ IX]]\ PRGHO IRU LQGRRU LOOXPLQDQFH FRQWUROOHG GD\OLJKWLQJ IRU YLVXDO DQG QRQYLVXDO HIIHFWV HVWLPDWLRQDVSUHVHQWHGKHUHLVQRWSULPDULO\LQWHQGHG /LJKWLQJ5HVHDUFKDQG7HFKQRORJ\ WREHXVHGDVDVWDQGDORQHDSSOLFDWLRQDOWKRXJKLWFDQ >@ 7UREHF/DK0=XSDQþLþ%3HWHUQHOM-.UDLQHU$'D\ EH EXW WR EH LQFOXGHG LQ EURDGHU WHVW HQYLURQPHQWV OLJKWLOOXPLQDQFHFRQWUROZLWKIX]]\ORJLF6RODU(QHUJ\ XVHIXO IRU D YDULHW\ RI VWXGLHV IRU LQVWDQFH EXLOGLQJ DXWRPDWLRQOLYLQJFRPIRUWHQHUJ\FRQVHUYDWLRQFRQWURO >@ .ULVWOä.RãLU07UREHF/DK0.UDLQHU$)X]]\FRQ GHVLJQHWF6LQFHVXFKVWXGLHVDUHSUDFWLFDOO\LPSRVVLEOH WURO V\VWHP IRU WKHUPDO DQG YLVXDO FRPIRUW LQ EXLOGLQJ WR SHUIRUP RQ UHDO V\VWHPV GXH WR YDU\LQJ ZHDWKHU 5HQHZDEOH(QHUJ\ FRQGLWLRQVDQGDOPRVWQRUHSHDWDELOLW\DFRPELQDWLRQRI >@ 2K0+/HH.+

SNE 26(4) – 12/2016 275 Logar et al. Modelling of Indoor Lighting Conditions in Buildings TN

>@ @ 5HLQKDUW&)$QGHUVHQ0'HYHORSPHQWDQGYDOLGDWLRQ OLJKWLQJ XVH SDWWHUQV RQ OLJKWLQJ HQHUJ\ FRQVXPSWLRQ RID5DGLDQFHPRGHOIRUDWUDQVOXFHQWSDQHO(QHUJ\DQG (QHUJ\DQG%XLOGLQJV %XLOGLQJV >@ .RãLU0.UDLQHU$.ULVWOä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|I'$IDVWGD\OLJKWPRGHOVXLWDEOHIRUHPEHGGHG >@ @ /DUVRQ*:6KDNHVSHDUH5$5HQGHULQJZLWK5DGLDQFH 0RUJDQ.DXIPDQQ3XEOLVKHUV6DQ)UDQFLVFR >@ *ROEHUJ 55 0HWKRGV RI UHDO DQDO\VLV -RKQ :LOH\ DQG >@ 5HLQKDUW & %UHWRQ 3) ([SHULPHQWDO YDOLGDWLRQ RI GV 6RQV1HZ@ /LQ &+ 6LVR QRQOLQHDU V\VWHP LGHQWLILFDWLRQ XVLQJ D DO,%36$&RQIHUHQFH *ODVJRZ6FRWODQG IX]]\QHXUDO K\EULG V\VWHP ,QWHUQDWLRQDO -RXUQDO RQ 1HXUDO6\VWHPV >@ /DRXDGL$$UVHQDXOW&9DOLGDWLRQRI6N\YLVLRQ,QVWL WXWHRI5HVHDUFKLQ&RQVWUXFWLRQ2WWDZD

276 SNE 26(4) – 12/2016 SNE Simulation News EUROSIM Data and Quick Info

EUROSIM 2019 10th EUROSIM Congress on Modelling and Simulation La Rioja, Logroño, Spain, July 2019

Contents SNE Reports Editorial Board Short Info EUROSIM ...... N2 EUROSIM Emilio Jiménez, [email protected] Andreas Körner, [email protected] Short Info ASIM, CAE-SMSG ...... N3 Miguel Mujica Mota, [email protected] Short Info CROSSIM, CSSS, DBSS, FRANCOSIM ...... N4 ASIM A. Körner, [email protected] Short Info HSS, ISCS, LIOPHANT, LSS ...... N5 CEA-SMSG Emilio Jiminez, [email protected] Short Info KA-SIM, PSCS, SIMS ...... N6 CROSSIM Vesna Dušak, [email protected] CSSS Mikuláš Alexík, [email protected] Short Info SLOSIM, RNSS, UKSIM ...... N7 DBSS M. Mujica Mota, [email protected] Short Info ROMSIM, MIMOS, Albanian Soc...... N8 FRANCOSIM Karim Djouani, [email protected] HSS András Jávor, [email protected] Simulation Notes Europe SNE is the official membership ISCS M. Savastano, [email protected] journal of EUROSIM and distributed / available to members of LIOPHANT F. Longo, [email protected] the EUROSIM Societies as part of the membership benefits. LSS Yuri Merkuryev, [email protected] If you have any information, announcement, etc. you want to PSCS Zenon Sosnowski, [email protected] see published, please contact a member of the editorial board RNSS Y. Senichenkov, [email protected] in your country or the editorial office. For scientific publica- SIMS Esko Juuso, [email protected] tions, please contact the EiC. SLOSIM Vito Logar, [email protected] UKSIM A. Orsoni, [email protected] This EUROSIM Data & Quick Info compiles data from EU- KA-SIM Edmond Hajrizi, [email protected] Paolo Proietti, [email protected] ROSIM societies and groups: addresses, weblinks, and officers MIMOS Marius Radulescu, [email protected] of societies with function and email, to be published regularly ROMSIM Albanian Society Kozeta Sevrani, [email protected] in SNE issues. This information is also published at EU- ROSIM’s website www.eurosim.info. SNE Editorial Office /ARGESIM ĺ www.sne-journal.org, www.eurosim.info | [email protected], Andreas Körner, (info, news) | [email protected], Felix Breitenecker (publications) SNE Editorial Office, Andreas Körner c/o ARGESIM / Mathe- matical Modelling & Simulation Group, TU Wien /101, Wiedner Haupstrasse 8-10, 1040 Vienna , Austria

SNE 26(4) – 12/2016 N 1

Information EUROSIM and EUROSIM Societies

EUROSIM EUROSIM Board / Officers. EUROSIM is governed by a Federation of European board consisting of one representative of each member society, president and past president, and representatives Simulation Societies for SNE Simulation Notes Europe. The President is nom- General Information. EUROSIM, the Federation of Eu- inated by the society organising the next EUROSIM ropean Simulation Societies, was set up in 1989. The Congress. Secretary, Secretary to the Board, and Treas- purpose of EUROSIM is to provide a European forum for urer are elected out of members of the board. simulation societies and groups to promote advance- ment of modelling and simulation in industry, research, Emilio Jiménez (CAE-SMSG), President and development. ĺ www.eurosim.info [email protected] Member Societies. EUROSIM members may be na- Past President Esko Juuso (SIMS) tional simulation societies and regional or international [email protected] societies and groups dealing with modelling and simula- Secretary M. Mujica Mota (DBSS), tion. At present EUROSIM has 16 Full Members and 2 [email protected] Felix Breitenecker (ASIM) Observer Members: Treasurer [email protected] ASIM Arbeitsgemeinschaft Simulation Secretary to the Andreas Körner Austria, Germany, Switzerland Board [email protected] Felix Breitenecker CEA-SMSG Spanish Modelling and Simulation Group SNE Repres. Spain [email protected] CROSSIM Croatian Society for Simulation Modeling Croatia SNE – Simulation Notes Europe. SNE is a scientific CSSS Czech and Slovak Simulation Society journal with reviewed contributions as well as a mem- Czech Republic, Slovak Republic bership newsletter for EUROSIM with information from DBSS Dutch Benelux Simulation Society the societies in the News Section. EUROSIM societies Belgium, Netherlands are offered to distribute to their members the journal FRANCOSIM Société Francophone de Simulation Belgium, France SNE as official membership journal. SNE Publishers are HSS Hungarian Simulation Society; Hungary EUROSIM, ARGESIM and ASIM. ISCS Italian Society for Computer Simulation Italy Editor-in-Chief Felix Breitenecker KA-SIM Kosovo Simulation Society, Kosovo [email protected]

LIOPHANT LIOPHANT Simulation Club Italy & International ĺ www.sne-journal.org, LSS Latvian Simulation Society; Latvia | [email protected] PSCS Polish Society for Computer Simulation EUROSIM Congress. EUROSIM is running the triennial Poland conference series EUROSIM Congress. The congress is MIMOS Italian Modelling and Simulation organised by one of the EUROSIM societies. Association, Italy, Observer Member RNSS Russian National Simulation Society EUROSIM 2019, the 10th EUROSIM Congress, will be Russian Federation organised by CAE-SMSG, the Spanish simulation so- ROMSIM Romanian Society for Modelling and Sim- ciety, in La Rioja, Logroño, Spain, in July 2019. ulation, Romania, Observer Member SIMS Simulation Society of Scandinavia Denmark, Finland, Norway, Sweden Chairs / Team EUROSIM 2019 SLOSIM Slovenian Simulation Society Emilio Jiménez, EUROSIM President, Slovenia [email protected] UKSIM United Kingdom Simulation Society Juan Ignacio Latorre, [email protected] UK, Ireland ĺ www.eurosim.info

N 2 SNE 26(4) – 12/2016 Information EUROSIM and EUROSIM Societies

EUROSIM Member Societies ASIM Working Committee Methods in Modelling and Simulation GMMS ASIM Th. Pawletta, [email protected] Simulation in Environmental Systems SUG German Simulation Society J.Wittmann, [email protected] Arbeitsgemeinschaft Simulation Simulation of Technical Systems STS ASIM (Arbeitsgemeinschaft Simulation) is the associa- Walter Comerell, [email protected] Simulation in Production and Logistics tion for simulation in the German speaking area, servic- SPL ing mainly Germany, Switzerland and Austria. ASIM Sigrid Wenzel, [email protected] Simulation in Education/Education in Simulation was founded in 1981 and has now about 600 individual EDU members, and 90 institutional or industrial members. A. Körner, [email protected] Working Group Data-driven Simulation in Life ĺ www.asim-gi.org with members’ area BIG DATA Sciences; [email protected] | [email protected], [email protected] ASIM – Inst. f. Analysis and Scientific Computing Working Groups for Simulation in Business Vienna University of Technology Administration, in Traffic Systems, for Standardisa- Wiedner Hauptstraße 8-10, 1040 Vienna, Austria tion, etc.

ASIM Officers CEA-SMSG – Spanish Modelling and President Felix Breitenecker Simulation Group [email protected] Vice presidents Sigrid Wenzel, [email protected] CEA is the Spanish Society on Automation and Control T. Pawletta, [email protected] and it is the national member of IFAC (International Secretary Ch. Deatcu, [email protected] Federation of Automatic Control) in Spain. Since 1968 A. Körner, [email protected] CEA-IFAC looks after the development of the Automa- Treasurer Anna Mathe, [email protected] tion in Spain, in its different issues: automatic control, Membership S. Wenzel, [email protected] robotics, SIMULATION, etc. In order to improve the ef- Affairs W. Maurer, [email protected] ficiency and to deep into the different fields of Automa- Ch. Deatcu, [email protected] tion. The association is divided into national thematic F. Breitenecker, [email protected] groups, one of which is centered on Modeling, Simula- Repr. EUROSIM F. Breitenecker, [email protected] tion and Optimization, constituting the CEA Spanish A. Körner, [email protected] Modeling and Simulation Group (CEA-SMSG). It looks Int. Affairs – N. Popper, [email protected] after the development of the Modelling and Simulation GI Contact O. Rose, [email protected] (M&S) in Spain, working basically on all the issues concerning the use of M&S techniques as essential en- Editorial Board T. Pawletta, [email protected] gineering tools for decision-making and optimization. SNE Ch. Deatcu, [email protected] ĺ http://www.ceautomatica.es/grupos/ Web EUROSIM A. Körner, [email protected] ĺ [email protected] Last data update December 2016 [email protected]

ASIM is organising / co-organising the following inter- CEA-SMSG / Emilio Jiménez, Department of Electrical Engineering, University of La Rioja, San José de Calasanz national conferences: 31, 26004 Logroño (La Rioja), SPAIN • ASIM Int. Conference ‘Simulation in Production and Logistics’ – bi-annual CEA - SMSG Officers • ASIM ‘Symposium Simulation Technique’ President Emilio Jiménez, [email protected] – bi-annual Vice president Juan Ignacio Latorre juanigna- • MATHMOD Int. Vienna Conference on [email protected]

Mathmatical Modelling – tri-annual Repr. EUROSIM Emilio Jiminez, [email protected] Furthermore, ASIM is co-sponsor of WSC – Winter Edit. Board SNE Emilio Jiminez, [email protected] Web EUROSIM Mercedes Perez [email protected] Simulation Conference Last data update June 2016

SNE 26(4) – 12/2016 N 3

Information EUROSIM and EUROSIM Societies

CROSSIM – Croatian Society for CSSS Officers Simulation Modelling President Miroslav Šnorek, [email protected] Vice president Mikuláš Alexík, [email protected] CROSSIM-Croatian Society for Simulation Modelling Scientific Secr. A. Kaviēka, [email protected] was founded in 1992 as a non-profit society with the Repr. EUROSIM Miroslav Šnorek, [email protected] goal to promote knowledge and use of simulation me- Edit. Board SNE Mikuláš Alexík, [email protected] thods and techniques and development of education. Web EUROSIM Petr Peringer, [email protected] CROSSIM is a full member of EUROSIM since 1997. Last data update December2012 ĺ www.eurosim.info | [email protected] DBSS – Dutch Benelux Simulation Society CROSSIM / Vesna Dušak The Dutch Benelux Simulation Society (DBSS) was Faculty of Organization and founded in July 1986 in order to create an organisation Informatics Varaždin, University of Zagreb Pavlinska 2, HR-42000 Varaždin, Croatia of simulation professionals within the Dutch language area. DBSS has actively promoted creation of similar organisations in other language areas. DBSS is a mem- CROSSIM Officers ber of EUROSIM and works in close cooperation with its President Vesna Dušak, [email protected] Vice president Jadranka Božikov, [email protected] members and with affiliated societies. Secretary Vesna Bosilj-Vukšiđ, [email protected] ĺ www.DutchBSS.org Executive board Vlatko eriđ, [email protected] | [email protected] members Tarzan Legoviđ, [email protected] DBSS / A. W. Heemink Repr. EUROSIM Jadranka Božikov, [email protected] Delft University of Technology, ITS - twi, Edit. Board SNE Vesna Dušak, [email protected] Mekelweg 4, 2628 CD Delft, The Netherlands Web EUROSIM Jadranka Bozikov, [email protected] Last data update December2012 DBSS Officers President A. Heemink, [email protected] Vice president M. Mujica Mota, [email protected]

Treasurer M. Mujica Mota, [email protected] Secretary P. M. Scala, [email protected] Repr. EUROSIM M. Mujica Mota, [email protected] CSSS – Czech and Slovak Edit. SNE/Web M. Mujica Mota, [email protected] Simulation Society Last data update June 2016 CSSS -The Czech and Slovak Simulation Society has about 150 members working in Czech and Slovak nation- FRANCOSIM - Société Francophone de Simulation al scientific and technical societies (Czech Society for FRANCOSIM was founded in 1991 and aims to the pro- Applied Cybernetics and Informatics, Slovak Society for motion of simulation and research, in industry and aca- Applied Cybernetics and Informatics). The main objec- demic fields. tives of the society are: development of education and | [email protected] training in the field of modelling and simulation, organis- FRANCOSIM / Yskandar Hamam ing professional workshops and conferences, disseminat- Groupe ESIEE, Cité Descartes, ing information about modelling and simulation activities BP 99, 2 Bd. Blaise Pascal, in Europe. Since 1992, CSSS is full member of EU- 93162 Noisy le Grand CEDEX, France ROSIM. ĺ www.fit.vutbr.cz/CSSS FRANCOSIM Officers | [email protected] President Karim Djouani, [email protected] Treasurer François Rocaries, [email protected] CSSS / Miroslav Šnorek, CTU Prague FEE, Dept. Computer Science and Engineering, Repr. EUROSIM Karim Djouani, [email protected] Karlovo nam. 13, 121 35 Praha 2, Czech Republic Edit. Board SNE Karim Djouani, [email protected] Last data update December2012

N 4 SNE 26(4) – 12/2016 Information EUROSIM and EUROSIM Societies

HSS – Hungarian Simulation Society

The Hungarian Member Society of EUROSIM was estab- LIOPHANT Simulation lished in 1981 as an association promoting the exchange of information within the community of people involved Liophant Simulation is a non-profit association born in in research, development, application and education of order to be a trait-d'union among simulation developers simulation in Hungary and also contributing to the en- and users; Liophant is devoted to promote and diffuse hancement of exchanging information between the the simulation techniques and methodologies; the Asso- Hungarian simulation community and the simulation ciation promotes exchange of students, sabbatical years, communities abroad. HSS deals with the organization of organization of International Conferences, courses and lectures, exhibitions, demonstrations, and conferences. internships focused on M&S applications. ĺ www.eurosim.info ĺ www.liophant.org | [email protected] | [email protected] HSS / András Jávor, Budapest Univ. of Technology and Economics, LIOPHANT Simulation, c/o Agostino G. Bruzzone, Sztoczek u. 4, 1111 Budapest, Hungary DIME, University of Genoa, Savona Campus via Molinero 1, 17100 Savona (SV), Italy

HSS Officers LIOPHANT Officers President András Jávor, [email protected] President A.G. Bruzzone, [email protected] Vice president Gábor Szƾcs, [email protected] Director E. Bocca, [email protected] Secretary Ágnes Vigh, [email protected] Secretary A. Devoti, [email protected] Repr. EUROSIM András Jávor, [email protected] Treasurer Marina [email protected] Deputy Gábor Szƾcs, [email protected] Repr. EUROSIM A.G. Bruzzone, [email protected] Edit. Board SNE András Jávor, [email protected] Deputy F. Longo, [email protected] Web EUROSIM Gábor Szƾcs, [email protected] Edit. Board SNE F. Longo, [email protected] Last data update March 2008 Web EUROSIM F. Longo, [email protected] Last data update June 2016 ISCS – Italian Society for Computer Simulation LSS – Latvian Simulation Society The Latvian Simulation Society (LSS) has been founded The Italian Society for Computer Simulation (ISCS) is a in 1990 as the first professional simulation organisation scientific non-profit association of members from indus- in the field of Modelling and simulation in the post- try, university, education and several public and research Soviet area. Its members represent the main simulation institutions with common interest in all fields of com- centres in Latvia, including both academic and industri- puter simulation. al sectors. ĺ www.eurosim.info ĺ briedis.itl.rtu.lv/imb/ | [email protected] | [email protected] ISCS / Mario Savastano, LSS / Yuri Merkuryev, Dept. of Modelling c/o CNR - IRSIP, and Simulation Riga Technical University Via Claudio 21, 80125 Napoli, Italy Kalku street 1, Riga, LV-1658, LATVIA ISCS Officers LSS Officers President M. Savastano, [email protected] President Yuri Merkuryev, [email protected] Vice president F. Maceri, [email protected] Secretary Artis Teilans, [email protected] Repr. EUROSIM F. Maceri, [email protected] Repr. EUROSIM Yuri Merkuryev, [email protected] Secretary Paola Provenzano, Artis Teilans, [email protected] [email protected] Deputy Edit. Board SNE M. Savastano, [email protected] Edit. Board SNE Yuri Merkuryev, [email protected] Web EUROSIM Vitaly Bolshakov, [email protected] Last data update December2010 Last data update June 2016

SNE 26(4) – 12/2016 N 5

Information EUROSIM and EUROSIM Societies

KA-SIM Kosovo Simulation Society PSCS Officers President Leon Bobrowski, [email protected] Kosova Association for Modeling and Simulation (KA – Vice president Tadeusz Nowicki, SIM, founded in 2009), is part of Kosova Association of [email protected] Control, Automation and Systems Engineering (KA – Treasurer Z. Sosnowski, [email protected] CASE). KA–CASE was registered in 2006 as non Profit Secretary Zdzislaw Galkowski, Organization and since 2009 is National Member of [email protected] IFAC – International Federation of Automatic Control. Repr. EUROSIM Leon Bobrowski, [email protected] KA-SIM joined EUROSIM as Observer Member in Deputy Tadeusz Nowicki, [email protected] 2011. In 2016, KA-SIM became full member. Edit. Board SNE Zenon Sosnowski, [email protected] KA-SIM has about 50 members, and is organizing the in- Web EUROSIM Magdalena Topczewska ternational conference series International Conference in [email protected] Business, Technology and Innovation, in November, in Last data update December2013 Durrhes, Albania, and IFAC Simulation Workshops in Pristina. SIMS – Scandinavian Simulation Society SIMS is the Scandinavian Simulation Society with ĺ www.ubt-uni.net/ka-case members from the four Nordic countries Denmark, Fin- | [email protected] land, Norway and Sweden. The SIMS history goes back MOD&SIM KA-CASE; Att. Dr. Edmond Hajrizi to 1959. SIMS practical matters are taken care of by the Univ. for Business and Technology (UBT) SIMS board consisting of two representatives from each Lagjja Kalabria p.n., 10000 Prishtina, Kosovo Nordic country (Iceland one board member).

KA-SIM Officers SIMS Structure. SIMS is organised as federation of re- President Edmond Hajrizi, [email protected] gional societies. There are FinSim (Finnish Simulation Vice president Muzafer Shala, [email protected] Forum), DKSIM (Dansk Simuleringsforening) and NFA Secretary Lulzim Beqiri, [email protected] (Norsk Forening for Automatisering). Treasurer Selman Berisha, [email protected] ĺ www.scansims.org Repr. EUROSIM Edmond Hajrizi, [email protected] Deputy Muzafer Shala, [email protected] | [email protected] Edit. Board SNE Edmond Hajrizi, [email protected] SIMS / Erik Dahlquist, School of Business, Society and Engineering, Department of Energy, Building and Envi- Web EUROSIM Betim Gashi, [email protected] ronment, Mälardalen University, P.O.Box 883, 72123 Last data update December 2016 Västerås, Sweden

SIMS Officers PSCS – Polish Society for Computer President Erik Dahlquist, [email protected] Vice president Bernd Lie, [email protected] Simulation Treasurer Vadim Engelson, PSCS was founded in 1993 in Warsaw. PSCS is a scien- [email protected] Repr. EUROSIM Erik Dahlquist, [email protected] tific, non-profit association of members from universi- Edit. Board SNE Esko Juuso, [email protected] ties, research institutes and industry in Poland with Web EUROSIM Vadim Engelson, common interests in variety of methods of computer [email protected] simulations and its applications. At present PSCS counts Last data update June 2016 257 members. ĺ www.eurosim.info (www.ptsk.man.bialystok.pl) | [email protected] PSCS / Leon Bobrowski, c/o IBIB PAN, ul. Trojdena 4 (p.416), 02-109 Warszawa, Poland

N 6 SNE 26(4) – 12/2016 Information EUROSIM and EUROSIM Societies

UKSIM / Prof. David Al-Dabass SLOSIM – Slovenian Computing & Informatics, Nottingham Trent University Society for Simulation Clifton lane, Nottingham, NG11 8NS and Modelling United Kingdom

SLOSIM - Slovenian Society for Simulation and Mod- UKSIM Officers elling was established in 1994 and became the full President David Al-Dabass, member of EUROSIM in 1996. Currently it has 90 mem- [email protected] bers from both Slovenian universities, institutes, and in- Secretary A. Orsoni, [email protected] dustry. It promotes modelling and simulation approach- Treasurer A. Orsoni, [email protected] es to problem solving in industrial as well as in academ- Membership chair G. Jenkins, [email protected] ic environments by establishing communication and co- Local/Venue chair Richard Cant, [email protected] operation among corresponding teams. Repr. EUROSIM A. Orsoni, [email protected]

Deputy G. Jenkins, [email protected] ĺ www.slosim.si Edit. Board SNE A. Orsoni, [email protected] | [email protected] Last data update March 2016 SLOSIM / Vito Logar, Faculty of Electrical

Engineering, University of Ljubljana, Tržaška 25, 1000 Ljubljana, Slovenia

SLOSIM Officers President Vito Logar, [email protected] RNSS – Russian Simulation Society Vice president Božidar Šarler, [email protected] NSS - The Russian National Simulation Society Secretary Aleš Beliē, [email protected] (ɇɚɰɢɨɧɚɥɶɧɨɟ Ɉɛɳɟɫɬɜɨ ɂɦɢɬɚɰɢɨɧɧɨɝɨ Ɇɨɞɟɥɢ- Treasurer Milan Simēiē, [email protected] ɪɨɜɚɧɢɹ – ɇɈɂɆ) was officially registered in Russian Repr. EUROSIM B. Zupanēiē, [email protected] Federation on February 11, 2011. In February 2012 NSS Deputy Vito Logar, [email protected] has been accepted as an observer member of EUROSIM, Edit. Board SNE B. Zupanēiē, [email protected] Vito Logar, [email protected] and in 2015 RNSS has become full member. Blaž Rodiē, [email protected] ĺ www.simulation.su Web EUROSIM Vito Logar, [email protected] | [email protected] Last data update December 2016 RNSS / R. M. Yusupov, St. Petersburg Institute of Informatics and Automation UKSIM - United Kingdom Simulation Society RAS, 199178, St. Petersburg, 14th lin. V.O, 39

The UK Simulation Society is very active in organizing RNSS Officers conferences, meetings and workshops. UKSim holds its President R. M. Yusupov, [email protected] annual conference in the March-April period. In recent Chair Man. Board A. Plotnikov, [email protected] years the conference has always been held at Emmanuel Secretary M. Dolmatov, [email protected] College, Cambridge. The Asia Modelling and Simula- Repr. EUROSIM R.M. Yusupov, [email protected] tion Section (AMSS) of UKSim holds 4-5 conferences Y. Senichenkov, per year including the EMS (European Modelling Sym- [email protected] posium), an event mainly aimed at young researchers, Deputy B. Sokolov, [email protected] organized each year by UKSim in different European Edit. Board SNE Y. Senichenkov, cities. [email protected] Membership of the UK Simulation Society is free to Last data update June 2016 participants of any of our conferences and their co- authors.

ĺ www.uksim.org.uk | [email protected]

SNE 26(4) – 12/2016 N 7

Information EUROSIM and EUROSIM Societies

EUROSIM OBSERVER MEMBERS MIMOS Officers President Paolo Proietti, [email protected] ROMSIM – Romanian Modelling and Secretary Davide Borra, [email protected] Simulation Society Treasurer Davide Borra, [email protected] Repr. EUROSIM Paolo Proietti, [email protected] ROMSIM has been founded in 1990 as a non-profit so- Deputy agosti- ciety, devoted to theoretical and applied aspects of mod- Agostino Bruzzone, [email protected] elling and simulation of systems. ROMSIM currently has about 100 members from Romania and Moldavia. Edit. Board SNE Paolo Proietti, [email protected] ĺ www.eurosim.info (www.ici.ro/romsim) Last data update December 2016 | [email protected] ROMSIM / Florin Hartescu, National Institute for Research in Informatics, Averescu CANDIDATES Av. 8 – 10, 71316 Bucharest, Romania Albanian Simulation Society ROMSIM Officers At the Department of Statistics and Applied Informatics, President Faculty of Economy, University of Tirana, Prof. Dr. Vice president Florin Hartescu, [email protected] Marius Radulescu, [email protected] Kozeta Sevrani at present is setting up an Albanian Repr. EUROSIM Florin Stanciulescu, [email protected] Simulation Society. Kozeta Sevrani, professor of Com- Deputy Marius Radulescu, [email protected] puter Science and Management Information Systems, Edit. Board SNE and head of the Department of Mathematics, Statistics Web EUROSIM Zoe Radulescu, [email protected] and Applied Informatic, has attended a EUROSIM Last data update partly June 2016 board meeting in Vienna and has presented simulation activities in Albania and the new simulation society. The society – constitution and bylaws are being worked MIMOS – Italian Modelling and out - will be involved in different international and local Simulation Association simulation projects, and will be engaged in the organisation of the conference series ISTI – Information Sys- MIMOS (Movimento Italiano Modellazione e Simula- tems and Technology. The society intends to become a zione – Italian Modelling and Simulation Association) is EUROSIM Observer Member. the Italian association grouping companies, professionals, universities, and research institutions working in the | [email protected] field of modelling, simulation, virtual reality and 3D, Albanian Simulation Goup, attn. Kozeta Sevrani with the aim of enhancing the culture of ‘virtuality’ in University of Tirana, Faculty of Economy Italy, in every application area. rr. Elbasanit, Tirana 355 Albania MIMOS became EUROSIM Observer Member in 2016 and is preparing application for full membership. Albanian Simulation Society- Officers (Planned) President Kozeta Sevrani, ĺ www.mimos.it [email protected] | [email protected] – [email protected] Secretary MIMOS – Movimento Italiano Modellazione e Simulazio- Treasurer ne; via Ugo Foscolo 4, 10126 Torino – via Laurentina Repr. EUROSIM Kozeta Sevrani, 760, 00143 Roma [email protected] Edit. Board SNE Albana Gorishti, [email protected] Majlinda Godolja, [email protected] Last data update December 2016

N 8 SNE 26(4) – 12/2016 EUROSIM 2019 9th EUROSIM Congress on Modelling and Simulation

La Rioja, Logroño, Spain, July 2019

EUROSIM Congresses are the most important modelling and simulation events in Europe. For EUROSIM 2019, we are soliciting original submissions describing novel research and developments in the following (and related) areas of interest: Continuous, discrete (event) and hybrid modelling, simulation, identification and optimization approaches. Two basic contribution motivations are expected: M&S Methods and Technologies and M&S Applications. Contributions from both technical and non-technical areas are welcome.

Congress Topics The EUROSIM 2019 Congress will include invited talks, parallel, special and poster sessions, exhibition and versatile technical and social tours. The Congress topics of interest include, but are not limited to:

Intelligent Systems and Applications Bioinformatics, Medicine, Pharmacy Simulation Methodologies and Tools Hybrid and Soft Computing and Bioengineering Parallel and Distributed Data & Semantic Mining Water and Wastewater Treatment, Architectures and Systems Neural Networks, Fuzzy Systems & Sludge Management and Biogas Operations Research Evolutionary Computation Production Discrete Event Systems Image, Speech & Signal Processing Condition monitoring, Mechatronics Manufacturing and Workflows Systems Intelligence and and maintenance Adaptive Dynamic Programming Intelligence Systems Automotive applications and Reinforcement Learning Autonomous Systems e-Science and e-Systems Mobile/Ad hoc wireless Energy and Power Systems Industry, Business, Management, networks, mobicast, sensor Mining and Metal Industry Human Factors and Social Issues placement, target tracking Forest Industry Virtual Reality, Visualization, Control of Intelligent Systems Buildings and Construction Computer Art and Games Robotics, Cybernetics, Control Communication Systems Internet Modelling, Semantic Web Engineering, & Manufacturing Circuits, Sensors and Devices and Ontologies Transport, Logistics, Harbour, Shipping Security Modelling and Simulation Computational Finance & Economics and Marine Simulation

Congress Venue / Social Events The Congress will be held in the City of Logroño, Capital of La Rioja, Northern Spain. The main venue and the exhibition site is the University of La Rioja (UR), located on a modern campus in Logroño, capital of La Rioja, where 7500 students are registered. The UR is the only University in this small, quiet region in Northern Spain. La Rioja is where the Monasteries of San Millán de la Cogolla, cradle of the first words written in the Spanish language, are situated, sites included in UNESCO’s World Heritage List in 1996. Of course, social events will reflect this heritage – and the famous wines in la Rioja.

Congress Team: The Congress is organised by CAE CAE-SMSG, the Spanish simulation society, and Universidad de la Rioja. Info: Emilio Jiménez, EUROSIM President, [email protected] Juan Ignacio Latorre, [email protected] www.eurosim.info A modern approach to modeling and simulation

With MapleSim, educators have an industry-proven tool to help bridge the gap between theory and practice.

• MapleSim illustrates concepts, and helps students learn the connection between theory and physical behavior

• A wide variety of models are available to help get started right away

MapleSim is built on Maple, which combines the world’s most powerful mathematical computation engine with an intuitive, “clickable” user interface.

To learn more about how you can reinforce engineering concepts using a combination of theory, simulation, and hardware, view this webinar. www.maplesoft.com/SNEWebinar

www.maplesoft.com | [email protected]

© 2016 Maplesoft, ein Bereich von Waterloo Maple Inc., 615 Kumpf Drive, Waterloo, ON, N2V1K8, Kanada. Bei Maplesoft, Maple und MapleSim handelt es sich jeweils um Warenzeichen von Waterloo Maple Inc. Alle anderen Warenzeichen sind Eigentum ihrer jeweiligen Inhaber.