Computational Methods and Experimental Measurements XIV (WIT Transactions on Modelling and Simulation) (Wit Transactions on Modeling and Simulation; Fourteenth ... Methods and Experimental Measurements)

Computational Methods and Experimental Measurements XIV WIT Press publishes leading books in Science and Technology. V...

Author: C. A. Brebbia | G. M. (editors) Carlomagno | C. A. Brebbia | G. M. Carlomagno

51 downloads 988 Views 20MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Computational Methods and Experimental Measurements XIV

WIT Press publishes leading books in Science and Technology. Visit our website for new and current list of titles. www.witpress.com

WITeLibrary Home of the Transactions of the Wessex Institute. Papers presented at Computational Methods and Experimental Measurements XIV are archived in the WIT eLibrary in volume 48 of WIT Transactions on Modelling and Simulation (ISSN 1743-355X). The WIT eLibrary provides the international scientific community with immediate and permanent access to individual papers presented at WIT conferences. Visit the WIT eLibrary at www.witpress.com.

FOURTEENTH INTERNATIONAL CONFERENCE ON COMPUTATIONAL METHODS AND EXPERIMENTAL MEASUREMENTS

CMEM XIV CONFERENCE CHAIRMEN C.A. Brebbia Wessex Institute of Technology, UK G. M. Carlomagno University of Naples di Napoli, Italy

INTERNATIONAL SCIENTIFIC ADVISORY COMMITTEE I. Abdalla A. Beinat Z. Bielecki A. Britten L. Cascini R. Cerny G. Dalla Fontana J. Everett S. Fattorelli L. Fryba W. Graf L. Guerriero

C. Karayannis J. Kompenhans R. Liebe O. Manca P. Prochazka H. Sakamoto K. Takayama M. Trajkovic M. Tsutahara F. Viadero Rueda M. Wnuk G. Zappala

Organised by Wessex Institute of Technology, UK Sponsored by WIT Transactions on Modelling and Simulation

WIT Transactions Transactions Editor Carlos Brebbia Wessex Institute of Technology Ashurst Lodge, Ashurst Southampton SO40 7AA, UK Email: [email protected]

Editorial Board B Abersek University of Maribor, Slovenia Y N Abousleiman University of Oklahoma, USA P L Aguilar University of Extremadura, Spain K S Al Jabri Sultan Qaboos University, Oman E Alarcon Universidad Politecnica de Madrid, Spain A Aldama IMTA, Mexico C Alessandri Universita di Ferrara, Italy D Almorza Gomar University of Cadiz, Spain B Alzahabi Kettering University, USA J A C Ambrosio IDMEC, Portugal A M Amer Cairo University, Egypt S A Anagnostopoulos University of Patras, Greece M Andretta Montecatini, Italy E Angelino A.R.P.A. Lombardia, Italy H Antes Technische Universitat Braunschweig, Germany M A Atherton South Bank University, UK A G Atkins University of Reading, UK D Aubry Ecole Centrale de Paris, France H Azegami Toyohashi University of Technology, Japan A F M Azevedo University of Porto, Portugal J Baish Bucknell University, USA J M Baldasano Universitat Politecnica de Catalunya, Spain J G Bartzis Institute of Nuclear Technology, Greece A Bejan Duke University, USA

M P Bekakos Democritus University of Thrace, Greece G Belingardi Politecnico di Torino, Italy R Belmans Katholieke Universiteit Leuven, Belgium C D Bertram The University of New South Wales, Australia D E Beskos University of Patras, Greece S K Bhattacharyya Indian Institute of Technology, India E Blums Latvian Academy of Sciences, Latvia J Boarder Cartref Consulting Systems, UK B Bobee Institut National de la Recherche Scientifique, Canada H Boileau ESIGEC, France J J Bommer Imperial College London, UK M Bonnet Ecole Polytechnique, France C A Borrego University of Aveiro, Portugal A R Bretones University of Granada, Spain J A Bryant University of Exeter, UK F-G Buchholz Universitat Gesanthochschule Paderborn, Germany M B Bush The University of Western Australia, Australia F Butera Politecnico di Milano, Italy J Byrne University of Portsmouth, UK W Cantwell Liverpool University, UK D J Cartwright Bucknell University, USA P G Carydis National Technical University of Athens, Greece J J Casares Long Universidad de Santiago de Compostela, Spain, M A Celia Princeton University, USA A Chakrabarti Indian Institute of Science, India

A H-D Cheng University of Mississippi, USA J Chilton University of Lincoln, UK C-L Chiu University of Pittsburgh, USA H Choi Kangnung National University, Korea A Cieslak Technical University of Lodz, Poland S Clement Transport System Centre, Australia M W Collins Brunel University, UK J J Connor Massachusetts Institute of Technology, USA M C Constantinou State University of New York at Buffalo, USA D E Cormack University of Toronto, Canada M Costantino Royal Bank of Scotland, UK D F Cutler Royal Botanic Gardens, UK W Czyczula Krakow University of Technology, Poland M da Conceicao Cunha University of Coimbra, Portugal A Davies University of Hertfordshire, UK M Davis Temple University, USA A B de Almeida Instituto Superior Tecnico, Portugal E R de Arantes e Oliveira Instituto Superior Tecnico, Portugal L De Biase University of Milan, Italy R de Borst Delft University of Technology, Netherlands G De Mey University of Ghent, Belgium A De Montis Universita di Cagliari, Italy A De Naeyer Universiteit Ghent, Belgium W P De Wilde Vrije Universiteit Brussel, Belgium L Debnath University of Texas-Pan American, USA N J Dedios Mimbela Universidad de Cordoba, Spain G Degrande Katholieke Universiteit Leuven, Belgium S del Giudice University of Udine, Italy G Deplano Universita di Cagliari, Italy I Doltsinis University of Stuttgart, Germany M Domaszewski Universite de Technologie de Belfort-Montbeliard, France J Dominguez University of Seville, Spain

K Dorow Pacific Northwest National Laboratory, USA W Dover University College London, UK C Dowlen South Bank University, UK J P du Plessis University of Stellenbosch, South Africa R Duffell University of Hertfordshire, UK A Ebel University of Cologne, Germany E E Edoutos Democritus University of Thrace, Greece G K Egan Monash University, Australia K M Elawadly Alexandria University, Egypt K-H Elmer Universitat Hannover, Germany D Elms University of Canterbury, New Zealand M E M El-Sayed Kettering University, USA D M Elsom Oxford Brookes University, UK A El-Zafrany Cranfield University, UK F Erdogan Lehigh University, USA F P Escrig University of Seville, Spain D J Evans Nottingham Trent University, UK J W Everett Rowan University, USA M Faghri University of Rhode Island, USA R A Falconer Cardiff University, UK M N Fardis University of Patras, Greece P Fedelinski Silesian Technical University, Poland H J S Fernando Arizona State University, USA S Finger Carnegie Mellon University, USA J I Frankel University of Tennessee, USA D M Fraser University of Cape Town, South Africa M J Fritzler University of Calgary, Canada U Gabbert Otto-von-Guericke Universitat Magdeburg, Germany G Gambolati Universita di Padova, Italy C J Gantes National Technical University of Athens, Greece L Gaul Universitat Stuttgart, Germany A Genco University of Palermo, Italy N Georgantzis Universitat Jaume I, Spain G S Gipson Oklahoma State University, USA P Giudici Universita di Pavia, Italy F Gomez Universidad Politecnica de Valencia, Spain R Gomez Martin University of Granada, Spain

D Goulias University of Maryland, USA K G Goulias Pennsylvania State University, USA F Grandori Politecnico di Milano, Italy W E Grant Texas A & M University, USA S Grilli University of Rhode Island, USA R H J Grimshaw, Loughborough University, UK D Gross Technische Hochschule Darmstadt, Germany R Grundmann Technische Universitat Dresden, Germany A Gualtierotti IDHEAP, Switzerland R C Gupta National University of Singapore, Singapore J M Hale University of Newcastle, UK K Hameyer Katholieke Universiteit Leuven, Belgium C Hanke Danish Technical University, Denmark K Hayami National Institute of Informatics, Japan Y Hayashi Nagoya University, Japan L Haydock Newage International Limited, UK A H Hendrickx Free University of Brussels, Belgium C Herman John Hopkins University, USA S Heslop University of Bristol, UK I Hideaki Nagoya University, Japan D A Hills University of Oxford, UK W F Huebner Southwest Research Institute, USA J A C Humphrey Bucknell University, USA M Y Hussaini Florida State University, USA W Hutchinson Edith Cowan University, Australia T H Hyde University of Nottingham, UK M Iguchi Science University of Tokyo, Japan D B Ingham University of Leeds, UK L Int Panis VITO Expertisecentrum IMS, Belgium N Ishikawa National Defence Academy, Japan J Jaafar UiTm, Malaysia W Jager Technical University of Dresden, Germany Y Jaluria Rutgers University, USA C M Jefferson University of the West of England, UK

P R Johnston Griffith University, Australia D R H Jones University of Cambridge, UK N Jones University of Liverpool, UK D Kaliampakos National Technical University of Athens, Greece N Kamiya Nagoya University, Japan D L Karabalis University of Patras, Greece M Karlsson Linkoping University, Sweden T Katayama Doshisha University, Japan K L Katsifarakis Aristotle University of Thessaloniki, Greece J T Katsikadelis National Technical University of Athens, Greece E Kausel Massachusetts Institute of Technology, USA H Kawashima The University of Tokyo, Japan B A Kazimee Washington State University, USA S Kim University of Wisconsin-Madison, USA D Kirkland Nicholas Grimshaw & Partners Ltd, UK E Kita Nagoya University, Japan A S Kobayashi University of Washington, USA T Kobayashi University of Tokyo, Japan D Koga Saga University, Japan A Konrad University of Toronto, Canada S Kotake University of Tokyo, Japan A N Kounadis National Technical University of Athens, Greece W B Kratzig Ruhr Universitat Bochum, Germany T Krauthammer Penn State University, USA C-H Lai University of Greenwich, UK M Langseth Norwegian University of Science and Technology, Norway B S Larsen Technical University of Denmark, Denmark F Lattarulo, Politecnico di Bari, Italy A Lebedev Moscow State University, Russia L J Leon University of Montreal, Canada D Lewis Mississippi State University, USA S lghobashi University of California Irvine, USA K-C Lin University of New Brunswick, Canada A A Liolios Democritus University of Thrace, Greece

S Lomov Katholieke Universiteit Leuven, Belgium J W S Longhurst University of the West of England, UK G Loo The University of Auckland, New Zealand J Lourenco Universidade do Minho, Portugal J E Luco University of California at San Diego, USA H Lui State Seismological Bureau Harbin, China C J Lumsden University of Toronto, Canada L Lundqvist Division of Transport and Location Analysis, Sweden T Lyons Murdoch University, Australia Y-W Mai University of Sydney, Australia M Majowiecki University of Bologna, Italy D Malerba Università degli Studi di Bari, Italy G Manara University of Pisa, Italy B N Mandal Indian Statistical Institute, India Ü Mander University of Tartu, Estonia H A Mang Technische Universitat Wien, Austria, G D, Manolis, Aristotle University of Thessaloniki, Greece W J Mansur COPPE/UFRJ, Brazil N Marchettini University of Siena, Italy J D M Marsh Griffith University, Australia J F Martin-Duque Universidad Complutense, Spain T Matsui Nagoya University, Japan G Mattrisch DaimlerChrysler AG, Germany F M Mazzolani University of Naples “Federico II”, Italy K McManis University of New Orleans, USA A C Mendes Universidade de Beira Interior, Portugal, R A Meric Research Institute for Basic Sciences, Turkey J Mikielewicz Polish Academy of Sciences, Poland N Milic-Frayling Microsoft Research Ltd, UK R A W Mines University of Liverpool, UK C A Mitchell University of Sydney, Australia

K Miura Kajima Corporation, Japan A Miyamoto Yamaguchi University, Japan T Miyoshi Kobe University, Japan G Molinari University of Genoa, Italy T B Moodie University of Alberta, Canada D B Murray Trinity College Dublin, Ireland G Nakhaeizadeh DaimlerChrysler AG, Germany M B Neace Mercer University, USA D Necsulescu University of Ottawa, Canada F Neumann University of Vienna, Austria S-I Nishida Saga University, Japan H Nisitani Kyushu Sangyo University, Japan B Notaros University of Massachusetts, USA P O’Donoghue University College Dublin, Ireland R O O’Neill Oak Ridge National Laboratory, USA M Ohkusu Kyushu University, Japan G Oliveto Universitá di Catania, Italy R Olsen Camp Dresser & McKee Inc., USA E Oñate Universitat Politecnica de Catalunya, Spain K Onishi Ibaraki University, Japan P H Oosthuizen Queens University, Canada E L Ortiz Imperial College London, UK E Outa Waseda University, Japan A S Papageorgiou Rensselaer Polytechnic Institute, USA J Park Seoul National University, Korea G Passerini Universita delle Marche, Italy B C Patten, University of Georgia, USA G Pelosi University of Florence, Italy G G Penelis, Aristotle University of Thessaloniki, Greece W Perrie Bedford Institute of Oceanography, Canada R Pietrabissa Politecnico di Milano, Italy H Pina Instituto Superior Tecnico, Portugal M F Platzer Naval Postgraduate School, USA D Poljak University of Split, Croatia V Popov Wessex Institute of Technology, UK H Power University of Nottingham, UK D Prandle Proudman Oceanographic Laboratory, UK

M Predeleanu University Paris VI, France M R I Purvis University of Portsmouth, UK I S Putra Institute of Technology Bandung, Indonesia Y A Pykh Russian Academy of Sciences, Russia F Rachidi EMC Group, Switzerland M Rahman Dalhousie University, Canada K R Rajagopal Texas A & M University, USA T Rang Tallinn Technical University, Estonia J Rao Case Western Reserve University, USA A M Reinhorn State University of New York at Buffalo, USA A D Rey McGill University, Canada D N Riahi University of Illinois at UrbanaChampaign, USA B Ribas Spanish National Centre for Environmental Health, Spain K Richter Graz University of Technology, Austria S Rinaldi Politecnico di Milano, Italy F Robuste Universitat Politecnica de Catalunya, Spain J Roddick Flinders University, Australia A C Rodrigues Universidade Nova de Lisboa, Portugal F Rodrigues Poly Institute of Porto, Portugal C W Roeder University of Washington, USA J M Roesset Texas A & M University, USA W Roetzel Universitaet der Bundeswehr Hamburg, Germany V Roje University of Split, Croatia R Rosset Laboratoire d’Aerologie, France J L Rubio Centro de Investigaciones sobre Desertificacion, Spain T J Rudolphi Iowa State University, USA S Russenchuck Magnet Group, Switzerland H Ryssel Fraunhofer Institut Integrierte Schaltungen, Germany S G Saad American University in Cairo, Egypt M Saiidi University of Nevada-Reno, USA R San Jose Technical University of Madrid, Spain F J Sanchez-Sesma Instituto Mexicano del Petroleo, Mexico

B Sarler Nova Gorica Polytechnic, Slovenia S A Savidis Technische Universitat Berlin, Germany A Savini Universita de Pavia, Italy G Schmid Ruhr-Universitat Bochum, Germany R Schmidt RWTH Aachen, Germany B Scholtes Universitaet of Kassel, Germany W Schreiber University of Alabama, USA A P S Selvadurai McGill University, Canada J J Sendra University of Seville, Spain J J Sharp Memorial University of Newfoundland, Canada Q Shen Massachusetts Institute of Technology, USA X Shixiong Fudan University, China G C Sih Lehigh University, USA L C Simoes University of Coimbra, Portugal A C Singhal Arizona State University, USA P Skerget University of Maribor, Slovenia J Sladek Slovak Academy of Sciences, Slovakia V Sladek Slovak Academy of Sciences, Slovakia A C M Sousa University of New Brunswick, Canada H Sozer Illinois Institute of Technology, USA D B Spalding CHAM, UK P D Spanos Rice University, USA T Speck Albert-Ludwigs-Universitaet Freiburg, Germany C C Spyrakos National Technical University of Athens, Greece I V Stangeeva St Petersburg University, Russia J Stasiek Technical University of Gdansk, Poland G E Swaters University of Alberta, Canada S Syngellakis University of Southampton, UK J Szmyd University of Mining and Metallurgy, Poland S T Tadano Hokkaido University, Japan H Takemiya Okayama University, Japan I Takewaki Kyoto University, Japan C-L Tan Carleton University, Canada M Tanaka Shinshu University, Japan E Taniguchi Kyoto University, Japan

S Tanimura Aichi University of Technology, Japan J L Tassoulas University of Texas at Austin, USA M A P Taylor University of South Australia, Australia A Terranova Politecnico di Milano, Italy E Tiezzi University of Siena, Italy A G Tijhuis Technische Universiteit Eindhoven, Netherlands T Tirabassi Institute FISBAT-CNR, Italy S Tkachenko Otto-von-GuerickeUniversity, Germany N Tosaka Nihon University, Japan T Tran-Cong University of Southern Queensland, Australia R Tremblay Ecole Polytechnique, Canada I Tsukrov University of New Hampshire, USA R Turra CINECA Interuniversity Computing Centre, Italy S G Tushinski Moscow State University, Russia J-L Uso Universitat Jaume I, Spain E Van den Bulck Katholieke Universiteit Leuven, Belgium D Van den Poel Ghent University, Belgium R van der Heijden Radboud University, Netherlands R van Duin Delft University of Technology, Netherlands P Vas University of Aberdeen, UK W S Venturini University of Sao Paulo, Brazil

R Verhoeven Ghent University, Belgium A Viguri Universitat Jaume I, Spain Y Villacampa Esteve Universidad de Alicante, Spain F F V Vincent University of Bath, UK S Walker Imperial College, UK G Walters University of Exeter, UK B Weiss University of Vienna, Austria H Westphal University of Magdeburg, Germany J R Whiteman Brunel University, UK Z-Y Yan Peking University, China S Yanniotis Agricultural University of Athens, Greece A Yeh University of Hong Kong, China J Yoon Old Dominion University, USA K Yoshizato Hiroshima University, Japan T X Yu Hong Kong University of Science & Technology, Hong Kong M Zador Technical University of Budapest, Hungary K Zakrzewski Politechnika Lodzka, Poland M Zamir University of Western Ontario, Canada R Zarnic University of Ljubljana, Slovenia G Zharkova Institute of Theoretical and Applied Mechanics, Russia N Zhong Maebashi Institute of Technology, Japan H G Zimmermann Siemens AG, Germany

Computational Methods and Experimental Measurements XIV EDITORS C.A. Brebbia Wessex Institute of Technology, UK G.M. Carlomagno University of Naples Federico II, Italy

Editors: C.A. Brebbia Wessex Institute of Technology, UK G.M. Carlomagno University of Naples Federico II, Italy

Published by WIT Press Ashurst Lodge, Ashurst, Southampton, SO40 7AA, UK Tel: 44 (0) 238 029 3223; Fax: 44 (0) 238 029 2853 E-Mail: [email protected] http://www.witpress.com For USA, Canada and Mexico Computational Mechanics Inc 25 Bridge Street, Billerica, MA 01821, USA Tel: 978 667 5841; Fax: 978 667 7582 E-Mail: [email protected] http://www.witpress.com British Library Cataloguing-in-Publication Data A Catalogue record for this book is available from the British Library ISBN: 978-1-84564-187-0 ISSN: 1746-4064 (print) ISSN: 1743-355X (on-line) The texts of the papers in this volume were set individually by the authors or under their supervision. Only minor corrections to the text may have been carried out by the publisher. No responsibility is assumed by the Publisher, the Editors and Authors for any injury and/ or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. The Publisher does not necessarily endorse the ideas held, or views expressed by the Editors or Authors of the material contained in its publications. © WIT Press 2009 Printed in Great Britain by Athenaeum Press Ltd. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of the Publisher.

Preface

This book contains the majority of the papers presented at the 14th International Conference on Computational Methods and Experimental Measurements (CMEM/ 09) held in Algarve (Portugal) in 2009. The series of these events is unique in the field and now part of a quite long tradition. Indeed, the first conference took place in Washington DC in 1981 and has been reconvened every two years, in several locations, with uninterrupted success. The main scope of the meeting is to provide to the international technical and scientific community a forum to discuss the interaction and the complementary aspects of computational methods and experimental measurements, the main consideration and importance being committed to their advantageous integration. It is well known that the stable progresses in computers efficiency and numerical techniques are producing a steady growth of computational simulations which nowadays influence both an ever-widening range of engineering problems, as well as our everyday activities. As these simulations are continuously expanding and improving, there still exists the necessity of their validation, which can be only accomplished by performing dedicated experimental tests. Furthermore, because of their incessant development, experimental techniques are becoming more complex and sophisticated so that, both their running as well as data collection can only be performed by means of computers. Finally, it must be stressed that, for the majority of the measurements, the obtained data must be processed by means of numerical methods. This volume contains a substantial number of excellent scientific papers, which present advanced approaches to modern research problems. They have been grouped in the following sections: • • • •

Computational and experimental methods Experimental and computational analysis Direct, indirect and in-situ measurements Detection and signal processing

• • • • • • •

Data processing Fluid flow Heat transfer and thermal processes Material characterization Structural and stress analysis Industrial applications Forest fires

The Editors are very grateful to all the authors for their valuable contributions and to the Members of the International Scientific Advisory Committee, as well as other colleagues, for their help in reviewing the papers published in this book. The Editors Algarve, 2009

Contents Section 1: Computational and experimental methods Heat and moisture transport in porous materials involving cyclic wetting and drying R. Černý, J. Maděra, J. Kočí & E. Vejmelková ................................................... 3 Influence of material characteristics of concrete and thermal insulation on the service life of exterior renders J. Maděra, V. Kočí, E. Vejmelková, R. Černý, P. Rovnaníková, M. Ondráček & M. Sedlmajer............................................................................ 13 A procedure for adaptive evaluation of numerical and experimental data J. Krok, M. Stanuszek & J. Wojtas..................................................................... 25 Gear predictor of manual transmission vehicles based on artificial neural network A. M. Wefky, F. Espinosa, M. Mazo, J. A. Jiménez, E. Santiso, A. Gardel & D. Pérez ........................................................................................ 37 Non-thermal, chemical destruction of PCB from Sydney tar ponds soil extract A. J. Britten & S. MacKenzie ............................................................................. 47 Section 2: Experimental and computational analysis Multi-scale FE analyses of sheet formability based on SEM-EBSD crystal texture measurement H. Sakamoto, H. Kuramae, E. Nakamachi & H. Morimoto............................... 61 Direct simulation of sounds generated by collision between water drop and water surface M. Tsutahara, S. Tajiri, T. Miyaoka, N. Kobata & H. Tanaka .......................... 71

Experimental and numerical analysis of concrete slabs prestressed with composite reinforcement R. Sovják, P. Máca, P. Konvalinka & J. L. Vítek............................................... 83 Measures in the underground work method to determine the mathematical relations that predicts rock behaviour S. Torno, J. Velasco, I. Diego, J. Toraño, M. Menéndez, M. Gent & J. Roldán.......................................................................................... 95 Implementation and validation of a strain rate dependent model for carbon foam G. Janszen & P. G. Nettuno............................................................................. 105 Statics and dynamics of carbon fibre reinforcement composites on steel orthotropic decks L. Frýba, M. Pirner & Sh. Urushadze ............................................................. 117 Study of the thermo-physical properties of bitumen in hydrocarbon condensates A. Miadonye, J. Cyr, K. Secka & A. Britten..................................................... 125 Section 3: Direct, indirect and in-situ measurements Modelling energy consumption in test cells D. Braga, Y. Parte, M. Fructus, T. Touya, M. Masmoudi, T. Wylot & V. Kearley...................................................................................... 137 Evaluation of insulation systems by in situ testing I. Enache, D. Braga, C. Portet & M. Duran.................................................... 147 Monitoring coupled moisture and salt transport using single vertical suction experiment Z. Pavlík, J. Mihulka, M. Pavlíková & R. Černý.............................................. 157 Application of image analysis for the measurement of liquid metal free surface S. Golak ........................................................................................................... 169 Damage assessment by automated quantitative image analysis – a risky undertaking P. Stroeven....................................................................................................... 179

Section 4: Detection and signal processing Special session chaired by A. Kawalec Antenna radiation patterns indication on the basic measurement of field radiation in the near zone M. Wnuk........................................................................................................... 191 Sub-ppb NOx detection by a cavity enhanced absorption spectroscopy system with blue and infrared diode lasers Z. Bielecki, M. Leszczynski, K. Holz, L. Marona, J. Mikolajczyk, M. Nowakowski, P. Perlin, B. Rutecka, T. Stacewicz & J. Wojtas................... 203 Multispectral detection circuits in special applications Z. Bielecki, W. Kolosowski, E. Sedek, M. Wnuk & J. Wojtas........................... 217 Modification of raised cosine weighting functions family C. Lesnik, A. Kawalec & J. Pietrasinski .......................................................... 229 Technique for the electric and magnetic parameter measurement of powdered materials R. Kubacki, L. Nowosielski, R. Przesmycki...................................................... 241 Acoustic watermark server effectiveness Z. Piotrowski & P. Gajewski ........................................................................... 251 Intrapulse analysis of radar signal A. Pieniężny & S. Konatowski ......................................................................... 259 Neural detection of parameter changes in a dynamic system using time-frequency transforms E. Swiercz ........................................................................................................ 271 Section 5: Data processing A versatile software-hardware system for environmental data acquisition and transmission G. Zappalà ....................................................................................................... 283 Modelling of the precise movement of a ship at slow speed to minimize the trajectory deviation risk J. Malecki......................................................................................................... 295 Automated safe control of a Self-propelled Mine Counter Charge in an underwater environment P. Szymak......................................................................................................... 305

A novel financial model of long term growing stocks for the Taiwan stock market S.-H. Liang, S.-C. Liang, L.-C. Lien & C.-C. Liang ........................................ 315 Section 6: Fluid flow On the differences of transitional separated-reattached flows over leading-edge obstacles of varying geometries I. E. Abdalla..................................................................................................... 329 A second order method for solving turbulent shallow flows J. Fe & F. Navarrina ....................................................................................... 341 Numerical analysis of compressible turbulent helical flow in a Ranque-Hilsch vortex tube R. Ricci, A. Secchiaroli, V. D’Alessandro & S. Montelpare ............................ 353 Turbulence: a new zero-equation model K. Alammar...................................................................................................... 365 Mesh block refinement technique for incompressible flows in complex geometries using Cartesian grids C. Georgantopoulou, G. Georgantopoulos & S. Tsangaris............................. 369 Application of the finite volume method for the supersonic flow around the axisymmetric cone body placed in a free stream R. Haoui........................................................................................................... 379 Some aspects and aerodynamic effects in repairing battle damaged wings S. Djellal & A. Ouibrahim ............................................................................... 389 Section 7: Heat transfer and thermal processes Natural and mixed convection in inclined channels with partial openings A. Andreozzi, B. Buonomo, O. Manca & S. Nardini ........................................ 401 Heat flux reconstruction in the grinding process from temperature data J. Irša & A. N. Galybin .................................................................................... 413 Analysis of non-isothermal fluid flow past an in-line tube bank M. Alavi & H. Goshayeshi ............................................................................... 425

Transport phenomenon in a jet type mold cooling pipe H. Kawahara & T. Nishimura ......................................................................... 437 Two-phase modelling of nanofluid heat transfer in a microchannel heat sink C. T. Nguyen & M. Le Menn............................................................................ 451 Numerical investigation of sensible thermal energy storage in high temperature solar systems A. Andreozzi, N. Bianco, O. Manca, S. Nardini & V. Naso ............................. 461 Dynamic modelling of the thermal space of the metallurgical walking beams furnaces D. Constantinescu............................................................................................ 473 Section 8: Material characterisation Investigation of shape recovery stress for ferrous shape memory alloy H. Naoi, M. Wada, T. Koike, H. Yamamoto & T. Maruyama .......................... 485 Growth behavior of small surface cracks in coarse and ultrafine grained copper M. Goto, S. Z. Han, Y. Ando, N. Kawagoishi, N. Teshima & S. S. Kim ........... 497 Section 9: Structural and stress analysis Numerical simulation of structures using generalized models for data uncertainty W. Graf, J.-U. Sickert & F. Steinigen .............................................................. 511 A dynamic model for the study of gear transmissions A. Fernandez del Rincon, F. Viadero, R. Sancibrian, P. Garcia Fernandez & A. de Juan.................................................................. 523 Long-term behaviour of concrete structures reinforced with pre-stressed GFRP tendons J. Fornůsek, P. Konvalinka, R. Sovják & J. L. Vítek........................................ 535 Application of the finite element method for static design of plane linear systems with semi-rigid connections D. Zlatkov, S. Zdravkovic, B. Mladenovic & M. Mijalkovic ............................ 547

Blade loss studies in low-pressure turbines – from blade containment to controlled blade-shedding R. Ortiz, M. Herran & H. Chalons .................................................................. 559 Section 10: Industrial applications Finding the “optimal” size and location of treatment plants for a Jatropha oil plantation project in Thailand J. E. Everett ..................................................................................................... 571 An industrial ship system for the flat development of undevelopable surfaces: algorithm and implementation E. M. Soto, X. A. Leiceaga & S. García........................................................... 579 Section 11: Forest fires Assessment of the plume theory predictions of crown scorch or crown fire initiation using transport models V. Konovalov, J.-L. Dupuy, F. Pimont, D. Morvan & R. R. Linn..................................................................................................... 593 Spotting ignition of fuel beds by firebrands C. Lautenberger & A. C. Fernandez-Pello ...................................................... 603 Impact of fuel-break structure on fire behaviour simulated with FIRETEC F. Pimont, J.-L. Dupuy & R. R. Linn ............................................................... 613 A new model of information systems for public awareness about wildfires P.-Y. Badillo & C. Sybord................................................................................ 623 Combustion modelling for forest fires: from detailed to skeletal and global models P. A. Santoni .................................................................................................... 633 Author Index .................................................................................................. 645

Section 1 Computational and experimental methods

This page intentionally left blank


3

Heat and moisture transport in porous materials involving cyclic wetting and drying R. Černý, J. Maděra, J. Kočí & E. Vejmelková Czech Technical University in Prague, Faculty of Civil Engineering, Department of Materials Engineering and Chemistry, Czech Republic

Abstract Computational modeling of coupled heat and moisture transport in porous building materials with hysteretic moisture transport and storage parameters in the conditions of difference climate is presented in the paper. A diffusion-type model is used for the description of coupled heat and moisture transport. An empirical procedure is chosen to describe the path between the transport and storage parameters corresponding to wetting and drying. In a practical example of computer simulation, a concrete wall provided with exterior thermal insulation is analyzed. Computational results reveal very significant differences in moisture and relative humidity profiles calculated using the model with hysteretic parameters and without hysteresis. As the differences are on dangerous side from the hygrothermal point of view, the application of hysteretic moisture transport and storage parameters in computational models can be considered as quite important for service life analyses of multi-layered systems of building materials. Keywords: moisture transport, hysteresis, computer simulation.

1

Introduction

Heat and moisture transport calculations are quite common in service life analyses of multi-layered systems of building materials. They make it possible to identify potential weak points in building envelopes from the hygro-thermal point of view, thus allow to react to possible danger in time and prevent excessive damage caused for instance by accumulation of liquid water in specific parts of a structure. Although it is known for years that moisture transport and storage parameters of many porous materials may exhibit considerable hysteresis, most of the calculations are still performed with parameters measured WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090011

4 Computational Methods and Experimental Measurements XIV during the adsorption phase which is an apparent consequence of difficulties the experimentalists face in the measurements of some parameters in desorption phase. In this paper, the effect of hysteresis of moisture transport and storage parameters on calculated moisture and relative humidity fields is investigated for a characteristic case of concrete wall with exterior thermal insulation, and the possible consequences of neglecting hysteresis of these parameters for service life calculations are analyzed.

2

Mathematical model

The diffusion model proposed by Künzel [1] was used for description of coupled heat and moisture transport. The heat and moisture balance equations were formulated in the form

[

]

dH ∂T = div (λ grad T ) + Lv div δ p grad (ϕ p s ) , dT ∂t ∂ρ v ∂ϕ = div Dϕ grad ϕ + δ p grad (ϕ p s ) , ∂ϕ ∂t

[

]

(1) (2)

where H is the enthalpy density, Lv heat of evaporation, λ thermal conductivity, T temperature, ρv partial density of moisture, ϕ relative humidity, δp permeability of water vapor, ps partial pressure of saturated water vapor,

Dϕ = Dw

dρ v dϕ

(3)

is the liquid water transport coefficient, DW capillary transport coefficient (moisture diffusivity). The inclusion of cyclic wetting and drying processes into the model was done using different moisture transport and storage parameters functions in wetting and drying phase and calculating the path between the parameter functions corresponding to wetting and drying in every time step. For describing the path between the adsorption and desorption isotherms an empirical procedure was chosen which follows Pedersen’s hysteretic model [2]. The actual value of moisture content, w, is determined using equation w = wp + ξ ϕ a − ϕ p , (4)

(

)

where ϕa is the actual value of relative humidity and ϕp the value of relative humidity from previous calculation step, ξ is the slope of the hysteretic parameter which is calculated as

ad (wp − wa ) ξ d + aa (wp − wd ) ξ a 2

ξ=

(wd − wa )2

2

,

(5)

wp is the value of moisture content from previous calculation step, wa and wd are values of moisture content for adsorption and desorption cycles, ξa and ξd the values for tangent adsorption and desorption in the points wa and wd, aa and ad WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


5

the correction coefficients. An example of the calculation of the path between the adsorption and desorption isotherms is given in Figure 1.

0.6 Adsorption

ξd

Desorption

0.5

Calculated with hysteresis

w [m3/m3]

0.4

ξ 0.3 0.2

ξa

0.1 0 0

Figure 1:

0.2

0.4

ϕ [-]

0.6

0.8

1

Example of application of hysteresis to sorption isotherms during drying and wetting cycles.

In the case of moisture diffusivity, a modification of (5) was necessary in order to express the hysteretic effect in more accurate way. The modified equation can be expressed as

ad (ln κ p − ln κ a ) ln ξ d + aa (ln κ p − ln κ d ) ln ξ a 2

ξ=

(ln κ d − ln κ a )2

2

,

(6)

where κp is the value of moisture diffusivity from previous calculation step.

3

Materials and building envelope

A simplified building envelope system was chosen for the investigation of the effect of hysteresis on the calculated moisture and relative humidity fields. The load-bearing structure was made of high performance concrete containing metakaolin (600 mm). Mineral wool (140 mm) was used as exterior thermal insulation. Lime-cement plaster (10 mm) was on both exterior and interior side. The building envelope was exposed from inside to constant conditions (temperature equal to 21°C and relative humidity equal to 55%) and from outside to climatic conditions corresponding to the reference year for Prague (Fig. 2).

WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

6 Computational Methods and Experimental Measurements XIV INSIDE

OUTSIDE

Constant temperature T = 21 °C Climatical Data from TRY

Constant relative humidity ϕ = 55 %

10

600

140

10

Figure 2:

Scheme of the studied envelope including boundary conditions.

Table 1:

Basic parameters of materials of the studied building envelope.

Parameter ρ whyg wsat λdry λsat κ

Unit [kg/m3] [m3/m3] [m3/m3] [W/mK] [W/mK] [m2/s]

HPCM 2366 0.107 0.13 1.56 2.09 see Fig. 4

c µ

[J/kgK] [-]

730 21

MW 170 0.0073 0.89 0.055 1.20 5.1E-10. e3.12w 1000 45

LCP 1550 0.03 0.40 0.70 2.40 7.3E-7. e3.2w 1200 7

LPMH 1745 0.024 0.33 0.84 2.40 3.9E-8 610 13

The initial conditions were chosen as follows: relative humidity 89 % and constant temperature profile equal to 21°C. The basic material parameters of concrete (HPCM), mineral wool (MW), lime-cement plaster (LCP) and hydrophobized lime plaster modified by metakaolin (LPMH) are shown in Table 1 where the following symbols were used: ρ – bulk density, c – specific heat capacity, µ – water vapor diffusion resistance factor, λdry – thermal conductivity in dry conditions, λsat – thermal conductivity in water saturated conditions, κ - moisture diffusivity, whyg – hygroscopic moisture content by volume, wsat – saturated moisture content by volume. Fig. 3 presents the sorption isotherms of concrete used in the simulations. The data were obtained by experiments performed at the Department of Materials Engineering and Chemistry, Faculty of Civil Engineering, Czech Technical University in Prague [3, 4]. Fig. 4 presents the moisture diffusivity of concrete. Moisture diffusivity vs. moisture function for adsorption was derived according to (7)-(9), where two input parameters were used, the normalized pore distribution curve f(r), Rmin < r < Rmax, and the average value of moisture diffusivity κav determined in a common WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


7

0.12 0.1

Adsorption Desorption

3

w [m /m ]

0.08 3

0.06 0.04 0.02 0 0

0.2

0.4

0.6

0.8

1

ϕ [-]

Figure 3:

Sorption isotherm of concrete.

1.00E-05 Adsorption Desorption

2

κ [m /s]

1.00E-07

1.00E-09

1.00E-11

1.00E-13 0.00

0.02

0.04

0.06

0.08 3

0.10

0.12

3

w [m /m ]

Figure 4:

Moisture diffusivity of concrete.

sorptivity experiment [5], and the tortuosity effect was linearized for the sake of simplicity, i.e., an assumption of n=1 was adopted,

 w κ r (w) =   wsat

2 R 2  Rmax f (Rmax ) ∫ Rmin r f (r )dr  , R 2 f (R ) ∫ RRmax r 2 f (r )dr  min n


(7)

8 Computational Methods and Experimental Measurements XIV

2 3

 

κ  wsat  = κ av

(8)

R

w( R) = wsat

∫ f (r )dr

(9)

R min

Contrary to the sorption isotherms, where desorption curves can be obtained by common experimental techniques, the desorption curve of moisture diffusivity had to be set empirically as for its experimental determination no quite reliable techniques are available at present. Based on the results of experiments and computational analyses described in [6], it was estimated to be one order of magnitude lower than the adsorption curve. The moisture transport and storage parameters of mineral wool insulation and exterior and interior plasters were assumed to be the same in both adsorption and desorption phase. The water vapor adsorption and moisture diffusivity of mineral wool are so low that hysteretic effects are within the error range of experimental methods in any case. The thickness of renders is much lower as compared to both concrete and mineral wool so that the effect of hysteresis in their hygric parameters on moisture transport in the wall as a whole is supposed to be neglected. 1 Without hysteresis With hysteresis SI

0.06 0.05 0.04 0.03 0.02 0.01 0

0.8 0.7 0.6 0.5 0.4

0

Figure 5:

4

Without hysteresis With hysteresis SI

0.9

Relative humidity [-]

Moisture content [m3/m3]

0.07

200

400

Distance [mm]

600

800

0

100

200

300

400

500

Distance [mm]

600

700

800

Moisture content (a) and relative humidity (b) profiles for January 1.

Results of computer simulations and discussion

Three different simulations were performed, combining the effects of hysteresis of moisture transport and storage parameters, namely the simulation with hysteresis of sorption isotherm only, hysteresis of moisture diffusivity only and hysteresis of both sorption isotherm and moisture diffusivity. The calculations without hysteresis were done as well for the sake of comparison, using the data for adsorption phase which is usual in computational simulations where hysteresis is neglected.



9

4.1 Hysteresis of sorption isotherm Figs. 5(a), (b) show moisture content and relative humidity profiles calculated for January 1, which can be considered as characteristic for the winter period. The calculation with hysteretic effects led to increase of moisture content in the envelope, whereas relative humidity was significantly lower. Figs. 6(a), (b) summarize the moisture and relative humidity fields during the whole simulated time period of 5 years calculated with hysteresis.

Figure 6:

Moisture content (a) and relative humidity (b) fields during a 5-year period.

4.2 Hysteresis of moisture diffusivity The moisture and relative humidity profiles for January 1 (Figs. 7(a), (b)) show that both moisture content and relative humidity calculated with hysteresis were higher, which was a consequence of lower moisture diffusivity in drying phase. 0.9 Without hysteresis With hysteresis κ

0.07

Without hysteresis With hysteresis κ



0.08

0.06 0.05 0.04 0.03 0.02

0.8

0.7

0.6

0.5

0.01 0

0

Figure 7:

200

400

Distance [mm]

600

800

0.4

0

100

200

300

400

500

Distance [mm]

600

700

800


Figs. 8(a), (b) show the moisture and relative humidity fields during the whole simulated time period of 5 years calculated with hysteresis.



Figure 8:

Moisture content (a) and relative humidity (b) fields during a 5-year period. 1 Without hysteresis With hysteresis both values

0.07 0.06 0.05 0.04 0.03 0.02

0.8 0.7 0.6 0.5 0.4

0.01 0

Without hysteresis With hysteresis both values

0.9



0.08

0

200

400

Distance [mm]

600

800

0

100

200

300

400

500

Distance [mm]

600

700

800

Figure 9:


Figure 10:

Moisture content (a) and relative humidity (b) fields during a 5-year period.

4.3 Hysteresis of both sorption isotherm and moisture diffusivity The moisture and relative humidity profiles (Figs. 9(a), (b)) were very similar to the simulations presented in 4.1, which meant that the hysteretic effect of sorption isotherm had remarkably higher influence on simulation results. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


11

Figs. 10(a), (b) present the moisture and relative humidity fields during the whole simulated time period of 5 years calculated with hysteresis.

Figure 11:

Calculation of hysteresis of moisture diffusivity (a) and sorption isotherm (b).

Figs. 11(a), (b) show how the hysteretic effects were manifested in the values of moisture diffusivity and sorption isotherm of concrete used by the model during the whole time period of 5-years simulation. While the water vapour sorption parameters were mostly near to the centerline of the area demarcated by the adsorption and desorption curves, the moisture diffusivities oscillated between the maximum and minimum values. 1 Without hysteresis With hysteresis SI With hysteresis κ With both hysteresis

0.07 0.06 0.05 0.04 0.03 0.02

Without hysteresis With hysteresis SI With hysteresis κ With both hysteresis

0.9



0.08

0.8 0.7 0.6 0.5 0.4

365

Figure 12:

730

1095

1460

Time [days]

1825

2100

365

730

1095

1460

Time [days]

1825

2100

Moisture content (a) and relative humidity (b) in a characteristic locality in concrete, 1 cm from the interface between concrete and thermal insulation.

4.4 Comparison of the effects of hysteresis of moisture transport and storage parameters Figs. 12(a), (b) present a comparison of the effects of hysteresis of moisture transport and storage parameters on moisture content and relative humidity for a characteristic locality in concrete, 1 cm from the interface between concrete and thermal insulation. As for the relative humidity profiles, it is obvious that all values came close each other in the 5th year of simulation. Moisture content WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

12 Computational Methods and Experimental Measurements XIV profiles were quite different. Results of simulation without hysteretic effect and with hysteresis of moisture diffusivity only were very similar, but they differed from the other simulations in a significant way. Besides, the yearly oscillations of water content were not steady yet after 5 years. Involving hysteretic effects of both moisture diffusivity and sorption isotherm caused the highest increase of moisture content in the chosen point. Similar results were also obtained for other points within the concrete part of the wall.

5

Conclusions

The computer simulations of coupled heat and moisture transport in this paper have shown that the application of hysteretic moisture transport and storage parameters can be considered as quite important in service life analyses of multilayered systems of porous building materials. The results indicated very significant differences in moisture and relative humidity profiles calculated using the model with hysteretic parameters and without hysteresis. As the differences were always on the dangerous side, it can be concluded that neglecting the hysteretic effects while desorption is in progress can lead to underestimation of damage risk due to water presence in a structure which is rather undesirable in any service life analysis.

Acknowledgement This research has been supported by the Czech Science Foundation, under grant No 103/07/0034.

References [1] Künzel, H. M., Simultaneous Heat and Moisture Transport in Building Components, PhD Thesis, IRB Verlag Stuttgart, 1995. [2] Pedersen, C. R., Combined Heat and Moisture Transfer in Building Constructions, PhD Thesis, Report 214. Thermal Insulation Laboratory, TU Denmark, 1990. [3] Vejmelková, E. & Černý, R., Application of Alternative Silicate Binders in the Production of High Performance Materials Beneficial to The Environment. Proceedings of the 2008 World Sustainable Building Conference [CD-ROM]. Balnarring, Victoria: ASN Events Pty Ltd, 2008, p. 520-525, 2008. [4] Pernicová, R., Pavlíková, M., Pavlík, Z. & Černý, R., Vliv metakaolinu na mechanické, tepelné a vlhkostní vlastnosti vápenných omítek. Metakaolin 2007. Brno: VUT FAST, s. 70-77, 2007 [5] Černý, R. & Rovnaníková, P., Transport Processes in Concrete, 1. ed. London: Spon Press, pp. 26–29, 2002. [6] Pel L., Černý, R. & Pavlík Z, Moisture and Ion Transport. WP5 2-Years Report of the EU 6th Program Project SSPI-CT-2003-501571. TU Eindhoven, Eindhoven, 2006. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


13

Influence of material characteristics of concrete and thermal insulation on the service life of exterior renders J. Maděra1, V. Kočí1, E. Vejmelková1, R. Černý1, P. Rovnaníková2, M. Ondráček3 & M. Sedlmajer3 1

Czech Technical University in Prague, Faculty of Civil Engineering, Department of Materials Engineering and Chemistry, Czech Republic 2 Brno University of Technology, Faculty of Civil Engineering, Institute of Chemistry, Czech Republic 3 Brno University of Technology, Faculty of Civil Engineering, Institute of Technology of Building Materials and Components, Czech Republic

Abstract An assessment of the service life of exterior renders of building structures using combined computational-experimental approach is presented in the paper. In the experimental part, durability of selected renders and concretes is determined in terms of their frost resistance. A diffusion-type model is used for the description of coupled heat and moisture transport aimed at the identification of the number of frost cycles in a real structure. The computational implementation of the model leads to a system of two non-linear partial differential equations with the moisture accumulation function as additional condition. In a practical application of the model, a concrete wall provided with exterior thermal insulation system and both exterior and interior renders is analyzed. The influence of different material composition of building envelope in the service life of exterior renders is analyzed to meet the main objective of the paper. Different types of concrete, thermal insulation materials and renders are under consideration. Conclusions on the most advantageous material composition with respect to the service life of exterior renders are drawn. Keywords: computational analysis, coupled heat and moisture transport, concrete wall, thermal insulation system. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090021


1

Introduction

Degradation of exterior renders is caused by many factors. The most principal are chemical and mechanical corrosion. As a mechanical corrosion we can assume the influence of weather conditions, particularly the destruction effect of freezing water which is contained in the exterior renders. Phase conversion of this water goes along with volume increase and this is the main mechanism leading to material destruction. Different exterior render can be frost-resistant in varying degrees which depends on its material characteristics. The frost resistance could be defined by experimental methods. However, more complicated is to determine the amount of freezing cycles which arise during the year in the investigated exterior render applied on real building structure. Freezing cycle can appear only when two conditions are met. The first condition is the presence of overhygroscopic (liquid) moisture, the second a temperature below water freezing point. So we have to observe the hygrothermal performance of the exterior render and compare the thermal and hygric state in parallel. Computational analysis is the best instrument for this operation. The amount of freezing cycles depends in the first instance on climatic conditions and material composition used in building envelope. Based on experimental and computational results we can then design optimal material composition of building envelope with respect to its service life.

2

Experimental

Determination of frost resistance of exterior renders’ materials and concretes was accomplished under laboratory conditions. For renders, the specimens in size of 40 × 40 × 160 mm were made, for concretes 100 x 100 x 400 mm. Temperature in the laboratory was 21±1°C, relative humidity was 45±5%. Water saturated specimens of renders - lime-cement plaster (LCP) and hydrophobic lime plaster modified by metakaolin (LPMH) - were cyclically frosted and defrosted until their damage got obvious. One frosting and defrosting cycle meant to put the specimens into plastic bag and then into freezing box for 6 hours, after removal to keep the specimens in laboratory with temperature of 20±1°C for 2 hours and then to put the specimens into the water for 16 hours. These cycles were repeated until the damage of specimens was visible. Damage of LPMH specimens is shown on Figure 1. Table 1:

Number of freezing cycles causing damage of material.

LCP Number of freezing cycles > 103

LPMH 40

CF > 100

CM > 100

CS > 100

CR > 100

Frost resistance tests of concretes - concrete modified by fly ash (CF), metakaolin (CM), slag (CS) and reference concrete without any modification (CR) - were carried out according to ČSN 73 1322/Z1:1968 [1]. The samples WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


15

were tested after 28 days of concrete maturing and standard curing. The total test required 100 freezing and thawing cycles. One cycle consisted of 4 hours freezing at -20°C and 2 hours thawing in 20°C warm water.

Figure 1:

3

Hydrophobic lime plaster modified by metakaolin after 40 freezing cycles.

Computational

3.1 Description of construction In this paper we assumed concrete wall made from different types of concrete (CF, CM, CS or CR) provided by thermal insulation system (EPS or mineral wool). The wall is provided by lime-cement plaster on interior side and by modified lime plaster on exterior side. By the same token we assumed the cases when the thermal insulation system is not present. To compare different hygrothermal behaviour of concrete in dependence on its modification we made simulation of simple concrete wall only. The material combination is shown in Figure 2. Hygrothermal performance was investigated in exterior plaster in a point just under the surface and in concrete in a point close to the interface with render. 3.2 Input parameters As the input parameters we need to know characteristics of used materials, boundary and initial conditions. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


Figure 2: Table 2: ρ [kg/m3] ψ [%] c [J/kgK] µdry cup [-] µwet cup [-] λdry [W/mK] λsat [W/mK] κ [m2/s] whyg [m3/m3]

ρ [kg/m3] ψ [%] c [J/kgK] µ [-] λdry [W/mK] λsat [W/mK] κ [m2/s] whyg [m3/m3]

Basic material characteristics of concretes.

CF 2356 12.5 692 44.63 17.18 1.550 1.940 6.49e-9 0.074685

Table 3:

Scheme of building envelope.

CM 2366 13.0 728 32.44 20.99 1.565 2.085 4.09e-9 0.106943

CS 2334 9.7 720 17.70 8.99 1.632 2.077 3.77e-9 0.089000

CR 2380 12.3 672 15.80 6.60 1.660 2.085 7.15e-9 0.083300

Basic material characteristics of plasters. LCP 1550 40 1200 7 0.700 2.40 7.3e-7 0.040

LPMH 1745 33 610 10 0.845 2.40 3.9e-8 0.024

Basic material characteristics of analyzed materials are shown in Tables 2, 3 and 4. We used following symbols: ρ – bulk density [kg/m3], ψ − porosity [%], c – specific heat capacity [J/kgK], µ – water vapour diffusion resistance factor [-], λdry – thermal conductivity in dry conditions [W/mK], λsat – thermal conductivity in water saturated conditions [W/mK], κ - moisture diffusivity [m2/s], whyg – hygroscopic moisture content by volume [m3/m3]. All these parameters were WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


17

measured in laboratory of transport processes at the Department of Materials Engineering and Chemistry, Faculty of Civil Engineering, Czech Technical University in Prague [2–3]. Table 4:

Basic material characteristics of thermal insulation materials. EPS 50 97 1300 50 0.040 0.560 2.1e-11 0.001

3

ρ [kg/m ] ψ [%] c [J/kgK] µ [-] λdry [W/mK] λsat [W/mK] κ [m2/s] whyg [m3/m3]

Mineral wool 170 89 840 3 0.055 1.200 5.1e-10 0.0073

Initial and boundary conditions should be as realistic as possible. Therefore, we used climatic data in exterior for Prague in the form of a Test Referent Year (TRY), which contains average climatic data for 30 years. On interior side we used constant value of relative humidity 55% and temperature 21°C. The simulation started on 1 July and was done for 5 years. 3.3 Computational model The computations were accomplished by the computational program TRANSMAT 7.1, which was developed at the Department of Material Engineering and Chemistry, Faculty of Civil Engineering, Czech Technical University in Prague on the basis of the general finite element package SIFEL. The mathematical formulation of coupled transport of heat and moisture leads to a system of partial differential equations, which are solved by finite element method. In the particular case in this paper, Künzel’s model was used [4]:

dρ v ∂ϕ = div Dϕ gradϕ + δ p grad (ϕps ) dϕ ∂t dH ∂T = div(λgradT ) + Lv div δ p grad (ϕp s ) dT ∂t

[

]

[

]

(1) (2)

where ρv is the partial density of moisture, ϕ relative humidity, δp permeability of water vapour, ps partial pressure of saturated water vapour, H enthalpy density, Lv heat of evaporation, λ thermal conductivity and T temperature,

Dϕ = Dw

dρ v dϕ

is liquid moisture diffusivity coefficient, DW capillary transport coefficient.


(3)

18 Computational Methods and Experimental Measurements XIV 3.4 Results of computational simulation The results are summarized in a set of figures. Every figure shows time dependence of moisture and temperature at the same time during fourth year of simulation. This is advantageous with regard to interpretation of simulation results. There are two horizontal lines in each figure which represent hygroscopic moisture content and temperature of freezing point. From the vast number of figures produced in computational simulations only the most representative are chosen, rest of them is only described. At the beginning, hygrothermal properties of simple concrete wall are analyzed. This allows us to get real image about differences in hygrothermal behaviour of different types of concrete in dependence of their modifications. Analysis of other building envelopes then follows which utilizes the previous knowledge. 3.4.1 Simple concrete wall The most predisposed type of concrete to creation of freezing cycles is the reference concrete. However, due to low moisture content during the studied period there is not any freezing cycle. As we can see on Figure 3, overhygroscopic moisture content is reached only in summer month when the temperature is above zero. Freezing point of water Temperature Hygroscopic moisture content Moisture content

0.180000

0.160000

0.140000

300

Temperature [K]

0.120000 280 0.100000 260

0.080000

0.060000

240

0.040000 220

Moisture content by volume [m3/m3]

320

0.020000

200 1275

0.000000 1325

1375

1425

1475

1525

1575

1625

Time [days]

Figure 3:

Hygrothermal performance of concrete, simple concrete wall (CR), 2 mm under the surface.

In concrete modified by fly ash, the overhygroscopic moisture content is reached once per a reference year, but as in the previous case this happens in summer month so there are not any possibilities of creation of freezing cycles. In other types of concrete the overhygroscopic moisture content is not reached at all. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


19

3.4.2 Concrete wall without thermal insulation system If we consider concrete wall provided only by plasters, the worst material combination is to use reference concrete without any modifications. This material combination leads to creation of two freezing cycles in exterior plaster. The cycles took 2 and 13 hours and arisen in 325th and 326th day of reference year (Figure 4). Other types of concrete do not give suitable conditions for freezing of water. Freezing point of water Temperature Hygroscopic moisture content Moisture content

0.080000

0.070000

Temperature [K]

300

0.060000

0.050000

280

0.040000 260 0.030000 240 0.020000 220

200 1275


320

0.010000

0.000000 1325

1375

1425

1475

1525

1575

1625

Time [days]

Figure 4:

Hygrothermal performance of exterior plaster, concrete wall (CR) provided by exterior plaster, 2 mm under the surface.

In all cases, the water contained in concrete does not get frozen. All the results are similar to result on Figure 5. 3.4.3 Concrete wall provided with thermal insulation system EPS as the insulation material provided with plaster reliably keeps the concrete wall from effects of freezing cycles no matter which type of concrete is under consideration. Nevertheless, disadvantage of this material combination lies in abnormal strain of exterior plaster caused by weather conditions. In the simulation we counted more than 25 freezing cycles in every material combination in exterior plaster per reference year (Figure 6). Duration of freezing cycles is different. The longest one takes 36 hours, the shortest one only 1 hour. Single cycles are separated by tiny temperature or moisture fluctuations which raise their final number. Mineral wool has the same effect as EPS and protects the concrete wall from increase of moisture content and decrease of temperature at the same time. This prevents water in the wall getting overhygroscopic and getting frozen. Anyway, exterior plaster applied on mineral wool is abnormally exposed too. The number of freezing cycles during one referent year is little bit smaller then in plaster WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

20 Computational Methods and Experimental Measurements XIV applied on EPS and counting about 25 cycles in every material combination (Figure 7). As we can see, the cycles are separated by tiny moisture and temperature fluctuations, too. Freezing point of water Temperature Hygroscopic moisture content Moisture content

300

0.180000

0.160000

0.140000

Temperature [K]

280

0.120000

0.100000 260 0.080000

240

0.060000

0.040000 220


320

0.020000

200 1275

0.000000 1325

1375

1425

1475

1525

1575

1625

Time [days]

Hygrothermal properties of concrete, concrete wall (CM) provided by exterior plaster, 2 mm from material interface.

Freezing point of water Temperature Hygroscopic moisture content Moisture content

320

0.080000

0.070000

Temperature [K]

300

0.060000

0.050000

280

0.040000 260 0.030000 240 0.020000 220

200 1275


Figure 5:

0.010000

0.000000 1325

1375

1425

1475

1525

1575

1625

Time [days]

Figure 6:

Hygrothermal performance in exterior plaster, concrete wall (CS) provided by EPS with exterior plaster, 2 mm under the surface.


Computational Methods and Experimental Measurements XIV Freezing point of water Temperature Hygroscopic moisture content Moisture content

0.080000

0.070000

Temperature [K]

300

0.060000

0.050000

280

0.040000 260 0.030000 240 0.020000 220

200 1275


320

21

0.010000

0.000000 1325

1375

1425

1475

1525

1575

1625

Time [days]

Figure 7:

4

Hygrothermal performance in exterior plaster, concrete wall (CF) provided by mineral wool with exterior plaster, 2 mm under the surface.

Discussion

In an assessment of the impact of freezing cycles on service life of exterior renders, it is not enough to consider only the frost resistance determined in the laboratory. It is important to realize, that the number of freezing cycles depends on material combination, not only on material characteristics of used materials. As important as the freezing cycles’ time is the lag between two cycles. If the freezing cycles’ time is too short, water cannot solidify into ice and cannot disrupt the structure of plaster. By the same token, if the lag between two cycles is too short, ice cannot melt back into water and then refreeze again, so its destruction effect can be neglected and we can consider both as one. When we evaluate the number of freezing cycles taking these considerations into account, we get new number of freezing cycles which is reduced (Table 5). In this computer simulation only the liquid moisture appearance caused by rain was considered. However, there can be locations on building, which are exposed to water originating from other sources. This can be the case of a socle part of building and places with bad construction details solution. In these cases, the number of freezing cycles during one year could be much higher. Although simple concrete wall built from all the types of investigated concrete does not show indications of freezing of the contained moisture, the wall made from reference concrete is very close to it. As we considered only reference year which is based on long-term average of relative humidity and temperature, the freezing cycles cannot be completely excluded in every particular year; in real weather conditions deviations from the average values may appear which could lead to creation of some freezing cycles. Basically, this WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

22 Computational Methods and Experimental Measurements XIV Table 5:

Simple concrete wall

Concrete wall provided with exterior plaster Concrete wall provided with EPS and exterior plaster Concrete wall provided with mineral wool and exterior plaster

Number of freezing cycles after reduction. LCP 0 0 0 0 0 0 0 0 0 0 0 0

CF 0 0 0 0 -

CM 0 0 0 0 -

CS 0 0 0 0 -

CR 0 0 0 0

EPS 0 0 0 0 -

MW 0 0 0 0

LPMH 0 0 0 2 20 20 19 20 20 21 19 21

is caused by the lowest value of moisture diffusivity of reference concrete which does not allow for a relatively fast release of contained moisture. Considering the wall provided with exterior plaster only, the reference concrete gives us the worst results again. There are some freezing cycles in all types of concrete, however water contained in exterior plaster applied on reference concrete gets frozen twice a reference year. This determines the service life of plaster approximately to 15 years. In rest of cases, damage caused by weather conditions is not the main cause of degradation. If we investigate concrete wall provided with thermal insulation system, namely EPS or mineral wool, due to propitious thermal insulating properties the temperature in the concrete will never drop below zero which makes freezing of water impossible. However, both materials of thermal insulation have very low value of moisture diffusivity. Therefore, the moisture cannot be transported to concrete in sufficient amount and its level remains relatively high for long time. Pull of low temperature causes then freezing of this water. Surface layer (2 mm) of exterior plaster will resist only for 2 years. But overall thickness of the plaster is 5 mm, so the service life could be doubled.

5

Conclusions

Comparing all the results obtained in this paper, from the point of view of frost resistance the best option is to use concrete modified by fly ash. Although in most of cases, concrete is not the material which will be damaged by freezing of water, its type has still relatively high influence on service life of other materials of the wall, renders in particular. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


23

Nowadays, most of concrete buildings are provided with thermal insulation. Based on results of this simulation, more advantageous is to use EPS, but the differences between EPS and mineral wool are not too high. So, the sole criterion of water freezing in the exterior render is not the decisive one for a choice of thermal insulation material.

Acknowledgement This research has been supported by the Czech Science Foundation, under grant No 103/07/0034.

References [1] ČSN 73 1322/Z1:1968 Concrete testing – Hardened concrete – Frost resistance. Czech Standardization Institute, Prague 2003. [2] VEJMELKOVÁ, E. – ČERNÝ, R.: Application of Alternative Silicate Binders in the Production of High Performance Materials Beneficial to the Environment. In: Proceedings of the 2008 World Sustainable Building Conference [CD-ROM]. Balnarring, Victoria: ASN Events Pty Ltd, 2008, pp. 520-525. [3] PERNICOVÁ, R. – PAVLÍKOVÁ, M. – PAVLÍK, Z. - ČERNÝ, R.: Vliv metakaolinu na mechanické, tepelné a vlhkostní vlastnosti vápenných omítek. In: Metakaolin 2007. Brno: VUT FAST, 2007, pp 70-77. [4] KÜNZEL, H.M.: Simultaneous Heat and Moisture Transport in Building Components, Ph.D. Thesis. IRB Verlag, Stuttgart, 1995.




25

A procedure for adaptive evaluation of numerical and experimental data J. Krok1, M. Stanuszek2 & J. Wojtas2 1

Institute of Computer Methods in Mechanics, Cracow University of Technology, Cracow, Poland 2 Institute of Computer Modeling, Cracow University of Technology, Cracow, Poland

Abstract The work addresses extended formulation of a new approach proposed to control error of experimental data. It includes: development of postprocessing techniques to approximate data given in a discrete form, a’posteriori error estimation (evaluation) of measured data and definition of reliability index of experimental data. Theoretical consideration and numerical analysis are based on the Adaptive Meshless Finite Difference (MFDM) approach. Keywords: meshless FDM, experimental data approach and smoothing, a’posteriori error estimation of experimental data, adaptive methods.

1

Introduction

Almost all numerical procedures of computational mechanics consist of the discretization process in which the continuous model is transformed into a discrete one. The discretization process is made as well within experimental setup and constitutes the key point of the computer simulation. It has a strong influence on the exactness, efficiency, generality and usefulness of obtained results. Correct discretization strategy and control of the discretization process very often decide whether the solution of analysed task is obtained or not. One may investigate the exactness of calculations (simulation) i.e.: how to measure the error due to numerical simulation (coming from the discretization process), minimize obtained error and effectively eliminate it. The question of verification applies here as well, i.e. if one can assess received results by the WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090031

26 Computational Methods and Experimental Measurements XIV theoretical consideration (using available classical solution), will be the proposed procedure of estimation safe and applicable to other difficult and complex task as for example estimation of the different errors of the experimental data. In present work the efficiency of a new, coherent concept of a'posteriori error estimation of experimental or numerical (results of FEM (Finite Element Method) or FDM (Finite Difference Method)) data, together with estimation of the mesh density, taking into consideration the equal error distribution, is considered. Approach using the discrete function and MWLS (Moving Weighted Least Square) method with constrains, defined by the differential theory equations (e.g. equations of equilibrium, boundary conditions etc.) was applied. Several ways of error estimation as well as experimental points distribution were proposed. The suggested procedures of error estimation and density prediction of experimental points distribution were tested on solution of certain mechanical problems. The introduced approach relates to the so-called, problem - oriented a'posteriori constructed estimators. The special attention was given to define an error functional with additional conditions i.e. constrains. Such an approach makes the estimation of the error of separate components possible (e.g. one component of strain or stress state or one element of body potential energy [3] may be analyzed). Mainly, error estimators base on the behaviour of the total energy, recording and responding to its change (e.g. FEM estimators of Zienkiewicz-Zhu [9]). Followed by the change of the total energy of the deformed body a part of the information on change of its components is lost. The problem of approach and error estimation of experimental and/or numerical data was commonly considered by many researchers during the last three decades. Among these the paper of Karmowski and Orkisz [1] provide first concept of coupling of analytical as well as physical information on the case within the solution of the problem. Following this idea, a concept of physically based local - global approach was introduced. Simultaneously Lukasiewicz and Stanuszek [5, 6] developed their own concept of filtering and verification of numerical as well as experimental data. They formulated an error functional, as a combination of least square approximation of error with constrains in the form of theoretical equation approach. Presented idea turned out very attractive, especially in problems of mechanics, where equations of equilibrium and continuity as well as boundary conditions have to be satisfied. The extension of the idea of physically based approach to experimental data was delivered recently by Magiera [7], where certain non-statistical considerations for a’posteriori estimation were presented. Next important stage in constructing a correct estimator of experimental data was achieved by Krok and Wojtas. They have defined [2, 4] the density distribution of experimental points as the function of error distribution. Several ideas on converting the data errors to the node density distribution were also considered there. Current paper follows and extends the approaches presented in those two works.



2

27

Formulation of the problem

2.1 General remarks During the process of collecting experimental and/or calculating numerical data several typical situations may be encountered: a) it is impossible to avoid the errors of measurement or calculation, in other words it is impossible to obtain the results of such processes without errors, b) zones with large gradients of measured and/or calculated function, indicate the domains in which high density of points is required, c) application of data smoothing (approach technology) lowers the amount of information available to assess the results of FEM and/or FDM analysis. The user is strongly encouraged to make use of all available physical, theoretical as well as numerical information to estimate the error distribution and based on this, predict the corrected node distribution. 2.2 Definition of error functional In order to begin, let us assume a “raw” data vector u~{~u1,....,~un } with components being the values measured and/or numerically calculated at discrete n points of a grid θ. Additional information on the model of the analyzed system is provided by constraints and can be presented as the following set of equations: (1) Ηu = f in which H[n×k] is the matrix resulting from the application of constraints and u is a vector of unknown, corrected data u{ u1,....,um }, usually located at m points of a grid γ differing form grid θ; vector f{f1,....,fk } represents the right side of the constraints equations.

Figure 1:

~ - set The considered domain and points: o – with measured data u θ, ∆ - with unknown data u - set γ.


28 Computational Methods and Experimental Measurements XIV When the grid of points with measured or calculated data differs from the grid of verified (unknown) data (θ≠ γ), one has to approximate the vector u{ u1,....,um } using for example the MWLS (Moving Weighted Least Square) technique: (2) u = Au where u{ u1,....,un } are the unknown values of verified data on the grid θ used for

measured and/or calculated data. A matrix A[m×k] results from the procedure (MWLS) for verified (γ) and measured/calculated (θ) points which in general are different. The constraints described by eq. (1) can be satisfied in a sense of the least squares technique. Therefore the calculated value of u has to minimize the following error function R

1 ~ ) 2 + ( H u − f )T λ (3) R(u, λ) = ( A u − u 2 where λ represents a vector of Lagrange multipliers. In other words, the vector u should satisfy the equality constraints (1) and be as close as possible to the ~ . The set of equations (3) has to be represented numerically calculated values u by a discrete model of the system in terms of finite difference operators constructed on irregular grids based on the approach (2). Based on the standard minimizing procedure with respect to u and λ vectors ∂R ~ + H T λ = 0 , ∂R = Hu − f = 0 (5) = A T Au − A T u ∂λ ∂u one can get two separated sets of linear equations leading to u and λ −1 ~ − (H(A T A) −1 H T )−1 f (6) λ = (H(A T A) −1 H T ) H(A T A) −1 A T u T −1 T ~ T −1 T (7) u = (A A) A u − (A A) H C

{(

where C = H(AT A)−1 HT and in the matrix form

)

−1

(

~ − H T H(AT A)−1 HT H(AT A)−1 AT u

)

−1

f

}

~  A T A H T  u   Au (8)  ×  =   0  λ   f   H Solution of the set of equations (8) gives the vector u for which the error function reaches the minimum value and the constraint equations (1) are exactly satisfied. Procedure described above was implemented to verify and correct data determined while calculating stress distribution in 2D disc loaded by the concentrated force.

3

Numerical implementation

In order to demonstrate the functionality of the proposed verification procedure, a number of trial calculations have been made. With reference to works [5, 6] the examples of verification of the plane stress state in a half-plane, loaded by a concentrated force P at the boundary is presented. Based on [8] theoretical stress distribution (σxx and σyy) inside the half-plane loaded along the edge can be presented in the form WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


29

2

3

xy ; σ yy = 2 P ; (9) 2 2 2 2 2 2 2 π ( x + y )( x + y ) (x + y ) Force P was assumed to be equal to 5[kN/mb]. The size of the domain (fig.1) was depicted in [m] and the stress distribution in [kPa]. Calculations were performed on irregular grid of nodes with equilibrium and continuity equations in the forms: 2 ∂ 2σ xx ∂ σ yy − = 0 ; ∆ ( σ xx + σ yy ) = 0 (10) ∂x 2 ∂y 2 The relative error implemented in the tests has taken the form: x

σ xx = 2 P π

εR =

σ exact − σ ⋅ 100% σ exact

(11)

Irregular distribution of nodes was presented on fig.2. Figure 3 shows the exactly calculated (9) σxx stress component. 2.50 2.00

P

1.50 1.00 0.50 0.00 -0.50 -1.00 -1.50 -2.00 -2.50 1.00

Figure 2:

2.00

3.00

4.00

5.00

6.00

Discretization of the domain.

2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 3:

2.00

3.00

4.00

σxx stress distribution. Input data - exact values.


5.00

6.00

30 Computational Methods and Experimental Measurements XIV To find out the effectiveness of the proposed filtering procedure the above exact solution was artificially disturbed by the local errors reaching up to 50% of initial values of stresses. Stresses σyy remained unchanged. Figure 4 shows a view of disturbed stress state and fig. 5 depicts the measure of relative error of corrupted data.

2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 4:

2.00

3.00

4.00

5.00

6.00

σxx stress distribution with random errors - corrupted data.

2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 5:

2.00

3.00

4.00

5.00

6.00

Relative error of σxx – corrupted data.

Prepared corrupted data were subjected to filtering process with two types of constrained equations applied: a) constrained equations (10) were imposed only on the internal nodes of the domain; b) in addition to the previous case the boundary conditions of I order (assumed values of function on the edge of the domain) were introduced. The results of filtration in first case were presented in fig. 6 and 7 below. One can observe high efficiency of the procedure for the internal nodes and low for the boundary ones (fig.7). This is the effect of FDM approach of constrained equations only on internal nodes of the domain. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


31

2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 6:

2.00

3.00

4.00

5.00

6.00

Stress σxx distribution after filtering. Constrains (10) imposed only on internal nodes of the domain.

2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 7:

2.00

3.00

4.00

5.00

6.00

Relative (11) errors of σxx after filtering of corrupted data. Constrains (10) imposed only on internal nodes of the domain.

Very often certain boundary conditions in the analysed domains are explicitly known and should be imposed within the solution obtained from the filtering process. The result of such a procedure is presented on fig. 8, where theoretically calculated values of the approximated function were applied at the boundary points of analysed domain. It is worthy to point out that the maximum local relative error of σxx was reduced from 50% to less than 8% (fig.9). To estimate the quality of the numerical solution one can calculate the distribution of so called local approach error defined by the equation: σ approx − σ (12) εL = σ approx The distribution of such an error was presented in fig. 10 and 11. It is easy to notice, that estimation of the required density of node distribution based on such an approach is destined to fail. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 8:

2.00

3.00

4.00

5.00

6.00

Stress distribution σxx after filtering of corrupted data. Constrains (10) with boundary condition imposed. 2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 9:

2.00

3.00

4.00

5.00

6.00

Relative (11) errors of σxx. after filtering of corrupted data. Constrains (10) with boundary condition imposed. 2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 10:

2.00

3.00

4.00

5.00

6.00

Relative (12) errors of σxx. Constrains (10) imposed on internal nodes of the domain.



33

2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 11:

2.00

3.00

4.00

5.00

6.00

Relative (12) errors of σxx. Constrains (10) with boundary condition imposed.

For further consideration the parameter called “efficiency index” will be introduced and defined by the equation: e ψ = approx (13) e exact i

This parameter equals Ψ = 0.41 when only constraints (10) are imposed, while when boundary conditions are satisfied Ψ = 0.84. Efficiency index takes here a role of a measure of approach quality. The perfect situation occurs when the estimated and exact solutions are identical e.g. Ψ = 1.00. High value of the efficiency index in second case is explained by the fig. 12 where one can note a similar distribution of absolute errors of corrupted and estimated σxx .

Figure 12:

Errors σ approx − σ and σ exact − σ

of σxx . Second case.


34 Computational Methods and Experimental Measurements XIV Now we are going to provide a way for expressing the local error in terms of the nodal densities for which one can obtain even error distribution. This is the essential task underlying adaptive technologies. Based on our considerations it turns out that to define such a nodal distribution it is required to apply local as well as global concentration index. Lets introduce: Local index of nodal density correction defined as: σ iexact − σ iapprox (14) ξσLi = e Average error in the analysed domain may be calculated as (n number of nodes): n 1 (σ iexact − σ iapprox )2 eσ= (15) ∑ n i =1 Total weighted norm of stresses is depicted by: 2

 n exact  (16)  ∑σ i  σ n  i =1  Finally, global index of nodal density correction, grid refinement, defined as: eσ ξσ = (17) η⋅ U σ U

1

=

where η corresponds to the level of imposed error distribution. Based on the norms listed above, global – local index of increasing grid density (I type) may be defined using square of the global index as: 2

 e σ  σ iexact − σ iapprox e  ⋅ = 2 σ 2 σ iexact − σ iapprox (18) η ⋅ U  e U σ η ⋅ σ  σ  Other possible global – local index of increasing grid density (II type) may be defined using linear form of global refinement index as  e σ  σ iexact − σ iapprox 1 ⋅ (19) ξ σLi = ξ σ ⋅ ξ σLi =  = σ iexact − σ iapprox η ⋅ U  e U η ⋅ σ  σ σ 

ξσLi = ξσ ⋅ ξσLi =  2

2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 13:

2.00

3.00

4.00

5.00

6.00

Local-global mesh refinement index (18). Permissible error η=0.10.



35

The distribution of the global – local refinement index (19) is depicted in fig.14. 2.00

1.00

0.00

-1.00

-2.00 1.00

Figure 14:

2.00

3.00

Local-global mesh refinement index (19). η=0.10.

4.00

5.00

6.00

Permissible error

It is easy to observe on the results presented above that the refinement of the grid was necessary within the areas exhibiting high gradient of the sought function. Moreover it is worthy to note that in I-st case, the required grid density was overestimated (20) while in II-nd attempt the refinement index takes realistic value of 4 (verified during analysis).

4

Conclusions

In this work the development of an adaptive approach for the experimental as well as numerical data collections was presented. An adaptive procedure of experimental and numerical data collections based on a’posteriori error analysis of data was proposed. It includes estimation of the new grid density taking into account equal distribution of an error (using different error norms). Thus adaptive procedure of experiment or numerical discretization planning is possible. Numerical approach was executed using Adaptive Meshless Finite Difference Method with certain additional constrains taken into consideration. The paper also presents a new concept of error estimation with the use of so called global – local estimator. The implementation of such estimators lets to refine the grid of numerical or experimental nodes. Obtained results show a good efficiency of proposed adaptive procedure.

References [1] W. Karmowski, J. Orkisz, Physically Based Method of Enhancement of Experimental Data - Concept, Formulation, and Application to Identification of Residual Stresses, Proc. of the IUTAM Symp. on Inverse Problems in Engng Mech., Tokyo, Japan, Springer-Verlag, 1993, 61-70. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

36 Computational Methods and Experimental Measurements XIV [2] J. Krok, An Extended Approach to Error Control in Experimental and Numerical Data Smoothing and Evaluation Using the Meshless FDM, Revue Europenne des elements finis, no 7-8/2002, pages 913-935. [3] J. Krok, Meshless FDM based Approach to Error Control and Evaluation of Experimental or Numerical Data. II MIT Conf. on Comp. Fluid and Solid Mechanics, 2003, Cambridge, MA, USA. [4] J. Krok, J. Wojtas, An Adaptive Approach to Experimental Data Collection Based on A Posteriori Error Estimation of Data, Comp. Meth. in Mechanics – CMM-2007, June 2007, Spała-Łódź, pp.1-13. [5] S. A. Łukasiewicz, M. Stanuszek, J.A. Czyż, Filtering of the Experimental or FEM Data in Plane Stress and Strain Fields, Experimental Mechanics, 1993, June, pp. 139-147. [6] S. A. Łukaszewicz, M. Stanuszek, Constrained, weighted, least square technique for correcting experimental data, 6th Int. Conf. on Comp. Methods and Experimental Measurements 93, Vol 2: Stress analysis, pp. 467-480, Elsevier Applied science, London New York, 1993. [7] J. Magiera, Non-statistical physically reasonable technique for a posteriori estimation of experimental data error, Computer Assisted Mechanics and Engineering Sciences, 13, 593-611, 2006 [8] S. Timoshenko, J. N. Goodier Theory of elasticity, New York Toronto London, 1951. [9] O.C. Zienkiewicz, R.L. Taylor, The Finite The Finite Element Method, Vols. I-III, Sixth ed. Butterworth-Heinemann, Oxford, 2005.



37

Gear predictor of manual transmission vehicles based on artificial neural network A. M. Wefky, F. Espinosa, M. Mazo, J. A. Jiménez, E. Santiso, A. Gardel & D. Pérez Department of Electronics, University of Alcala, Spain

Abstract Nearly all mechanical systems involve rotating machinery (i.e., a motor or a generator), with gearboxes used to transmit power or/and change speed. Concerning vehicles, there is a specific nonlinear relationship between the size of the tires, linear velocity, engine RPM, gear ratio of the differential, and the gear ratio of the transmission. However, for each car there is a specific range of gear ratio of the transmission for each gear. On the other hand, the gear value is an indication of the driver behaviour and the road conditions, therefore it should be considered to establish non-pollutant driving guidelines. In this paper, two novel feed-forward artificial neural network (ANN) models have been developed and tested with the gear as the network output and the velocity of the engine (RPM) and the velocity of the car in (Km/h) as the network inputs. A lot of experiments were made using two commercial cars. The prediction efficiency of the proposed models is superior (i.e., the generalization mean square error is about 0.005). However after testing with two different vehicles, the conclusion is that on one hand the structure of the ANN model is suitable. On the other hand each vehicle has its specific model parameters. This paper shows that it is difficult to develop a universal model that predicts the gear based on the RPM and speed of any car. Keywords: feed-forward artificial neural networks, gear predictor, manual transmission.

1

Introduction

0B

The drivetrain system of the automobile engine consists of the following parts: engine, transmission, drive shaft, differential, and driven wheels. Firstly, the WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090041

38 Computational Methods and Experimental Measurements XIV transmission is a gear system that adjusts the ratio of engine speed or engine regime (RPM) to the vehicle speed. Mainly, it enables the engine to operate within its optimal performance range regardless of the vehicle speed. In a manual transmission, the driver selects the correct gear ratio from a set of possible gear ratios (usually five of six for modern passenger cars). For each gear, there is a specific gear ratio. But an automatic transmission selects this gear ratio by means of an automatic control system. Secondly, the drive shaft is used on front-engine rear wheel drive vehicles to couple the transmission output shaft to the differential input shaft. However, in front wheel drive automobiles, a pair of drive shafts couples the transmission to the drive wheels through flexible joints known as constant velocity (CV) joints. Thirdly, the differential has the following three purposes. The first is the right angle transfer of the rotary motion of the drive shaft to the wheels. The second is to allow each driven wheel to turn at a different speed, because the external wheel must turn faster than the internal wheel when the vehicle is turning a corner. The third is the torque increase provided by the gear ratio. The gear ratio also affects fuel economy. In front wheel drive cars, the transmission, differential, and drive shafts are known collectively as the transaxle assembly. The combination of drive shaft and differential completes the transfer of power from the engine to the rear wheels [1]. Finally, the car’s tires can almost be thought of as a third type of gearing. In other words, if the circumference of the tires is L, then for every complete revolution of the wheel, the car travels L meters. Eqn. (1) shows the formula relating the overall gear ratio (i.e., gear ratios of the transmission (grt) and differential (grd)), the size of the tires (Ct), the speed of the car (vc), and the engine speed (ve). The overall gear ratio gro is the product of grd and grt [2].

gro = grd * grt = Ct *

ve vc

(1)

One of the challenges of the MIVECO research project, in which the authors are involved, is to establish a relationship between the driver behaviour and the road conditions with the non-pollutant driving guidelines. Consequently, the knowledge of the gear value and its relationship with the gases measurement is required. However, in most engine control systems it’s difficult to find a sensor for the transmission gear selector position.

RPM

Artificial Neural Network

Vehicle’s Gear

Velocity

Figure 1:

Procedure used to test and train the ANN.

This paper proposes a novel artificial neural network (ANN) model to predict the overall gear ratio gro based on the engine RPM and the corresponding speed WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


39

of the car, see Figure 1. The model has been trained, validated, and tested with experimental tests using two commercial vehicles: Peugeot 205 and 405. ANNs have been used widely in recent years in various fields such as finance [3], medicine [4], industry [5] and engineering [6, 7], due to their computational speed, their ability to handle complex non-linear functions and their robustness and great efficiency, even in cases where full information for the studied problems is absent.

2

Methodology

1B

Figure 2 shows the process followed to evaluate the capability of a feedforward artificial neural network to predict the gear based on the corresponding velocity of the car and RPM of the engine. That procedure was applied to two different vehicles, as explained in the following sections.

Measure the engine RPM and the car speed

Deduce the corresponding gear

Normalize and divide the entire data set into training , validation, and testing data sets .

Train the neural network model using the training data set

Test the neural network model with the data of the same car

Test the neural network model with the data of the other car Figure 2:

Procedure used to test and train the ANN.

2.1 First case of study: Peugeot 205 The engine RPM and the car speed, shown in Figure 3, as well as the overall gear ratio gro were measured in different zones with different driving conditions. The instantaneous values of the gear were calculated where there is a specific WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

40 Computational Methods and Experimental Measurements XIV range of the overall gear ratio for each gear as illustrated in Table 1. The overall gear ratio, the corresponding gear signals, and the filtered gear signal are shown in Figure 4. The horizontal dashed lines in the graph of the overall gear ratio represent the boundaries of the ranges indicated in Table 1. The overall gear ratio and therefore the gear signal oscillates back and fourth around some boundaries in some areas marked by ellipses. Obviously, these oscillations in the gear signal happen in a very short time period which is impossible in reality. Consequently the gear signal was filtered in order to get rid of these repetitive changes. Table 1:

Ranges of gro with the corresponding gear.

Range of the overall gear ratio (gro) From To 20 -11,87628 20 6,74544 11,87628 4,87968 6,74544 3,83916 4,87968 3,12156 3,83916

Figure 3:

Gear Neutral First Second Third Fourth Fifth

Engine RPM and car velocity of the PEUGEOT 205.

2.2 Second case of study: Peugeot 406 6b

In INSIA laboratories in Madrid, the engine RPM was measured during a test of the New European Driving Cycle (NEDC) on rolling roads. The NEDC consists WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


41

of four repeated ECE-15 driving cycles and an Extra-Urban driving cycle (EUDC). The EUDC and only one of the four ECE-15 driving cycles were used. Concerning the speed of the vehicle, the standard values were used. The gear signal was deduced using the clutch signal and the RPM signal. In other words, the gear change event is always synchronized with activating the clutch signal. On the other hand, if the gear changes from a low to a higher position, the RPM level is suddenly decreased. On the contrary, when the gear changes from a high to a lower position, the RPM level decreases smoothly. The instantaneous values of the RPM, reference velocity, clutch, and the deduced gear signals are plotted in Figure 5.

Figure 4:

Overall gear ratio, gear, and filtered gear.

2.3 Artificial neural network modelling Numerous neural networks are available for function approximation problems. A multilayer Perceptron MLP feedforward neural network trained with backpropagation was chosen to analyze the data because it has many properties useful for the vehicle gear prediction problem. It can efficiently learn large data sets. To obtain a good generalization, the entire data set was divided into training (60%), validation (20%), and testing (20%) groups. MLPs are more powerful than single layer networks because single layer networks are only able to solve linearly separable classification problems [11]. For example, a single-hidden layer feedforward network with a sufficiently large number of neurons can satisfy the "universal approximation" property [8, 12, 13, and 14]. A singlehidden layer neural network (1-S-1), with S sigmoid neurons in the hidden layer and linear neurons in the output layer, can produce a response that is a superposition of S sigmoid functions [11]. Moreover, a single-hidden layer feedforward network with any bounded nonlinear transfer function with N-1 WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

42 Computational Methods and Experimental Measurements XIV hidden neurons can represent any N input-target relations exactly (with zero error) [15, 16, 17, and 18]. However, a double-hidden layers network can represent any N input-target relations with a negligible small error using only (N/2) + 3 hidden neurons. This means that a network with 2 hidden layers is better than a network with one hidden layer in terms of number of training samples [15]. Consequently, in this paper both single and double hidden layer networks were used with sigmoid hidden neurons and linear output neurons.

Figure 5:

RPM, reference velocity, clutch, and the gear.

When a particular training algorithm fails on a MLP; it could be due to one of two reasons. The learning rule fails to converge to the proper values of the network parameters, perhaps due to unsuitable network initialization. Or the inability of the given network to implement the desired function, perhaps due to insufficient number of hidden neurons. To avoid the first possibility, the neural network models were trained and tested 10 times and the network with the lowest mean square error was chosen. Concerning the second possibility, there is no theory yet to tell you how many hidden neurons are needed to approximate any given function. In most situations, there is no way to determine the best number of hidden neurons without training several networks and estimating the generalization error of each. However, the designer must put into consideration that the MLP with the minimum size is less likely to learn noise during the training phase; consequently generalizes better to unseen data. The methods to achieve this design objective are: network growing and network pruning. In network growing, we start with a small MLP, and then add a new hidden neuron or new hidden layer when we are unable to meet the design specifications. On the other hand, in network pruning, we start with a large MLP, and then prune it by eliminating certain weights in an orderly manner. If there were too few hidden neurons, high training error and high generalization error would result due to underfitting and high statistical bias. On the other hand, if there were too many hidden neurons, low training WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


43

error, but still have high generalization error, would result due to overfitting and high variance [7, 10, 11, and 13].

3

Results

The mean square errors resulting from testing single and double feedforward neural networks with the data of the same vehicle are shown in Tables 2 and 3. The minimum size model architecture that met the design goal (about 0.005) for the PEUGEOT 205 and PEUGEOT 406 was 2-10-1 with 10 hidden neurons. Table 2: Number Neurons 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

MSE resulting from testing single hidden layer networks.

of

Table 3: Number neurons S1 1 2 3 4 5 6 7 8 9 10

Hidden

Testing Mean Square Error (MSE) PEUGEOT 205 PEUGEOT 406 0.022929804113799 0.030737847949962 0.008113754649602 0.013009639928021 0.007150878530542 0.007953683727784 0.006916124765493 0.007533362116994 0.006770174929462 0.007356262805218 0.006801600634453 0.006821096479489 0.007161435831671 0.006249000207199 0.007262385239038 0.005906782786126 0.006665429326786 0.005840597665814 0.006486881856067 0.005527620454196 0.006699752798641 0.004349509945806 0.006961214546624 0.005195854106452 0.006687424328677 0.004042310392445 0.006733573608595 0.005189990010043 0.006861530582299 0.003762549738464

MSE resulting from testing double hidden layer networks. of S2 1 2 3 4 5 6 7 8 9 10

hidden

Testing Mean Square Error (MSE) PEUGEOT 205 0.021251738513978 0.007897066497980 0.007403216890455 0.007136833232957 0.007266126627429 0.006853332450965 0.006382907932383 0.006284173922234 0.006515939055022 0.006873106767166

PEUGEOT 406 0.029045230801033 0.007546159875325 0.007741855207261 0.003671035573348 0.002963149051319 0.002682340983058 0.002440590330384 0.002303214429815 0.002130067441743 0.002144081137843



Figure 6:

Figure 7:

Testing results of the proposed model for P-205 9B.

Testing results of the proposed model for P-406.



45

The proposed models showed satisfactory results when it was tested with the data of the same vehicle as can be shown in Figures 6 and 7. On the other hand, the models failed to predict the gear signal when they were tested with the data of the other vehicle.

4

Discussion

3B

The neural network models succeeded to predict the gear given only the corresponding engine RPM and car speed. Two structures models have been studied. Considering the value of 0.005 for the mean square error as a goal, it can be deduce that the simplest solution is the following structure model: only one hidden layer and 10 neurons per layer. On the other hand the same neural network model failed to predict the gear signal when it was tested with the data of the other vehicle. This means that the parameters (weights and biases) of the model should be calculated for each vehicle.

5

Conclusion

An approach to predict the vehicle’s gear based on the engine regime (RPM) and the vehicle’s velocity (Km/h) using feedforward neural networks is presented. Two neural network models were evaluated for two different vehicles. The proposed ANN model structure is: only one hidden layer and 11 neurons [10 hidden plus 1 out]. This model allows one an acceptable mean square error (about 0.005) to predict the gear of manual transmission vehicles. However, the calculated model parameters for a car cannot be extended to another vehicle; they should be checked for any different case.

References [1] Norman P. M., Gerald L., Charles W. Battle, Edward C.J, Understanding Automotive Electronics, 2003, Elsevier Science, USA. [2] http://en.wikipedia.org/wiki/Main [3] Y. Bodyanskiy, S. Popov, Neural network approach to forecasting of quasiperiodic financial time series, European Journal of Operational Research, 175(3) (2006) 1357-1366. [4] M. Frize, C.M. Ennett, M. Stevenson, and H.C.E. Trigg, Clinical decision support systems for intensive care units: using artificial neural networks, Medical Engineering & Physics 23(3) (2001) 217-225. [5] M. Sloleimani-Mohseni, B. Thomas, Per Fahlen, Estimation of operative temperature in buildings using artificial neural networks, Journal of Energy and Buildings, 38 (2006) 635-640. [6] Y.J. Chen, Y.M. Chen, C.B. Wang, H.C. Chu, T.N. Tsai, Developing a multi-layer reference design retrieval technology for knowledge management in engineering design, Expert Systems with Applications, 29(4) (2005) 839-866. HU

UH


46 Computational Methods and Experimental Measurements XIV [7] S. Haykin, Neural Networks: a comprehensive foundation. New York:

[8] [9]

[10]

MacMillan College Publishing Company 1994. O. Nolles, Nonlinear system identification: from classical approaches to neural networks and fuzzy models, Spring-Verlag, Berlin, 2001. Mukesh Khare, S.M. Shiva Nagendra, Artificial Neural Networks in Vehicular Pollution Modelling, Springer 2007 T. I. Maris, L. Ekonomou, G.P. Fotis, A. Nakulas, E. Zoulias, Electromagnetic field identification using artificial neural networks, Proceedings of the 8th WSEAS International Conference on Neural Networks, Vancouver, British Columbia, Canada, June 19-21, 2007. Khalaf, A.A.M.; Abo-Eldahab, M.A.M.; Ali, M.M., System Modelling Using Neural Networks in the Presence of Noise, Electronics, Circuits and Systems, 2003. ICECS 2003. Proceedings of the 2003 10th IEEE International Conference on Volume 2, 14-17 Dec. 2003 Page(s):467 - 470 Vol.2. Martin T. Hagan, Howard B. Demuth, Neural Network Design, 1996 by PWS Publishing Company. Bishop, C.M. (1995), Neural Networks for Pattern Recognition, Oxford: Oxford University Press. Sarle, W., 1997. Neural network frequently asked questions. ftp://ftp.sas.com/pub/neural/ FAQ.html. Guang-Bin Huang; Lei Chen; Chee-Kheong Sien, Universal approximation using incremental constructive feedforward networks with random hidden nodes, Neural Networks, IEEE Transactions on Volume 17, Issue 4 , July 2006 Page(s):879 – 892. S.I. Tamura, M. Tateishi, Capabilities of a four-layered feedforward neural network: four layers versus three”, IEEE Transactions on Neural Nets, 8(2) (1997) 251-255. Sartori, M.A.; Antsaklis, P.J.; A simple method to derive bounds on the size and to train multilayer neural networks, Neural Networks, IEEE Transactions on Volume 2, Issue 4 , July 1991 Page(s):467 – 471. Huang, S.-C.; Huang, Y.-F.; Bounds on number of hidden neurons of multilayer perceptrons in classification and recognition; Circuits and Systems, 1990., IEEE International Symposium on 1-3 May 1990 Page(s):2500 - 2503 vol.4. Guang-Bin Huang; Babri, H.A.; Upper bounds on the number of hidden neurons in feedforward networks with arbitrary bounded nonlinear activation functions, Neural Networks, IEEE Transactions on Volume 9, Issue 1 , Jan. 1998 Page(s):224 – 229. H

H

[11] [12] [13]

HU

[14]

U

H

[15] [16]

H

H

H

H

H

[17]

H

H

H

H

[18]

H

H

H


H


47

Non-thermal, chemical destruction of PCB from Sydney tar ponds soil extract A. J. Britten & S. MacKenzie Cape Breton University, Sydney, Nova Scotia, Canada

Abstract The Sydney Tar Ponds, in Sydney, Nova Scotia, Canada, contains more than 700,000 tonnes of contaminated sediments including PAH, hydrocarbon compounds, coal tar, PCB, coal dust, and municipal sewage. An important source of contamination are the PCB which cause adverse health affects to humans as well as environmental problems for the surrounding ecosystems through bioaccumulation and resistance to environmental breakdown. There are various processes for the remediation of contaminated sites. The most commonly used methods include incineration, solvent washing and/or extraction, stabilization/solidification and base catalyzed soil remediation. A recent and more environmentally friendly method for remediation is the “SonoprocessTM.” The claim is that PCB are destroyed in a non-thermal way using a sodium reaction and high frequency vibration to remove the chlorine atoms from the biphenyl. In this study, the process is modified to suit the Tar Ponds matrix and is tested on samples of PCB and PAH contaminated soil from the Tar Ponds. A steel bar (with a chamber containing the contaminated soil, sodium, and solvent attached to the end) is brought to its resonance frequency to destroy harmful contaminants. The energy which is generated is used to vibrate the PCB extract with sodium to break the C-Cl bonds. The soil mixture is removed and washed, resulting in clean, safe soil and sodium chloride byproduct. The remaining solution from the extraction has a possibility of being used as a low-grade fuel. GC-ECD and GC-MS were used to identify and to quantify the compounds present before and after the PCB destruction process. PCB present at 160mg/kg in soil were reduced to 0, we look for Wˆβ that solves 1 W 1 2 min hΩT (W ) + β (9) W 2(nI + 1)nH In this minimization problem weights are initialized to the value obtained at the end of first training phase. We denote by βˆ the value of β for which hΩG (Wˆβ ) is the smallest. 2.2.3 Final training phase We perform finally a new training phase on the set ΩT ∪ΩG using the regularization parameter βˆ provided by the previous step and Wˆβ as initial weights. 1 1 2 ˆ W min hΩT ∪ΩG (W ) + β (10) W 2(nI + 1)nH 2.3 Levenberg-Marquardt algorithm Levenberg-Marquardt algorithm is a combination of steepest descent method and Gauss-Newton algorithm [8–10]. The iterative descent algorithms consist in defining a descent direction d, and the new point W+ is obtained from the current point W using the following update rule W+ = W + d The descent direction, d is given by [10] J(W )T J(W ) + αI d = −J(W )T r(W )

(11)

(12)

where α > 0 and J(W ) = ∇r(W )T is the Jacobian matrix. The LevenbergMarquardt algorithm is then the following: • Choose an initial point W0 and a real number α0 > 0, k = 0 WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


141

• Compute dk , solution to J(Wk )T J(Wk ) + αk I dk = −J(Wk )T r(Wk ) • If hΩ (Wk + dk ) < hΩ (Wk ), set Wk+1 = Wk + dk , choose αk+1 < αk , increase k and goto previous step; otherwise, decrease αk and goto previous step. Stop if hΩ (Wk + dk ) < η 2.3.1 Memory reduction and adjoint computation As seen in eqn. (12), the Levenberg-Marquardt algorithm usually requires the computation of the inverse of J(W )T J(W ) + αI, whose size may be quite large in some cases. For memory reduction, at least in terms of storage, the linear system in eqn. (12) can be solved using the conjugate gradient method, which requires only matrix-vector products. We only need to compute the left-hand side of eqn.(12) in an efficient way. This can be done in two steps. 1. We first compute z = J(W )d. This quantity can be rewritten as follows ∂r(W + εd) r(W + εd) − r(W ) = (13) J(W )d = lim ε→0 ε ∂ε ε=0 and J(W )d corresponds to the differentiation of a vector-valued function r with respect to a single parameter ε. This can be done very efficiently using the forward mode of the algorithmic differentiation. 2. Then, we have to compute J(W )T z, which can be rewritten n nP P T ∇ri (W )zi = ∇ ri (W )zi = ∇ r(W )T z (14) J(W ) z = i=1

i=1

In this form, J(W )T z corresponds to the differentiation of a scalar function with respect to several parameters and the reverse mode of the algorithmic differentiation is particularly efficient in this case [11–13]. The computation of the right-hand side of (12) is realized in the same manner as in the second step of the computation of the left-hand side.

3 Experimental setup 3.1 Description of the test cells Two test cells, located at Limoux, France, with a roof surface area of 35 m2 , outside dimensions of 4×7 m2 on floor level and height of 3 m were used for the test. Each cell is built without windows and there is no ventilation. The roof with inclination of 36◦ is made up of rafter of 8 × 11 cm with spacing of 48 cm between adjacent rafters and has clay tiles. The roof ridge has north-south orientation. The floor is made up of wood paving and the under floor gap is over insulated with 40 cm of mineral wool. The access to each test volume is by an airlock in the gable wall and thus the thermal exchange takes place through walls and roof alone. Inside temperature of each cell is maintained at 23◦ C using two fan heater of 1 KW output. The airlock is heated to 1◦ C less than the main cell and acts as a guard cell. One test WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

142 Computational Methods and Experimental Measurements XIV cell has TS9 multi-foil insulation of ACTIS [14] and other has 20 cm of mineral wool insulation. The layout of each cell is in accordance with the manufacturer’s instructions [14]. Infra red pictures made outside and inside of the test cells do not reveal any significant differences between each kind of cells. 3.2 Instrumentation and measurements Each cell is equipped with two temperature sensors located 1.5 m above the floor and placed in an open gray PVC tube to shield them from air movements. Energy consumption in each test cell is measured by recording current and voltage using calibrated instruments. Weather parameters namely, the outside temperature, relative humidity, wind direction, wind speed and total solar radiation were recorded per minute by a dedicated weather station at the site. The period of measurement was from 1 December 2005 to 28 February 2006. Measurements are carried out per minute and are recorded by dedicated data logger units. All quality control checks pertaining to instrumentations were made. [14]

4 Simulations using GAP 4.1 Training of neural network

Consumption (KWh)

The consumption in a test cell is a function of meteorological parameters, namely, the difference between outside temperature and temperature inside the chalet, wind speed, wind direction, relative humidity and global solar radiation. Using GAP, two neural networks pertaining to the data of two test cells are developed. Total 1513 observations spanned over the period 1/12/2005 to 31/12/2005 are used to train the network. These networks are validated by predicting the consumption values for the period of 1/1/2006 to 28/02/2006. Fig. 1, 2 shows the comparison between measured and simulated energy consumptions corresponding to these experiments. Table 1 shows the measured and simulated energy consumption in these test cells. The difference between measured and simulated consumption is less than 1%. It is observed that the energy consumption in a test cell with TS9 material is 4% less than the energy consumption for a test cell with mineral wool.

Simulated

0.2 0.15 0.1 0.05

0

500

Hours→ 1000

Measured

1500

Figure 1: Comparison of measured and simulated energy consumption in a test cell with TS9 insulation, located at Limoux, France. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

Consumption (KWh)


Simulated

0.2 0.15 0.1 0.05

Measured

1500

Hours→ 1000

500

0

143

Figure 2: Comparison of measured and simulated energy consumption in a test cell with mineral wool insulation, located at Limoux, France.

Table 1: Measured and simulated net energy consumption in test cells located at Limoux, France. Energy consumption in kWh Cell with TS9 Measured Simulated

Consumption (KWh)

137

136

Cell with mineral wool Measured Simulated 143

142

Simulated

0.2 0.15 0.1 0.05 0

500

Hours→ 1000

Measured

1500

Figure 3: Comparison of measured and simulated energy consumption in a test cell with multi-foil insulation, located at TRADA, U.K.

4.2 Prediction using neural network These trained and validated neural networks are used to predict the energy consumption of similar test cells with multi-foil insulation similar to TS9 and mineral wool insulation located at TRADA in United Kingdom [15]. The weather data and the consumption were measured for the period 1/1/2006 to 28/2/2006. Fig. 3, 4 shows the comparison of measured and simulated consumption. Table 2 provides the values of measured and simulated consumption. For the cell with multi-foil insulation the simulated consumption differs by 4% where as for the test cell with mineral wool insulation the difference is 3%. As part of the long term in situ data collection strategy, ACTIS planned to set up new test cells in United Kingdom. The weather data spanned over the WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

Consumption (KWh)


Simulated

0.2 0.15 0.1 0.05

0

500

Measured

1500

Hours→ 1000

Figure 4: Comparison of measured and simulated energy consumption in a test cell with mineral wool insulation, located at TRADA, U.K.

Table 2: Measured and simulated net energy consumption in test cells at TRADA, United Kingdom. Energy consumption in kWh Cell with multi-foil insulation

Cell with mineral wool

Measured

Simulated

Measured

Simulated

165

159

138

134

period of 1/2/2006 to 28/02/2006 at eight different location in United Kingdom, namely, Manchester, Norwich, London, Plymouth, Cardiff, Aberdeen, Newcastle and Belfast is available. The weather data characteristics at these locations are different than those at Limoux, France. Fig. 5 shows the range of measured values for outside temperature, wind direction, wind speed and global solar radiation at Limoux, France and at these location in United Kingdom. The minimum temperature at these locations is higher than Limoux whereas global solar radiation values are small compared to Limoux. In order to estimate the energy consumption in test cells that would be built in near future at these sites, the neural network model trained using weather data and consumption details in Limoux, France is used to simulate the in situ energy consumption in test cells with TS9 and mineral wool insulation materials. Table 3 shows predicted energy consumption for the test cell with TS9 and mineral wool insulation to be built at each of the eight sites. The difference in energy consumption for the test cell with TS9 and mineral wool insulations is also shown. The predicted consumption in test cell with multi-foil insulation is less compared to test cell with mineral wool insulation at Plymouth, Cardiff, Aberdeen, Newcastle and Belfast whereas it is more by 2% at Manchester and by 1% at Norwich and London. The standard deviation of predicted consumption values for test cell with mineral wool is 9.7 whereas the corresponding values for the test cell with multifoil insulation is 6.5, possibly indicating that multifoil insulation is more robust to varying weather conditions as compared to mineral wool insulation. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

Solar radiation (W/m2) Wind direction (deg) Temperature (oC) Wind speed (m/s)


145

30 20 10 0

Limoux Manchester Norwich

London Plymouth

Cardiff

Aberdeen Newcastle Belfast

Limoux Manchester Norwich

London Plymouth

Cardiff

Aberdeen Newcastle Belfast

400

300

200 100 0

17 15

10

5 0

Newcastle Belfast LimouxManchester Norwich London Plymouth Cardiff Aberdeen

800 600 400 200 0

Newcastle Belfast Norwich London Plymouth Cardiff Aberdeen LimouxManchester

Figure 5: Comparison of weather data at eight different locations in U.K. and of Limoux, France.

Table 3: Simulated consumptions of test cells with ACTIS TS9 and mineral wool insulation at eight different locations in United Kingdom. Location

Energy consumption (kWh)

Difference

Mineral wool

ACTIS TS9

(A)

(B)

(B-A)/B*100

Manchester

135

138

2%

Norwich

136

137

1%

London

129

130

1%

Plymouth

155

128

−17%

Cardiff

132

126

−5%

Aberdeen

154

143

−7%

Newcastle

144

143

−1%

Belfast

140

136

−3%

5 Conclusion GAP is a result of a powerful combination of several techniques such as the use of a zero memory minimization method, specific activation function that guarantees WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

146 Computational Methods and Experimental Measurements XIV the minimal change of weights, the use of a Tikhonov regularization technique in order to build smooth and stretched basis functions. GAP is used to generate neural network model to predict energy consumption of test cells as a function of meteorological parameters using in situ data. It is demonstrated with examples that properly trained networks can accurately predict the energy consumption of houses located at some other locations also.

References [1] ISO 8302, Thermal insulation. Determination of steady state thermal resistance and related properties-Gaurded hot plate apparatus. [2] ISO 8990:1994, Thermal insulation. Determination of steady state thermal transmission properties-Calibrated and guarded hot box. [3] ISO 6946, Building components and building elements-Thermal resistance and thermal transmittance-Calculation method. [4] Doran, S., Field investigation of thermal performance of construction elements as built. Technical Report Building Research Establishment Ltd. 78132, November, 2001. [5] Cybenko, G., Continuous valued neural networks with two hidden layers are sufficient. Technical report, Department of Computer Science,Tufts University, Mdeford, Massachusetts, 1988. [6] Cybenko, G., Approximation by superpositions of sigmoidal function. Mathematics of control, signals and systems, 2, pp. 303–314, 1989. [7] K. Hornik, M.S. & White, H., Multilayer feedforward networks are universal approximators. Neural networks, 2, pp. 359–366, 1989. [8] Levenberg, K., A method for the solution of certain problems in least squares. Quarterly of Applied Mathematics, 2, pp. 164–168, 1944. [9] Marquardt, D., An algorithm for least-squares estimation of nonlinear parameters. SIAM Journal on Applied Mathematics, 11, pp. 431–441, 1963. [10] Nocedal, J. & Wright, S., Numerical Optimization. Springer Series in Operations Research and Financial Engineering, Springer, 2nd edition, 2006. [11] J. Gilbert, G.V. & Masse, J., La differentiation automatique de fonctions representes par des programmes. Technical Report 1557, Technical report,INRIA, 1991. [12] Griewank, A., Evaluating derivatives: principles and techniques of algorithmic differentiation. SIAM: Philadelphia, USA, 2000. [13] Rall, L.B. & Corliss, G.F., An introduction to automatic differentiation. Computational Differentiation: Techniques, Applications, and Tools, eds. M. Berz, C.H. Bischof, G.F. Corliss & A. Griewank, SIAM: Philadelphia, USA, pp. 1–17, 1996. [14] ACTIS. http://www.actis-isolation.com. [15] Kearley, V., Multifoil testing at TRADA. Technical report, TRADA Technology Ltd, United Kingdom.



147

Evaluation of insulation systems by in situ testing I. Enache1, D. Braga1, C. Portet1 & M. Duran2 1 2

Research and Development Department, Actis Limoux, France EMM, Bruxelles, Belgium

Abstract Due to the global warming of the earth, the energy performance of buildings is now a crucial subject. In order to have an accurate comprehension of the energy loss of a building, we have developed real time in-situ tests for the thermal performance of building insulation systems. The test cells are pitched roofed with two gable walls, have the same interior and exterior dimensions, are placed in outside weather conditions and are constructed with the same materials (apart from the roof and gables insulation). All cells are heated in the same way. The temperature inside the test cell is maintained at the same specified level in winter by fan heaters. By comparing the air temperature within the test cell to the outside weather conditions and monitoring the energy required to maintain the internal temperatures, the real life thermal efficiency of each insulation system can be estimated. This paper deals with the results obtained over several test centres around Europe using a thin multi-layer reflective insulation product for the insulation of the first test cell, mineral wool for the insulation of the second one, and without insulation for the third one. Keywords: in situ testing, real life thermal efficiency, thin multifoil insulation.

1

Introduction

Reducing CO2 emissions has become one of the most important challenges for all industrial sectors. Concerning the building industry, there has to be a significant improvement in thermal performance. New regulations continuously reinforce the requirements on the contribution of building products to energy saving.

WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090141

148 Computational Methods and Experimental Measurements XIV Nowadays, the estimate of energy demand for a building is made with dedicated software that allows a study of the impact of different solutions contributing to the building energy requirement. Insulation products are a key parameter. Their thermal properties are estimated by standardized guarded hot plate or hot box measurement. However, important differences between the standard calculation [1,2] and in situ measurements can be observed. A detailed report from the Building Research Establishment Ltd [3] has concluded that there are important differences, in certain cases equivalent to almost 30%, between the calculation of the coefficient of thermal transmission U under the norm ISO 6946 [4] and the measurements in situ made by the Alba Building Science company on the walls of buildings built between 1995 and 1999. In this context, the European Multifoil Manufacturers association (EMM) has chosen to categories different insulation systems using in situ tests. This method has the advantage of taking into account the influence of real conditions on the thermal performance of different insulation solutions. In this way this technique gives much more realistic information about the thermal behaviour of the tested insulation product once installed. In situ tests developed by EMM and presented here concern three identical buildings insulated with different insulation systems: one insulated with a thin multi reflective product (MF), one insulated with 200 mm of mineral wool (MW) and the last one is not insulated. The in situ measurements are also compared with simulation results obtained with TRNSYS® software.

2

Experimental part

2.1 Structural description of the test cells Three test cells in timber frame representative of an attic that can be converted, with a roof surface about 35 m², outside dimensions of 4 x 7 m² on floor area and a height of 3 m were used for the described tests (figure 1). The access to each test cell is gained through an insulated airlock situated on the gable wall. The gable walls and the attached airlock are made of 23 mm thick plywood. The roof (36° pitch), is traditional, made up of rafter of 8 x 11 cm with a roof void of 48 cm whilst the roofing is made of clay tiles. The floor is timber with an under floor void and over-insulated with 20 cm of polystyrene and 10 cm of mineral wool. In order to obtain a reasonable accuracy of the thermal performance of the insulation system, the cells have no windows and no controlled ventilation system. Also, the temperatures of the airlock entries and under floor spaces are maintained at the same level as inside; therefore the thermal exchanges take place only through walls insulated with the tested material. 2.2 Insulation set-up Insulation products tested are (figure 2): - Non-commercial thin (45 mm) multi reflective insulation product (MF) with a core thermal resistance of 1.25 m²K/W measured with standards methods [1,2]. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


149

- Mineral wool of 200 mm (MW) with a declared thermal conductivity of 0,040 W/m.K (RD = 5 m²K/W). An additional and continuous vapour control layer made of polyethylene foil (0.2 mm) was placed between mineral wool and plasterboard. - The last cell had no insulation above the plasterboard. The layout has been contrived under the manufacturers’ instructions [5,6]. A HPV under tile liner was placed under the tiles in each cell. 2.3 Test method In order to analyse the thermal behaviour of the insulation systems in different conditions, the three cells have been placed on exposed sites in three different

Figure 1:

Photography of the test cells. Under tile liner and ventilated airgap IRMM

200 mmMW Vapor control layer Plasterboard

Figure 2:

Scheme of roof constructions (for the non-insulated cell, the same configuration was contrived without any insulation) Table 1:

Site North of Europe (lat 54°N) Center of Europe (43° N) South of Europe (40 °N)

Characteristics of the test sites. Average temperature Low High High

Temperature variation Very low High Very high


Wind speed High High Low

Solar radiation Low High Very high

150 Computational Methods and Experimental Measurements XIV regions of Europe with very different weather conditions. The typical weather characteristics of the three sites are shown in table 1. 2.3.1 Instruments inside the testing cells Each cell has been equipped with two temperature sensors located at 1.50m above the floor. The sensors are placed in an open grey PVC tube in order to protect them from air movements which might affect the measurements. Seven other temperature sensors are installed in the cell in order to control the temperature distribution in the cell volume. During test in winter conditions, the temperature in the cells is maintained at a constant level with two fan heaters of 1 kW. Current and voltage measurements using calibrated transducers allow, in each cell, the determination of the exact energy consumption. A meteorological station, located nearby the cells is equipped to permanently record the following weather parameters: temperature, relative humidity, pressure, global solar radiation and wind speed. All measurements are registered on a constant rate (one per minute) 2.3.2 Cells calibration Before the test starts, a calibration phase takes place in order to ensure that the 3 test cells are similar in terms of internal dimensions and thermal performance when insulated with the same product. During this phase, the three cells are insulated with 200 mm mineral wool and the temperature inside is maintained at the same level for five days. The actual recorded energy consumptions were similar (difference less than 5%) and therefore the cells are considered to be identical. 2.3.3 Test of different insulation systems After the calibration phase, the cell which consumed the lower amount of energy retained MW as insulation product. The cell which consumed the higher amount of energy is not insulated during the test. Finally, the last test cell is insulated with MF. Before the test started, new measurements of the cell interior dimensions and air tightness were performed. As we can see in table 2, the three cells are very similar. Table 2:

Interior dimensions and air tightness of the test cells.

Surface (m²) MF

North of MW Europe Not insulated MF

Centre of MW Europe Not insulated MF

South of MW Europe Not insulated

43.62 44.18 44.41 44.19 45.31 44.71 43.93 44.51 44.59

Volume (m3) 27.78 28.37 28.45 28.93 30.32 30.18 28.15 28.72 28.79


n50 (h-1) 4.60 5.00 4.85 4.50 4.35 4.40 5.29 5.57 5.72


Table 3:

151

Daily weather conditions and energy consumptions on the south of Europe site. South of Europe

Date

Temp (°C)

Wind speed (m/s)

29/01/2008 09:00 30/01/2008 09:00 31/01/2008 09:00 01/02/2008 09:00 02/02/2008 09:00 03/02/2008 09:00 04/02/2008 09:00 05/02/2008 09:00 06/02/2008 09:00 07/02/2008 09:00 08/02/2008 09:00 09/02/2008 09:00 10/02/2008 09:00 11/02/2008 09:00 12/02/2008 09:00 13/02/2008 09:00 14/02/2008 09:00 15/02/2008 09:00 16/02/2008 09:00 17/02/2008 09:00 18/02/2008 09:00 19/02/2008 09:00 20/02/2008 09:00 21/02/2008 09:00 22/02/2008 09:00 23/02/2008 09:00 24/02/2008 09:00 25/02/2008 09:00 26/02/2008 09:00 27/02/2008 09:00 28/02/2008 09:00 29/02/2008 09:00 01/03/2008 09:00 02/03/2008 09:00 03/03/2008 09:00 04/03/2008 09:00 05/03/2008 09:00 06/03/2008 09:00 07/03/2008 09:00 08/03/2008 09:00 09/03/2008 09:00 10/03/2008 09:00 11/03/2008 09:00

1.51 1.44 2.52 6.03 4.38 6.99 6.16 7.46 5.74 5.09 4.59 4.32 3.45 3.75 3.38 2.80 7.29 5.31 5.96 6.87 7.60 8.38 7.83 7.06 7.41 10.24 9.63 7.95 8.22 10.36

0.24 0.38 0.30 1.46 1.17 3.00 1.29 1.14 0.43 0.21 0.17 0.24 0.22 0.23 0.47 0.40 0.49 0.27 0.24 0.65 0.65 0.13 0.35 0.52 0.35 0.42 0.30 0.49 0.49 0.62

3

Solar radiation (W/m²) 152.82 97.79 157.27 57.26 144.94 49.90 130.35 74.17 165.90 166.90 164.42 172.50 170.31 174.30 170.93 173.84 113.83 125.39 166.95 140.67 83.84 55.08 111.04 195.91 199.26 141.79 85.40 189.18 124.73 116.47

kWh MF

kWh MW

kWh Not insulated

8.88 9.08 9.05 7.68 8.21 7.48 7.44 6.57 7.01 7.44 7.10 7.27 7.45 7.59 7.67 7.96 6.76 6.25 6.68 6.58 6.07 5.60 5.48 5.87 5.90 5.33 4.98 5.15 5.61 4.89

9.06 9.21 8.87 7.35 7.42 6.83 6.76 5.89 6.13 6.67 6.31 7.00 7.26 7.39 7.38 7.76 6.57 5.99 6.54 6.38 5.75 5.29 5.13 5.56 5.74 5.18 4.73 4.77 5.44 4.66

30.81 31.29 29.93 26.08 28.56 27.22 25.81 22.62 23.38 24.21 24.64 24.97 26.04 27.10 27.33 28.59 22.98 23.32 23.57 23.03 21.86 20.24 20.62 21.70 21.40 18.00 17.90 19.00 19.84 17.38

6.51 6.63 6.60 4.21

23.31 25.03 26.54 15.97

Invalid data

7.20 7.04 8.39 12.79

0.85 2.01 3.46 2.44

222.45 170.01 42.27 210.80

6.67 6.78 6.73 4.36

Experimental and simulation results

3.1 Experimental measurements In situ testing was performed during the 2007-2008 winter. The weather conditions encountered during this period and the energy needed to maintain the WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

152 Computational Methods and Experimental Measurements XIV temperature set point into the cells (21°C for North of Europe, 23°C for the two other sites) are presented in tables 3 to 5. The energy consumption of the three test cell for the entire test period on each site is given in the table 6. Obviously, the energy consumption depends on the number of days of test but also on the weather conditions and the temperature inside the cells. Table 7 gives the values of the thermal transmittance for the each cell on the three test sites. Table 4:

Daily weather conditions and energy consumptions on the centre of Europe site. Centre of Europe

Date 14/01/2008 15/01/2008 16/01/2008 17/01/2008 18/01/2008 19/01/2008 20/01/2008 21/01/2008 22/01/2008 23/01/2008 24/01/2008 25/01/2008 26/01/2008 27/01/2008 28/01/2008 29/01/2008 30/01/2008 31/01/2008 01/02/2008 02/02/2008 03/02/2008 04/02/2008 05/02/2008 06/02/2008 07/02/2008 08/02/2008 09/02/2008 10/02/2008 11/02/2008 12/02/2008 13/02/2008 14/02/2008 15/02/2008 16/02/2008 17/02/2008 18/02/2008 19/02/2008 20/02/2008 21/02/2008

09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00

Temp (°C)

Wind speed (m/s)

6.91 7.20 7.69 11.95 13.12 9.80 5.86

0.71 1.64 4.68 3.96 4.21 1.00 0.09

10.78 5.94 7.97

5.71 1.03 1.02

4.36 8.00 5.88 5.15 6.73 3.33 7.07 3.19 6.97 5.16 6.65 10.93 8.12 7.44 6.93 6.22 8.94 6.91 8.50 6.87 6.12 7.71 8.87 10.49 12.02 9.76 11.38

0.02 0.87 0.92 1.30 3.72 1.49 1.25 1.73 1.41 8.34 0.13 3.95 0.12 4.14 1.97 1.60 2.44 1.71 1.95 1.31 0.37 1.05 1.97 1.20 1.74 0.77 0.77

Solar kWh radiation MF (W/m²) 47.50 6.45 62.66 6.43 49.76 6.70 67.97 5.05 46.29 3.94 42.33 4.99 94.88 6.63 Invalid data 26.20 5.15 83.72 6.19 94.18 6.47 Invalid data 106.17 7.31 106.97 6.61 107.59 6.65 79.60 7.06 36.66 6.55 26.68 7.38 68.61 6.26 92.41 7.43 97.97 6.70 91.00 6.94 128.27 6.72 102.14 4.87 133.18 5.56 138.24 6.25 138.62 6.30 137.64 6.55 134.56 6.25 144.15 6.22 143.57 6.17 115.38 5.90 152.56 6.43 161.24 6.34 176.31 5.42 148.89 5.19 103.04 3.91 42.82 4.41 154.30 4.09


kWh MW

kWh Not insulated

5.45 5.55 5.91 4.29 3.30 4.15 5.73

23.54 23.49 26.33 18.60 15.84 19.24 25.19

4.85 5.44 5.72

20.70 24.50 23.09

6.46 5.94 5.87 6.24 5.94 6.51 5.62 6.67 6.00 6.08 5.97 4.33 4.89 5.58 5.60 5.79 5.62 5.46 5.48 5.10 5.83 5.71 4.73 4.59 3.47 3.81 3.70

27.56 23.81 25.49 26.28 25.68 28.85 23.11 29.57 23.78 27.26 24.77 17.02 21.42 23.19 23.58 24.54 23.38 23.58 22.61 22.09 24.62 22.82 20.59 17.92 14.21 16.49 16.67


Table 5:

153

Daily weather conditions and energy consumptions on the north of Europe site. North of Europe

Date 01/02/2008 02/02/2008 03/02/2008 04/02/2008 05/02/2008 06/02/2008 07/02/2008 08/02/2008 09/02/2008 10/02/2008 11/02/2008 12/02/2008 13/02/2008 14/02/2008 15/02/2008 16/02/2008 17/02/2008 18/02/2008 19/02/2008 20/02/2008 21/02/2008 22/02/2008 23/02/2008 24/02/2008 25/02/2008 26/02/2008 27/02/2008 28/02/2008 29/02/2008 01/03/2008 02/03/2008 03/03/2008 04/03/2008 05/03/2008 06/03/2008 07/03/2008 08/03/2008 09/03/2008

Temp (°C) 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00 09:00

1.61 1.88 0.44 1.73 2.14 4.00 3.76 2.30 3.88 3.44 0.59

Wind speed (m/s) 4.44 4.17 3.11 3.44 2.31 2.64 1.49 1.06 2.94 3.89 1.87

Solar radiation (W/m²) 14.11 31.64 66.52 66.69 7.53 25.43 33.06 26.18 52.27 31.72 7.21

kWh MF

kWh MW

kWh Not insulated

7.74 7.73 8.17 7.75 7.07 6.54 6.57 6.78 6.65 6.76 7.63

4.96 5.08 5.27 4.98 4.58 4.22 4.27 4.43 4.26 4.34 4.88

32.07 31.17 31.94 28.85 26.66 24.23 24.73 25.62 25.92 27.63 29.92

7.25 6.01 5.02 5.56 5.00 4.37 4.37 5.14 4.12 4.62 3.88 4.28 5.10 4.71 4.41 5.29 5.21 5.68 5.58 5.41 4.94 4.08 3.70

39.31 32.48 27.14 38.69 27.66 25.86 23.51 26.89 22.46 25.97 22.19 25.13 27.73 26.15 25.32 29.08 28.64 31.31 31.35 32.85 27.24 21.98 20.29

Invalid data -6.17 0.36 2.27 -0.57 2.27 3.85 6.78 4.36 6.79 4.28 6.75 5.39 3.69 3.32 3.56 2.44 1.85 -0.20 0.19 1.17 2.27 4.45 5.95

2.50 4.48 3.35 2.58 2.47 3.41 7.12 5.65 4.74 4.99 4.28 5.94 5.27 3.31 3.77 5.40 3.08 3.39 3.20 4.98 2.59 1.23 1.56

94.55 11.99 18.91 93.72 31.95 45.41 17.18 22.23 10.47 25.55 58.22 36.40 60.18 33.39 12.93 24.37 44.00 118.53 109.21 30.78 101.26 55.65 52.99

10.89 8.63 7.38 8.46 7.38 6.65 5.93 6.96 5.92 6.58 5.71 6.11 7.11 6.86 6.58 7.43 7.52 8.28 8.21 7.94 7.30 6.16 5.61

Table 6: Energy consumptions on the three test sites during the entire test period. Site South of Europe Centre of Europe North of Europe

Days of test 34 27 34

MF 229.5 223.5 245.0

Energy consumption (kWh) MW Not insulated 219.0 810.3 197.4 841.4 165.0 948.0



Thermal transmittance of the tested cells. Thermal transmittance (W/m²K) MF MW Not insulated 0.37 0.35 1.32 0.37 0.32 1.39 0.39 0.26 1.50

Site South of Europe Centre of Europe North of Europe

Energy savings MF MW 72% 73% 73% 77% 74% 83%

Comparison between in situ measurements and the Trnsys® simulation.

Table 8:

South of Europe MW Configuration Without thermal bridges Thermal bridges (Sext)

In situ (kWh) 218.96

Trnsys (kWh) 121.11 156.51

MF

Ecart Air change to fit (Trnsys-In situ) measurements (h-1) -45% 0.73 -29% 0.46


Trnsys (kWh) 243.83 328.99

Ecart (Trnsys-In situ) 6% 43%

Trnsys (kWh) 257.08 344.27

Ecart (Trnsys-In situ) 15% 54%

Trnsys (kWh) 241.54 327.81

Ecart (Trnsys-In situ) -1% 34%

Air change to fit measurements (h-1) -

Centre of Europe MW Configuration Without thermal bridges Thermal bridges (Sext)


Trnsys (kWh) 128.09 163.21

MF

Ecart Air change to fit (Trnsys-In situ) measurements (h-1) -35% 0.50 -17% 0.24


Air change to fit measurements (h-1) -

North of Europe MW Configuration Without thermal bridges Thermal bridges (Sext)


Trnsys (kWh) 126.95 166.43

Ecart Air change to fit (Trnsys-In situ) measurements (h-1) -23% 0.28 1% -

MF In situ (kWh) 245.03

Air change to fit measurements (h-1) 0.03 -

Table 7 shows that the thermal performance of MF is higher in the south and centre of Europe compared to the north of Europe. MW has an opposite effect with higher performance on the north of Europe and lower in the south. The same table presents the energy savings of each insulated cell compared to the non-insulated cell. One can see that in the north of Europe, the MW has higher energy savings than the MF. This is different on the two other sites where the cell insulated with MF and the one insulated with MW have very similar energy savings. 3.2 Simulation results The results presented in this section have been obtained with TRNSYS® software [7]. The simulations have been performed using the exact geometry of the cells and the weather conditions registered on each test centre. The input thermal properties of each wall of the structure are determined using standardised methods for measurement and calculation [1,2,5]. The simulation results in terms of energy consumption are compared with measurements in table 8. If the linear thermal bridges are not taken into account then the simulation results for the MW cell fit with the measurements using a level of air infiltration between 0.28 h-1 and 0.73 h-1. If thermal bridges are taken into account by considering the external surface of the cell as heat loss surface than the level of air infiltration to fit simulation results with measurements is considerably reduced (max value is 0.46 h-1).



155

For MF cell, simulations overestimated the energy needed to maintain the temperature set point into the cells even without taking into account the linear thermal bridges effect. Moreover, as shown in table 2, the cells air tightness is very similar and therefore the same level of infiltration found for MW cells should be applied to the MF one. In this case the difference between simulations and measurements is up to 70 %. This inconsistency could be explained by the underestimation of the thermal performance of MF + adjacent air gaps. This raises the question of the efficiency of traditional methods and the need for a new suitable method to determine the correct resistive characteristics of the cavities with reflective walls. A three dimensional CFD model coupling the different heat transfer mechanisms could allow a better understanding of the thermal behaviour of MF product.

4

Conclusion

The in situ tests performed in regions of Europe with different weather conditions have showed that the thermal performance of the MF product is clearly underestimated by the standard measurements and calculations currently employed. The weather conditions seem to have a high impact on the thermal performance of different insulation systems. The difference between calculation and in situ measurement is lower on the site situated in north of Europe probably because the test conditions are very similar with those imposed by the standard methods i.e. very low temperature variation. The highest difference between calculation and measurement is obtained in the south of Europe where the weather conditions are completely different from those imposed by standards: high temperature variation during the course of one day. This clearly shows that the actual standards are not appropriate to determine the true thermal performance of MF products. The protocol detailed in this paper also allows the direct determination of the energy saving using a given insulation system in comparison with a noninsulated cell. It appears that the MF solution allows a significant energy saving and, therefore, this solution can be an interesting alternative especially for old buildings where the space for thick insulation is not available.

Acknowledgement We would like to thank all people helping us in this project: T. Labrousse, F. Laché, B. Saintpeyre, B. Sanchez, T. Bonnafoux.

References [1] ISO 8302, Thermal Insulation. Determination of steady-state thermal resistance and related properties -- Guarded hot plate apparatus. [2] ISO 8990, Thermal insulation. Determination of steady-state thermal transmission properties – Calibrated and guarded hot box. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

156 Computational Methods and Experimental Measurements XIV [3] Report BRE n° 78132, Fields investigations of the thermal performance of construction elements as built, June 2001. [4] ISO 6946, Building components and building elements - Thermal resistance and thermal transmittance – Calculation method. [5] ACTIS guideline on http://www.actis-isolation.com [6] Example of set up on: http://www.knaufinsulation.co.uk/PDF/Book_3_2_2_ Pitched_Roofs_Rafter_Level.pdf [7] http://sortware.cstb.fr/soft/present.asp?page_id=fr!Trnsys [1] ISO 8302, Thermal Insulation. Determination of steady-state thermal resistance and related properties -- Guarded hot plate apparatus. [2] ISO 89 [3] Report BRE n° 78132, Fields investigations of the thermal performance of construction elements as built, 2001. [5] ACTIS guideline on http://www.actis-isolation.com [6] Example of set up on: http://www.knaufinsulation.co.uk/PDF/Book_3_2_2_Pitched_Roofs_Rafter_Lev el.pdf [7] http://sortware.cstb.fr/soft/present.asp?page_id=fr!Trnsys



157

Monitoring coupled moisture and salt transport using single vertical suction experiment Z. Pavlík, J. Mihulka, M. Pavlíková & R. Černý Department of Materials Engineering and Chemistry, Faculty of Civil Engineering, Czech Technical University in Prague, Czech Republic

Abstract A new method for simultaneous monitoring of coupled moisture and chloride ions transport is presented in the paper. The experiment is done in the conditions of one-sided 1.0 M NaCl solution vertical uptake into a sample of calcium silicate based material. In the experimental work, rod-shaped sample is used for the determination of moisture and chloride concentration profiles in simulated 1D water and chloride solution transport. For the measurement, advanced Time Domain Reflectometry (TDR) sensors are used. The sensors allow for moisture content assessment on the basis of relative permittivity measurement and chloride concentration monitoring based on electrical conductivity measurement. On the basis of measured data, moisture and chloride concentration profiles are obtained. Experimentally determined chloride concentration profiles and moisture profiles are then used for identification of apparent chloride diffusion coefficient and moisture diffusivity on the basis of inverse analysis using a simple diffusion model. Finally, the calibration procedure of the applied measuring method is discussed and practical recommendations for application of the combined TDR/electrical conductivity sensors for monitoring of coupled moisture and salt solution transport are given. Keywords: moisture, salt concentration, relative permittivity, electrical conductivity, calcium silicate.

WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090151


1

Introduction

Moisture and salt induced damage is considered to be one of the essentially problematic elements in the decay of building materials and structures. There is evident that the rising of moisture content in buildings and their materials leads to serious negative events, like degradation of materials (disintegration of inorganic plasters, porous stones and ceramic bricks, binder decomposition, surface erosion, etc.). It has also negative effects on biological devaluation of constructions (mould growing) and on the conditions of interior climate from the point of view of hygienic aspects. Water can deteriorate building materials and structure surfaces by acid decomposition reactions. Typical example is sulphur dioxide that dissolves in water and partly forms sulphurous acid and sulphur trioxide which forms acid as well. Both acids decompose lime and lime-mixed binders in coatings [1, 2]. The final result is formation of gypsum, CaSO4·2H2O. The reaction can be expressed in a simplified manner as: (1) CaCO3 + SO2 + 1 / 2O2 + H 2O → CaSO4 ⋅ 2 H 2O + CO2 and

CaCO3 + SO3 + 2 H 2O → CaSO4 ⋅ 2 H 2O + CO2 .

(2)

Gypsum has a large molar volume and, under favourable humidity conditions, large crystals of gypsum form. As a result, crystallisation pressure decays the surface layer of lime based material. Significant is also the negative effect of moisture on compressive and bending strength of bearing-structures materials. In the areas, where the temperature fluctuates around 0°C, water gives rise to freeze-thaw weathering that especially deteriorates porous structure of building materials. Ice, compared to the water volume in liquid phase, has a volume that is 9% higher and its crystallisation pressure causes the damage of solid porous structure of materials. A part of building damages assigned to the negative moisture effects would not arise, if only pure water would be present. Water represents very often only transport medium for other harmful pollutants that take part in the process of surface degradation of building materials. Water transport in porous materials makes possible salt transport or accumulation as well. Salt accumulation in specific places of building structures can lead in consequence to their failure or destruction. Among water soluble salt action in building materials, especially salt crystallization, salt hydration, hygroscopic water absorption, efflorescence and leaching represent the most harmful effects on the properties of building materials and structures. Salt crystallisation is physico-chemical degradation process that is related to formation of saturated and oversaturated salt solutions due to water evaporation. After overrun of the range of solubility, the salt crystals are growing and exert crystallisation pressures on walls of porous space. In dependence on the strength of materials, the damage of porous structure of materials is initiated. Salt hydration is related to salts that are able to bond in their crystal lattice certain defined number of water molecules. They form hydrates what is accompanied by volume changes and hydration pressures. For building materials WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


159

are the most dangerous salts changing their forms at standard climatic conditions; sodium sulphate, sodium carbonate and calcium nitrate. Example of hydration of calcium nitrate is given in equation 30° C 100° C Ca ( NO3 ) 2 ⋅ 4 H 2 0 ← → Ca ( NO3 ) 2 ⋅ 3H 2O ← → Ca ( NO3 ) 2 . (3) Although the moisture and water soluble salts related problems of buildings and their particular materials are generally known and proven, the exact description of coupled moisture and salt transport mechanism in porous medium remains still open field for building physicists and engineers. On this account, the main motivation of the presented work is to contribute to the explanation of salt solution transport and identification of parameters that can be used for its more exact characterization and description.

2

Method for simultaneous measurement of moisture and salt concentration

Water and salt ions possess many anomalous properties, which also affect the properties of a porous material. Therefore, there exist various methods of determination of moisture and salt content in porous materials, and various moisture and salt concentration meters. The presence of salt ions can negatively affect the accuracy of moisture measurement methods, especially of relative methods, where the measured physical quantity is dependent on salt concentration. Therefore, in case of simultaneous moisture and salt concentration measurement, proper method which accuracy is not influenced by presence of salt ions must be chosen. As stated in literature, the high frequency microwave methods based on permittivity determination correlated to moisture content can be used for such type of measurements. In this paper we introduce TDR (Time Domain Reflectometry) method for moisture measurement combined with simultaneous assessment of salt concentration by means of electrical conductivity measurement. TDR technique represents specific methodology among the microwave impulse techniques. The principle of TDR device consists in launching of electromagnetic waves and the amplitude measurement of the reflections of waves together with the time intervals between launching the waves and detecting the reflections. Time/velocity of pulse propagation depends on the apparent relative permittivity of the porous material, which can be expressed using the formula

 ct  εr =  p   2 Ls 

2

(4)

where εr is the complex relative permittivity of the porous medium, c the velocity of light (3·108 m/s), tp the time of pulse propagation along the probe rods measured by TDR meter and Ls the length of the sensor’s rod inserted into a measured porous medium. The determination of moisture content using the permittivity measurements is then based on the fact that the static relative permittivity of pure water is equal to WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

160 Computational Methods and Experimental Measurements XIV approximately 80 at 20°C [3], while for most dry building materials it ranges from 2 to 6. For evaluation of moisture content from measured relative permittivity values, three basic approaches can be used. The first possibility is utilization of empirical conversion functions generalized for a certain class of materials. On the basis of analysis performed in [4], it can be stated that empirical conversion functions used in current research for TDR data conversion are anything but universal. They are always limited to specific groups of materials. The second is application of dielectric mixing models, which assumes knowledge of the relative permittivities of the material matrix, water, air and other parameters, that cannot be measured directly but have to be determined by empirical calibration of the model. Dielectric mixing models were tested in many practical applications and their perspectives for further use seem to be better than those of the empirical conversion functions [5–7]. The third method for evaluation of moisture content from measured relative permittivity consists in empirical calibration for the particular material using a reference method, such as the gravimetric method. This method is the most reliable until now and was used also in this work for calibration of TDR method for moisture measurement in calcium silicate based material. As it was mentioned above, measurement of salt concentration (in our case of chlorides) was done by means of electrical conductivity measurement that is in clear relation with salt concentration. The calibration was done using measurement of salt concentration by ion selective electrode together with measurement of electrical conductivity. In this way, the salt-concentration dependent electrical conductivity was accessed.

3

Experimental

The experiment for determination of moisture and salt concentration profiles was done in the conditions of one-sided 1.0 M NaCl solution vertical uptake into the sample of calcium silicate based material. In the experimental work, rod-shaped sample was used for the determination of moisture and chloride concentration profiles in simulated 1-D water and chloride solution transport. The sample size was 50/100/300 mm and all the lateral sides of the sample were vapor proof insulated by epoxy resin to ensure 1-D moisture and salt solution transport. Into the studied sample, 8 two-rod miniprobes LP/ms (Easy Test) were placed for the monitoring of complex relative permittivity and electrical conductivity. The sensors are made of two 53 mm long parallel stainless steel rods, having 0.8 mm in diameter and separated by 5 mm [3]. The sphere of sensor’s influence creates the cylinder having diameter about 7 mm and height about 60 mm, circumference around the rods of sensor. The accuracy of relative permittivity and electrical conductivity reading, and the measuring range of applied sensors are given in Table 1. For the TDR measurements in this paper, the cable tester TDR/MUX/mts produced by Easy Test which is based on the TDR technology with sin2-like



Table 1:

161

Accuracy and measuring range of applied TDR sensors.

Measured quantity Relative permittivity ε

Measuring range 2 ÷ 90

Electrical conductivity σ

0 ÷ 1 S/m

Accuracy Absolute error: ± 1 for 2 ≤ ε ≤6 ± 2 for ε ≥ 6 Relative error: ± 5%

needle pulse having rise-time of about 250 ps, was used. The working frequency of this device is 1.8 GHz for the relative permittivity measurement. The measuring technology was as follows. At first, the sensors were placed into the sample and sealed by silicone gel. Since the material is rather soft, the sensors were placed into the sample by simple impress. Then, the sample was put into a vessel containing 1.0 M NaCl water solution and the suction has started. The complex relative permittivity and electrical conductivity were continuously monitored and stored in computer. After specific time interval, the experiment was interrupted and the sample was cut to eight separate pieces containing particular sensors. Finally, the sensors were removed and in each piece the moisture content and chloride concentration were measured by reference method. The moisture content was accessed by gravimetric method and chloride concentration by ion selective electrode (device pH/ION 340i) applied on leaches from particular sample pieces. In this way, the empirical calibration curves of TDR method for calcium silicate were determined. The suction experiment was realized on calcium silicate material. It is material having high thermal insulation properties, high total open porosity (87%), low bulk density (230 kg/m3) and from the chemical point of view is formed by Ca2SiO4. Sample arrangement and measuring technology is visible from Fig. 1.

4

Determination of moisture diffusivity and chloride diffusion coefficient

Moisture diffusivity and chloride diffusion coefficient represent necessary input data for computational modeling of coupled moisture and chloride ions transport in porous building materials. Their knowledge is important also for evaluation of moisture and salt solution transport properties of specific materials and plays important role in the process of design of building structures. In this paper, inverse analysis of experimentally measured moisture and chloride concentration profiles was used for the computational identification of moisture diffusivity and chloride diffusion coefficient of calcium silicate. Inverse analysis is based on the assumed mode of salt solution transport. In the presented work we have assumed only diffusion mechanism of moisture as well as of chloride ions transport. In this way, the chloride diffusion coefficient is considered to be apparent parameter that includes also the effect of chloride ions binding on porous space walls, advection of salt ions, surface diffusion and WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

162 Computational Methods and Experimental Measurements XIV osmotic effects. Compared to simple Fick’s diffusion, the dependence of moisture diffusivity on moisture content κ(w) and apparent diffusion coefficient on salt concentration D(C) is considered. In this case, the salt mass balance is expressed by equation

∂C = div ( D (C ) gradC ) , ∂t

(5)

where C [kg/m3] is salt concentration in kg per volume of the dry porous body, D [m2/s] the apparent salt diffusion coefficient. In this way the salt solution transport is formally described by the same parabolic equation with the same boundary and initial conditions usually used for description of water transport. Therefore, the calculation of concentration-dependent diffusion coefficients from the measured salt concentration profiles could be done using basically the same inverse methods as those for the determination of moisture-dependent moisture diffusivity or temperature-dependent thermal conductivity. In this paper, this type of model was employed for determination of both D(C) and κ(w) functions. In calculations, we assume that the concentration field C(x, t) and moisture content field w(x,t) are known from the experimental measurements as well as the initial and boundary conditions of the experiment. Using Matano method [8] and applying two Boltzmann transformations we arrive to the following final formulas for calculation of apparent salt (in our case chloride) diffusion coefficient

D(C0 ) =

1 dC 2t0 ( ) z = z0 dz

∞

dC

∫ z dz dz ,

z0

and moisture diffusivity

Figure 1:

Experimental setup of vertical suction experiment.


(6)


Figure 2:

163

Relative permittivity as function of moisture content.

κ ( w0 ) =

1 dw 2t0 ( ) z = z0 dz

∞

dw

∫ z dz dz ,

(7)

z0

where C0 = C(z0, t0) is salt concentration in the position z0 and time t0, w0 = w(z0, t0) the corresponding moisture content and z space variable. The integral in equations (6) and (7) is solved by common numerical methods, such as Simpson’s rule. The details on inverse analysis procedure can be found e.g. in [9].

5

Experimental and computational results

Figs. 2, 3 present calibration curves of applied measuring techniques for moisture and chloride concentration measurement. The measured data show the dependences of relative permittivity on moisture content and electrical conductivity on chloride concentration which knowledge is imperative for the applicability of used methods for salt solution transport monitoring. For calibration purposes, the experimental data were smoothed by simple polynomial relations that can be considered as empirical calibration curves of TDR for measurement of chloride water solution transport in calcium silicate. The experimentally measured moisture and chloride concentration profiles are presented in Figs. 4, 5. The data gives clear evidence about the velocity of moisture and chloride ions propagation in calcium silicate material. Fig. 6 shows the moisture diffusivity as a function of moisture content, Fig. 7 the apparent chloride diffusion coefficient as a function of chloride concentration. These figures reveal the necessity to consider in balance equations the dependence of transport parameters on moisture and salt concentration. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


Figure 3:

Dependence of measured electrical conductivity on chloride concentration.

Figure 4:

Moisture profiles for calcium silicate measured by TDR.

Looking at the data presented in Fig. 6, the moisture diffusivity varies within the range of four orders of magnitude. A similar trend was found also for concentration dependent chloride diffusion coefficient. In the range of lower concentrations, typically up to 0.005 [g/g], the apparent diffusion coefficient is close to diffusion coefficient of chlorides in water. On the other hand, in range of higher concentrations it increases very rapidly. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


Figure 5:

Chloride concentration profiles measured by TDR.

Figure 6:

Moisture diffusivity of calcium silicate


165


Figure 7:

6

Apparent chloride diffusion coefficient of calcium silicate.

Conclusions

The experiment presented in this paper has proven the capability of TDR combined sensors for simultaneous monitoring of moisture and salt concentration in porous building materials. This finding is very perspective for future work, especially for building practice that requires complex, precise and reliable methods for moisture and salinity measurement. The assessment of chloride diffusion coefficient and moisture diffusivity represents important information for application of studied calcium silicate material in practice regarding to its intended use in interior thermal insulation systems of building envelopes. This data can also find use in computational modeling of moisture and chloride ions transport in calcium silicate based materials what can be useful for example at damage assessment by means of salt action.

Acknowledgement This research has been supported by the Czech Ministry of Education, Youth and Sports, under project No MSM 6840770031.

References [1] Rovnaníková, P., Environmental pollution effects on other building materials (Chapter 7). Environmental Deterioration of Materials, ed. A. Moncmanová, WIT Press, Southampton, pp. 217-247, 2007.



167

[2] Moncmanová, A., Environmental factors that influence the deterioration of materials (Chapter 1). Environmental Deterioration of Materials, ed. A. Moncmanová, WIT Press, Southampton, pp. 1-21, 2007. [3] Malicki, M. & Skierucha, W.M., A Manually Controlled TDR Soil Moisture Meter Operating with 300 ps Rise-Time Needle Pulse. Irrigation Science, Vol. 10, pp. 153-163, 1989. [4] Fiala, L., Pavlík, Z., Jiřičková, M., Černý, R., Sobczuk, H. & Suchorab, Z., Measuring Moisture Content in Cellular Concrete Using The Time Domain Reflectometry Method. CD-ROM Proceedings of 5th International Symposium on Humidity and Moisture, J. Brionizio, P. Huang (eds.), Inmetro, Rio de Janeiro, paper No. 103, 2006. [5] Dobson, M.C., Ulaby, F.T., Hallikainen, M.T. & El-Rayes, M.A., Microwave dielectric behavior of wet soil, Part II: Dielectric mixing models, IEEE Trans. Geosci. Remote Sensing GE-23, pp. 35-46, 1985. [6] Jacobsen, O.H. & Schjonning, P., Comparison of TDR Calibration Functions for Soil Water Determination. Proceedings of the Symposium Time-Domain Reflectometry - Applications in Soil Science, L. W. Petersen and O. H. Jacobsen (eds.), Danish Institute of Plant and Soil Science, Lyngby, pp. 25-33, 1995. [7] Pavlík, Z., Fiala, L., Pavlíková, M., Černý, R., Sobczuk, H. & Suchorab, Z., Calibration of the Time Domain Reflectometry Method for Measuring Moisture Content in AAC of Various Bulk Densities, ISEMA 2007, Hamamatsu: Shizuoka University, pp. 151-158, 2007. [8] Matano, C., On the relation between the diffusion coefficient and concentration of solid metals. Jap. J. Phys., 8, pp. 109-115, 1933. [9] Fiala, L., Pavlík, Z., Pavlíková, M. & Černý, R., Water and Chloride Transport Properties of Materials of Historical Buildings. Recent Developments in Structural Engineering, Mechanics and Computation, Rotterdam: Millpress Science Publishers, pp 581-582, 2007.




169

Application of image analysis for the measurement of liquid metal free surface S. Golak Faculty of Materials Science and Metallurgy, Department of Electrotechnology, Silesian University of Technology, Poland

Abstract During the induction melting and stirring of molten metal a meniscus forms on the surface of the bath. This phenomenon affects the processes occurring on the metal-gas interface since they change the size of the real free surface of the metal. Numerical simulations indicate a significant scale of the phenomenon during typical processes of the induction melting of metals. Experimental measurements of the real free surface of induction melted metals can be quite difficult because of the rapid changes in the surface shape and high temperature of the molten metal. Therefore, a fast non-contact method with the use of laser and image analysis was proposed. However, in the case of liquid metals an optical method of measurement has its limitations as many molten metals shine. The problem was resolved by the use of a green laser and a narrow-band optical filter. This paper presents the methodology and problems of meniscus shape non-contact measurements. Keywords: liquid metal, free surface, laser measurement.

1

Introduction

The processes of induction melting, stirring and refining of liquid metals in crucibles and ladles are becoming more and more popular in the metallurgical industry. This situation promotes constant development of the equipment used in the above-mentioned processes, which in turn means continuous improvement of the designing methods. One of these is to expand the potential of simulating metallurgical processes and indeed it can be observed nowadays that more and more physical phenomena are taken into account in such simulations. At the WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090161

170 Computational Methods and Experimental Measurements XIV beginning just one-dimensional or quasi two-dimensional analytical models of electromagnetic and temperature fields were constructed. Then, together with the development of computer science, there appeared numerical tools for simulation of these fields and their coupling. Greater availability of computers and higher computing power contributed to progress in computational fluid dynamics, which made it possible to simulate liquid metal hydrodynamics occurring in real metallurgical devices. Initially, the simulations of electromagnetic and hydrodynamic fields were conducted as separate stages. First, the distribution of electromagnetic forces acting on the metal was calculated, and it was followed by the determination of the liquid metal velocity field induced by these forces. Those simulations assumed that the geometry of the liquid metal is not subject to change. The most frequent assumption was even that the surface of the liquid metal was perfectly flat. In fact, each metal forms on its surface a meniscus being the resultant of the surface energy of the crucible walls and total surface tension of the liquid metal. This process intensifies significantly under the influence of an electromagnetic field. The geometry of the liquid metal is distorted, and in effect the actual contact area between the metal and the atmosphere is bigger than in the case of a flat surface (Blacha et al. [1], Golak and Przylucki [2,3]). The simulations conducted in our department have proved that the surface area after deformation might even be increased by as much as 150 percent. Information of the real shape that the surface assumes is of great significance in the analysis of the melting, stirring and refining processes. During metal melting and heating it is the geometry of the melt that determines the actual power emitted in the charge, and consequently the capacity of the equipment in question. The distribution of the velocity field in induction stirrers determined for a flat free surface will be substantially different from the distribution for the deformed surface. Yet, the greatest importance of a free surface can be seen in the refining and other processes occurring at the interface of the phase’s liquid metal–atmosphere. Processes of this type show almost linear dependency on the actual size of the metal free surface. For this reason inaccurate estimation of the bath shape may completely distort the results of the most accurate simulations. The above-mentioned simulations conducted in our department revealed the scale of the phenomenon and its significance. However, the results of numerical simulation should be verified experimentally in order to validate the methodology applied and make any necessary corrections. To do this the measurements of the real shape of the bath must be performed and compared with the simulation results. In this way not only the calculations of the bath shape may be verified, but also, indirectly, the calculations of the electromagnetic forces and velocity field can be confirmed as the obtained meniscus is a resultant of the processes represented by these simulations. Measurements of a liquid metal are by no means an easy task because of its high temperature and chemical activity. That is why non-contact methods are often preferred. The level of liquid metal may be determined by point measurement performed with the use of radar, ultrasonic, and recently also laser sensors. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


171

Within the range of applications discussed, however, the point measurement proves to be insufficient. It turns out to be necessary that the whole shape of the liquid metal should be scanned with adequate resolution. The paper presents an attempt to adapt a widely used technique of 3D laser scanning to measure the geometry of a liquid metal in induction crucibles.

2

Method description

Laser measurement of the level is most often performed by triangulation method because of its accuracy and relative simplicity (e.g. Mass et al. [4]). It consists in a projection of a laser beam at a certain angle onto the surface from which the distance is to be measured and the camera recording of the beam reflection.

b

c

L

camera sensor

a laser

 y

x surface of liquid metal

Figure 1:

Level measurement by the method of diffused light recording.


172 Computational Methods and Experimental Measurements XIV Two techniques of measurement may be distinguished here. In one of them it is the light diffused on the surface that is recorded, while in the other it is the beam reflected from the surface and projected onto the screen. The choice between the above-mentioned methods of level measurement (and consequently surface profile measurement) of the liquid metal will depend on several factors. The first method is applicable when the laser beam is diffused on the surface of the liquid metal, which occurs when various kinds of impurities, micro-bubbles and the like can be found on it. Despite the inconvenience, the method is widely used for the simplicity of the measurement procedure. Figure 1 presents a schematic diagram of the measurement of metal level by the method of diffused light recording. As can be seen, a shift in the level of liquid causes a shift of the recorded light spot of the laser beam. Coordinates of the surface point can be estimated by the solution of equation set (1). c b  x  a y  L   y  tan    x 

(1)

where: x,y – coordinates of the surface point  – the tilt of the laser L – the position of the laser beam recorded by the camera a,b,c – geometric values (see fig. 1) In the case of some metals which melt at high temperatures (e.g. copper) a problem of their shining occurs. The spectrum of this light has a distribution similar to the light of a black body (fig. 2) described by Planck’s law. ucc [-]

1,40E-06 1,20E-06 1,00E-06 8,00E-07 6,00E-07 4,00E-07

2,00E-07 0,00E+00 300

Figure 2:

350

400

450

500

550

 [nm]

600

650

700

Normalized Planck distribution for the temperature of liquid copper.



173

Laser light reflected from the metal surface is then suppressed and its recording is not possible. The solution here is a narrow-band filter on the camera lens that will remove the spectrum components radiated by the metal and leave only the part of spectrum components in which the laser beam light is contained. Semiconductor red lasers, which are most frequently used, emit light of wavelength 650 nm, which is contained in the part of spectrum radiated by the heated metal. For this reason, a good idea is to use a laser with a shorter wavelength is suggested. Green semiconductor laser with wavelength of 532 nm have been recently available on the market. Within this range of radiation hot metal shines less brightly, and owing to filtration it is easier to distinguish between the laser light and metal shine. The method of point measurement of the liquid metal can be extended to linear measurement. In order to do this a dot line must be projected onto the metal surface. In 3D laser scanners a mechanical system deflecting the laser beam is commonly used. In the case of induction devices, however, quickly changing processes have to be often recorded. Having in mind fast digital cameras so easily available today, it must be concluded that the mechanical system would become the factor limiting the speed of measurement. Therefore what was applied in the presented solution was multiplication of laser beam by an optic method based on interference grid. The selection of a suitable head allows a dot line to be projected onto the surface of the liquid metal. Since the induction machines considered here are usually characterized by full axial symmetry, the measurement of the radial component of the curvature makes it possible to determine full geometry of the surface. The optic method can also be used to project a dot matrix, which will allow a full scanning of the surface. When the laser ray is not diffused on the metal surface, the method of the reflected beam should be applied. Figure 3 shows the idea of this measurement. Coordinates of the surface point are estimated by solution of equation set (2).

 y  b  H  a  x  tan   2    y  x tan  

(2)

where: x,y – coordinates of the surface point  - the tilt of surface  - the tilt of the laser H – the position of the laser beam spot on the projection panel recorded by the camera a,b – geometric values (see fig. 3) The technique is sensitive to the curvature of the surface from which the distance is measured. As Figure 4 presents, even small tilt of the surface which reflects the beam causes a significant change in its direction. A shift of the light spot on the screen resulting from the change of the beam direction may be far more serious than the shift resulting from the change in the metal level. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

174 Computational Methods and Experimental Measurements XIV Because in the application considered a meniscus occurs on the surface of the liquid metal, the method of point measurement based on the reflected beam is completely useless. However, multipoint measurement will allow the calculation of the metal surface curvature on the basis of the measurements performed by the method of reflected beam, according to solution of equation set (3).

0i N  y i  b  H i  a  x i  tan  i  2 i   y  x tan   0i N i  i i 2   x x x i i i 1 tan    y  i i 1  x i 1  x i xi 1  x i 1   0i N 2 x i  x i 1  x i 1 2 x i  x i 1  x i   y i x  x x  x   y i 1 x  x x  x  i i 1 i i 1 i 1 i 1 i 1 i  2 x 0  x1  x 2 2 x0  x0  x 2 tan    y y  0 0  x 0  x1 x 0  x 2  1 x1  x 0 x1  x 2   2 x 0  x 0  x1 y  2  x  x x  x  2 0 2 1  2 x N  x N 1  x N tan    y  N N 2  x N  2  x N 1 x N  2  x N   2x N  x N 2  x N 2 x N  x N  2  x N 1 y  N 1 x  x x  x   y N x  x  x  x  N 1 N 2 N 1 N N N 2 N N 1   x0  x2  x1  x 2   y 0  x  x  x  x   y1  x  x  x  x   1 0 1 2 0 1 0 2    x 0  x1 0  y2   x 2  x 0  x 2  x1 

(3)

where: xi,yi – coordinates of the i-th surface point i - the tilt of surface estimated from derivative of parabolically approximated function of shape i - the tilt of the laser i-th laser beam Hi – the position of the i-th laser beam spot on the projection panel recorded by the camera a,b– geometric values N – a number of laser beams and estimated surface points. The last equation of the set represents the known, zero value of the surface tilt on the axis of the crucible. Unfortunately, although the application of the above method solves the problem of the meniscus curvature effecting the measurement of the beam WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


175

reflection, it does not eliminate the errors resulting from local changes in the tilt of the surface. The solution to this problem lies in the assumption that the curvature of the surface caused by the meniscus is subject to changes much slower than the local deformations in the surface shape. This method relies on the measurement of the surface shape in a given period of time (short enough to record momentary shape of the meniscus), and then averaging the obtained result in time.

a laser



b y

projection panel

H

- -2 

x surface of liquid metal

Figure 3:

Level measurement by the method of reflected light recording.

laser

Figure 4:

projection panel

Influence of surface tilt.



3

Method application

On the basis of the measurement methods developed an experimental setup was designed where the technique of diffused beam was applied. A green semiconductor laser with wavelength of 532 nm was used. The model chosen for recording of the light (both reflected and diffused) was a colour camera, uEye UI-1640-C, resolution 1280 x 1024, with CMOS matrix and scanning frequency of 25 Hz with full resolution and the possibility of increasing the frequency (at the expense of resolution) do 254 Hz. The camera was equipped with the lens Pentax C1614-M with focal length of 16mm. So instrumented, the camera will monitor an area of 12 cm by 10 cm from the distance of 0,5 m. A 532 nm narrow-band interference filter with bandwidth of 3nm was also added to the measurement system. Figure 5 presents the arrangement of the stand for the measurement by the method of diffused beam. Theoretical accuracy of the measurement of coordinates in this arrangement can be calculated from eqn (4) and is equal to about 10-4 m. In reality, the accuracy is reduced along with the increase in the tilt of the metal surface because of the blurring of the light spot. For this reason the lowest accuracy is obtained nearby the crucible walls. y 

 ac  ctg    bc  L ; x  y  ctg   L  c  ctg  2

where: ∆x ,∆y– measurement errors of coordinates  – the tilt of the laser L – the position of the laser beam recorded by the camera sensor a,b,c – geometric values (fig. 1) ∆L – the accuracy of the camera sensor

green laser

head with inteference grid

Figure 5:

camera lens

interference filter crucible coil

Experimental setup.


(4)


4

177

Conclusion

The method of laser measurement of the surface profile may be successfully applied to record the surface of a liquid metal. The highest accuracy of results is obtained by a variant of this method in which light diffused on the metal surface is recorded. This technique avoids the problems with the influence of the surface curvature on the change in the direction of laser beam, which is why it was applied in our experimental setup. The measurements performed with the use of the method incorporating the reflected beam allows us to calculate the curvature of the liquid metal on which the laser beam is not diffused; however, the disadvantage is the complexity and lower accuracy of the measurement. Further studies on the issues discussed in this paper should be devoted to the expansion of the method so that the whole surface of the metal could be scanned, which would result in its application to the problems without axial symmetry.

Acknowledgement This research work was carried out within project No. N508 034 31/1889, financially sponsored by the Polish Ministry of Science and Higher Education.

References [1] Blacha L., Fornalczyk A., Przyłucki R., Golak S.: Kinetics of the evaporation process of the volatile component in induction stirred melts, 2nd International Conference Simulation and Modelling of Metallurgical Processes in Steelmaking STEELSIM 2007, Graz, Austria, pp. 389-395, 2007 [2] Golak S. , Przyłucki R.: Oxidation of the surface of a liquid metal in the induction furnaces., Acta Metallurgica Slovaca 13, pp. 256-259, 2007 [3] Golak S. , Przyłucki R. : The optimization of an inductor position for minimization of a liquid metal free surface, Electrotechnical Review, 11/2008, SIGMA-NOT, pp. 163–164, 2008 [4] Mass H. G., Hentschel B., Schreiber F.: An optical triangulation method for height measurements on water surfaces, Videometrics VIII (Electronic Imaging 2003), Ed. S. El Hakim, SPIE Proceedings Series Vol. 5013, pp. 103-109. 2003




179

Damage assessment by automated quantitative image analysis – a risky undertaking P. Stroeven Faculty of Civil Engineering and Geosciences, Delft University of Technology, The Netherlands

Abstract This paper presents the economic concept of describing damage in concrete by a partially linear planar system. The practical cases are elaborated of prevailing compressive or tensile stresses. They only require quantitatively analyzing the image patterns of vertical sections by sweeping test lines. Further it is demonstrated that automation of quantitative image analysis generally yields biased information. Keywords: concrete, image analysis, automation, sweeping test line, vertical section.

1

Introduction

Concretes are undergoing a process of degradation during the lifetime of the engineering structure. The degree of degradation can be reflected by characteristics of the internal damage structure. Nature of this problem is three dimensional (3D). To get access to the relevant 3D information would require a random set of section images, which is a laborious and time consuming operation. So, most investigations are of 2D nature only. A method introduced by the author renders possible reducing such efforts tremendously [1–3]. It assumes the crack structure for the most general case composed of 3D, 2D and 1D portion. The method is therefore discussed for the practical case of concrete under prevailing compressive stresses and prevailing tensile stresses, respectively. Damage characteristics under such conditions can be assessed on a single so-called vertical section. A parallel set of such sections can be necessary of course to reduce scatter to acceptable proportions. However, the efforts for sawing and quantitative image analysis are obviously dramatically reduced. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090171

180 Computational Methods and Experimental Measurements XIV The images of such vertical sections can be analyzed by the direct secants approach [4–9], whereby the number of intersections is determined per unit of line length for a stepwise rotating grid of parallel lines. Information can be obtained on total crack length per unit of area and on degree and direction of crack orientation. This is 2D information; however it is readily possible without investment of extra labour to assess the 3D specific crack surface area, and spatial orientation distribution of the cracks. This will be elaborated for the aforementioned practical cases. Due to the repetitive character of such investigations, one would be tempted to use an automated set up. However, the paper will demonstrate information generally to be seriously biased [10, 11]. This will be accomplished mathematically as well as by a visualization method proposed by Underwood [12]. This approach gives detailed insight into the level of bias as a function of conditions, such as the degree of prevailing orientation in the damage structure. Since damage is a fractal phenomenon, the obtained results will fundamentally be a function of magnification of the images. This allows performing a comparative study only.

2

Damage assessment

The damage structure according to the Stroeven concept [13] is denoted as a partially linear-planar structure. The 3D portion encompasses small flat crack element dispersed isotropic uniformly random (IUR) in space. In the 2D portion only small crack elements are collected that are parallel to an orientation plane, however otherwise they are “randomly” distributed. The 1D portion unites crack elements all parallel to an orientation axis, however otherwise “randomly” dispersed. When the 2D portion can be neglected, a so-called partially linearly oriented system is obtained. This model can be used in situations where compressive stresses are prevailing. Alternatively, for high tensile stresses the partially planar oriented damage model can be employed in which the 1D portion is neglected. Damage can be seen as surfaces distributed in space (representing the two crack surfaces at very small distances). Crack density is commonly expressed in total surface area, S, per unit of volume, V. So, leading descriptor of the damage structure is SV (in mm-1). Alternatively, in 2D the total crack length, L, per unit of area, A, yields information on LA. Measurements are made by superposition of line grids on images, of which fig. 1 (left) reveals only a small part (pore is visible at the bottom). Contrast was improved by applying a fluorescent spray. This author has extensively used this method of directed secants in the past 30 years. Incidentally, also other researchers in concrete technology used the very method [14–17]. Fig. 1 (right) shows grid orientations on a full-size hand-made copy of section image. Larger aggregate grains are visible as crack-free areas. 2.1 Concrete in compression Uniaxial compressive stresses produce predominantly cracks that are parallel to the stress direction. In a more general set up, we assume a portion of cracks WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


Figure 1:

181

Load-induced cracks in section plane visualized by fluorescent spray (left), and application of method of directed secants for assessment of intersection densities (right).

distributed isotropic uniformly random (IUR); this is denoted as the SV3 component. The resulting portion consists of cracks parallel to the orientation axis, denoted as the SV1 component. Total crack density is the summation of both components: SV 1 + SV 3 = SV . The proper approach (in technical as well as economic terms) is sampling by vertical sections. Hence, the specimen should be cut to yield one or more image planes parallel to the orientation axis. Such section images can provide the 3D information on SV. Averaging over more vertical images reduces the scatter around the average, and thus the reliability of the results. The results are unbiased, which means that averaging over an increasing number of images will bring the average closer and closer to the population value we are interested in. The analysis of the images is accomplished by line scanning. A grid of parallel lines is superimposed on the crack pattern, successively in the direction of the orientation axis (indicated by index & ) and perpendicular to it (indicated by index ⊥ ), as shown in fig. 1. The following relationships can be derived PL & =

1 2

SV 3

and

PL ⊥ =

1 2

SV 3 +

2

π

SV 1


(1)

182 Computational Methods and Experimental Measurements XIV Hence, crack density is obtained by simple mathematical manipulations, yielding SV =

π

PL ⊥ + (2 −

π

) PL & (2) 2 2 P in eqs. (1) and (2) stands for the number of intersections of grid lines and cracks. The constants account for probabilities that cracks appear in the section image [6,7].

2.2 Concrete in tension The methodology is very similar. The vertical section is again parallel to the tensile stresses. The grid is also successively superimposed in the stress direction and perpendicular to it, with the same indices accounting for the position of the grid. In this case we have 1 1 PL & = SV 3 + SV 2 and PL ⊥ = SV 3 (3) 2 2 Again, simple manipulation will yield SV = SV 2 + SV 3 = PL & + PL ⊥

(4)

Here, SV 2 stands for the portion of cracks perpendicular to the tensile stresses, with SV = SV 2 + SV 3 . To accomplish such operations, the contrast should be improved by ink penetration or by application of a fluorescent spray (applied in the case of fig. 1) or dye. Details can be found in the relevant literature [5,8].

3

Biases due to automated set up

3.1 Analogue images An elegant way to reveal differences in outcomes of quantitative image analysis approaches by directs secants to analogue and digitised images is to make use of the earlier mentioned Stroeven concept. Hence, LA is assumed consisting of two portions, a “random” one, denoted by LAr , and a fully oriented one, indicated by LAo . The latter “sticks” (short straight elements as part of the 2D crack) run parallel to the orientation axis that supposedly makes an angle β with the positive x-axis. This strategy allows dealing with both portions separately. The rose of intersections per unit of grid line length (intersection densities) of the random portion approximates (for very large images) a circle around the origin with radius PLr . The rose of intersection densities for the oriented portion approximates a circle through the origin with PLo ( β ) = 0 and PLo (ς ) = PLo max . Note that

β = ς + π 2 . PLo (θ ) = PLo max cos(θ − ς ) for an arbitrary angle θ.

When combined, the rose of intersection densities is obtained for a partially linear structure of lineal features in a plane, shown in fig. 3 [10]. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


183

Figure 2:

Rose of intersection densities for oriented (left) and random (right) line segments in a plane (together forming the 2D crack pattern) for an analogue image.

Figure 3:

Rose of intersection densities for random (dashed line) plus oriented line segments in a plane (continuous line; small circles) shown in fig. 2. Crack pattern is supposed to encompass only relatively small lineal portion (so-called weakly oriented pattern).

3.2 Digitized images The smooth contours of the cracks can be conceived in conventionally digitized images replaced by two orthogonal sets of mono-size sticks as shown in fig. 4. As before, a distinction can be made between the “random” portion and the oriented one running parallel to an orientation axis enclosing an angle β with the positive x-axis. As before, β = ς + π 2 . The random portion consists of two equally large sub-sets of sticks oriented in the respective coordinate directions {x,z}. This leads to two equally large roses of intersection densities that run through the origin and are orthogonally oriented. Circle diameter is PLr . The summation yields a symmetric flower-like rose displayed in fig. 5 at the bottom. In an arbitrary direction, the intersection density is given by


184 Computational Methods and Experimental Measurements XIV PLr (θ ) = PLr (sin θ + cos θ ) =

2 PLr cos(θ −

π 4

)

(5)

A striking but expected observation is the preferred orientation in a direction enclosing an angle π/4 with the positive x-axis; the random portion is reflected significantly biased by a digitised image with maximum value 2 PLr . The projected portions of the oriented fraction of the cracks in the x- and zdirections are LA (0) = LAo sin ς and LA (π 2) = LAo cos ς , respectively. This can '

'

be transformed to intersection counts per unit of grid line length, whereby the horizontal system of oriented sticks leads to two circles sharing the origin of the polar system and having the z-axis as symmetry line. The intersection with the zaxis is PLo sin ς . The second system of oriented sticks parallel to the z-axis is modelled in an analogous way by two circles sharing the origin of the polar system and having the x-axis as symmetry line. The intersection with x-axis is PLo cos ς . The two pairs of circles and their summation is displayed in fig. 5 (top). The intersection density in an arbitrary direction of the oriented portion is PLo (θ ) = PLo sin θ sin ς + PLo cos θ cos ς = PLo cos(θ − ς )

(6)

The direction of the principal axis of the rose is found for zero value of the first derivative of eqn. (6). This occurs for θ = ς , whereby PLo (ς ) = PLo .

Figure 4:

Detail of conventionally digitized section image of particulate structure of which the surfaces like cracks are dispersed in 3D space. Smooth surface contours are replaced by orthogonal set of straight line segments.



Figure 5:

185

Rose of PL (θ ) values for oriented (top) and random (bottom) line segments in digitized image of crack pattern.

The roses for the partially oriented lineal crack features are obtain by summation of the total roses at the top and bottom of fig. 5. The mathematical expression for this system is obtained upon adding up Eqs. (5) and (6). In both cases it becomes obvious that the direction of preferred orientation, β = ς + π 2 , is only reflected by the experimental data when the signal is very strong (i.e., PLo / Plr 1 ); when weak, the preferred orientation direction will be very close to π/4. Intermediate situations will be biased to an unknown degree when used for predicting the direction of preferred orientation.

4

Discussion

A more detailed insight into differences between quantitative image analysis outcomes obtained on either analogue or digitized images can be achieved by superimposing the relevant estimates assessed on the different images for the fully oriented and random portions separately. Fig. 6 (top) presents the fully oriented portions as reflected by the two approaches. In the first and third quadrant the solutions are identical, whereas they are distinctly different in the other two quadrants. Digitized images only offer correct information along the axes of (four-connexity) digitization (fig. 6: bottom). All other measurements WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

186 Computational Methods and Experimental Measurements XIV will be biased to an unknown degree. Only when confronted with very strong signals one might successfully assess such geometric parameters by an automated approach to digitized images. By equating the first derivative to zero of the aforementioned expression for the analogue and digitized images, this is confirmed by the respective equations that are obtained (in the first quadrant): PLo sin(θ − ς ) = 0 and PLo sin(θ − ς ) +

2 PLr sin(θ − π 4) = 0 , for analogue and

digitized images, respectively. The correct solution ( θ = ς ) is only for the analogue image. Estimation of total length of lineal features in a plane, LA, is based on π π π π π L = PL (θ ) = ∫ [ PLo cos(θ − ς ) + 2 PLr cos(θ − )]dθ / ∫ dθ . This holds for 2 2 0 4 0 analogue as well as for digitized images. The respective solutions are π π and LA = 2 cos(ς − ) PLo + 2 PLr (digit.) (7) LA = PLr + PLo (anal.) 4 2 LA as obtained from digitized image is always biased (i.e., overestimated); even for very strong signals, the ratio of digitized to analogue image information A

Figure 6: Digitization-induced biases for oriented (top) and random portion (bottom) of rose of intersection densities. Dashed line represents the analogue image, the solid line the digitized image subjected to the automated model. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


is

187

2 cos(ς − π 4) . For very weak signals this yields 4 / π . For mixed

2 and 4 / π , with even an situations, the bias can be anywhere between influence of the direction of preferred orientation. A parameter used to characterize the strength of preferred orientation is the degree of orientation, ω. This parameter can be derived for a space system (ω3), as well as for a planar one (ω2) by the information in this paper. As an example, for ω2 this leads to PLo .100 100 (%) (analogue) (8) ω2 = = PLr 1.572 PLr + PLo +1 1.572 PLo 1+

ω2 =

PLr PLo

[1 − 1.73 cos(ς +

PLr

π

π 4

)] 100 (%) (digitized)

(9)

[1 + cos(ς + )] 4 PLo The correct signal for the analogue image declines smoothly from strong to weak signals from 100% to 0%. For the digitized image it declines also from 100% for strong signals to anywhere between 100% and 0%, depending on the orientation direction in the image field, for weak signals. So, information will generally be biased to an unknown degree. 1+

5

Conclusions

Spatial damage structures can be analyzed by application of quantitative image analysis by sweeping test line system on orthogonal section images only. For the practical cases of prevailing compressive or tensile loadings, (a set of) vertical sections will do, restricting dramatically efforts required for preparation of samples and image analysis. The choice to automate the quantitative image analysis operation is a risky one, because characteristic measures for the damage structure, like total crack length (or specific crack surface area) and degree and direction of prevailing crack orientation will be seriously biased.

References [1]

[2]

Stroeven, P., Shah, S.P., Use of radiography-image analysis for steel fiber reinforced concrete, Testing and Test Methods of Fiber Cement Composites, ed. R.N. Swamy, Constr. Press: Lancaster, pp. 345-353, 1978. Stroeven, P., de Haan, Y.M., Structural investigations on steel fiber reinforced concrete, High Performance Reinforced Cement Composites, eds. H.W. Reinhardt, A.E. Naaman, E & FN Spon: London, pp. 407-418, 1992. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

188 Computational Methods and Experimental Measurements XIV [3]

[4] [5]

[6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17]

Stroeven P., Damage evolution in concrete; application of stereology to quantitative image analysis and modelling, Advanced Materials for Future Industries: Needs and Seeds, eds. Kimpara I, Kageyama K, Kagawa Y, SAMPE: Tokyo, pp. 1436-1443, 1991. Nemati, K.M., Stroeven, P., Stereological analysis of micro-mechanical behaviour of concrete. Mat. Struct. 34, pp. 486-494, 2000. Reinhardt, H.W., Stroeven, P., den Uijl, J.A., Kooistra, T.R., Vrencken, J.H.A.M., Einfluss von Schwingbreite, Belastungshöhe und Frequenz auf die Schwingfestigkeit von Beton bei niedrigen Bruchlastwechselzahlen. Betonw. & Fertigteil-Techn., 44, pp. 498-503, 1978. Stroeven, P., Hu, J., Gradient structures in cementitious materials, Cem. Concr. Comp., 29, pp. 313-323, 2007 Stroeven, P., Hu, J., Stereology: Historical perspective and applicability to concrete technology, Mat. Struct., 39, pp.127-135, 2005. Stroeven, P., Some observations on microcracking in concrete subjected to various loading regimes, Engr. Fract. Mech., 35(4/5), pp.775-782, 1990. Stroeven P., Geometric probability approach to the examination of microcracking in plain concrete, J. Mat. Sc. 14, pp.1141-1151, 1979. Stroeven, P., Stroeven, A.P., Dalhuisen, D.H., Image Analysis of ‘natural’ concrete samples by automated and manual procedures, Cem. Concr. Comp., 23, pp. 227-236, 2001. Chaix, J.M., Grillon, F., On the rose of direction measurements on the discrete grid of an automatic image analyser, J. Microsc., 184, pp. 208213,1996. Underwood E.E., Quantitative Stereology, Addison-Wesley: Reading (MA), 1970. Stroeven, P., Structural modelling of plain and fibre reinforced concrete, J. Comp., 13, pp. 129-139, 1982. Stang, H., Mobasher, B., Shah, S.P., Quantitative damage characterization in polypropylene fibre reinforced concrete, Cem. Concr. Res., 20, pp. 540558, 1990. Carcassès, M., Ollivier, J.P., Ringot, E., Analysis of microcracking in concrete, Acta Stereol., 8(2), pp. 307-312, 1989. Ringot, E., Automatic quantification of microcracks network by stereological method of total projections in mortars and concrete, Cem. Concr. Res., 18, pp. 35-43,1988. Nemati, K.M., Generation and interaction of compressive stress-induced microcracks in concrete, PhD Thesis, University of California: Berkeley, 1994.


Section 4 Detection and signal processing Special session chaired by A. Kawalec



191

Antenna radiation patterns indication on the basic measurement of field radiation in the near zone M. Wnuk Faculty of Electronics, Military University of Technology, Poland

Abstract The paper presents the method of antennas radiation patterns measurements in the near zone. Afterwards, using the analytical methods, the measured data is transformed into radiations patterns in the far field. This technique is used in measurements in the closed areas, which enable researchers to manage the environmental characteristics. The calculated radiation patterns are as precise as the range measurements in the far field. It needs to be outlined however that this method requires more complex and expensive regulation procedures as well as more sophisticated software, whereas the radiation patterns are not obtained in the real time. Keywords: antennas radiation, near zone, far field.

1

Introduction

An antenna is one of the important components of a radio communication system. It is designed to convert the input current into an electromagnetic field and to emit it into the surrounding space (transmitting antenna) to the contrary (receiving antenna). Therefore, the antenna is a device which adjusts the waveguide to free space. Due to its location between a transmitting or receiving device and the space, requirements set forth for an antenna are imposed both by the conditions of expansion of electromagnetic waves in space and by interaction of the antenna as an element of the given device on its operation [1]. Its parameters and patterns affect not only effective information transfer but also meeting the compatibility conditions, i.e. the antenna should not disturb operation of other systems, particularly along the lateral lobe radiation lines. That is why during the recent period the antenna measurement technique WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090181

192 Computational Methods and Experimental Measurements XIV develops very rapidly. It must ensure high accuracy of measurements, which must be taken at a receding level of signal. This results from the necessity to measure lateral lobes at below -40 dB. In order to meet these requirements the measurements should be taken in special conditions, which are ensured in anechoic chambers.

2

Antenna radiation zones

The area surrounding the antenna may be typically split into three zones of the electromagnetic field generated by that antenna: the reactive near-field, the radiating near-field, the far-field (Figure 1). The far-field expands to infinity and it is that area in space in which the electromagnetic field changes with distance r from the transmitting antenna according to the exp(-jkr)/r relationship, where k = 2π/λ, λ – wave length. It is assumed that the far-field expands between the distance from the studied antenna to infinity, where D is the largest geometrical dimension of the studied antenna.  2D 2   + λ R g =   λ 

(1)

Factor λ added to the 2D2/λ relationship covers the case in which the maximum geometrical dimension of the antenna is smaller than wave length λ. The space area stretching between the antenna and the conventional limit of the far-field is called the near-field, with the reactive near-field expanding between the studied antenna and distance λ. In the reactive near-field the electric and magnetic field phases are almost in quadrature, the Poynting vector is of a complex nature. The imaginary part of the Poynting vector is responsible for collecting the energy of the electromagnetic field near the antenna surface, and the real part relates to energy emitted by the antenna. Polenear-field indukcji Reactive Strefa bliska Near field

Strefa daleka

Far field

Rozkład pola

Field distribution

Odległość Distance λ

Figure 1:

2D /λ

Zones around the antenna.

At a distance greater than λ from the antenna the electromagnetic field has a complex nature and changes significantly in the function of distance from the WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


193

antenna, it is the so-called radiating near-field. Antenna measurements in the near-field are usually taken in the radiating near-field. Sometimes areas of the electromagnetic field around the antenna are named with terms borrowed from optics, such as the Fresnel zone and Fraunhofer zone. The Fraunhofer zone is a term equivalent with the far-field, whereas the Fresnel zone expands from distance to the conventional limit of the far-field [1]. 1

 D 3  D  Rd =     + λ  2λ   2 

(2)

Traditional methods require the measurements to be taken in the far-field. It is difficult to meet this requirement in case of antennas operating in the micro wave range. For instance, the radius of this zone, for an L=3m antenna with working frequency of f=9 GHz, expands to a distance of over 540m. It should be noted that the size of the largest anechoic chambers does not exceed 50m. Hence the need to develop such methods which allow reduction of the size of the measured structure do dimensions suitable for confined spaces such as an anechoic chamber. This problem may be solved by taking measurements in the near-field the socalled (4 ÷10)λ distance. There are many methods of measurements in the near-field. Three of them are generally used. These are: the planar method (Figure 2(a)), the cylindrical method (Figure 2(b)) and the spherical method (Figure 2(c)). Each of the methods has both advantages and disadvantages. Spherical scanning requires larger anechoic chambers, as compared with the remaining methods. Cylindrical scanning proves excellently for measurement of area monitoring radars. Planar scanning is limited by the angular sector which allows measurement of the main beam and the nearest lateral lobes. The main advantage of planar scanning consists in dense and uniform distribution of sampling points in the grid. In the case of scanning polar (uniaxial) and bipolar (biaxial) plane it will be possible to obtain a scanning plane larger than that offered by the anechoic chamber.

(a) spherical method Figure 2:

(b) cylindrical method

(c) rectangular planar method

View of setups for antenna measurements in the near-field.

In the methods of antenna measurement in the near-field, the values of the electromagnetic field are measured at discrete points in a preset surface. Sampling points are in nodes of suitably defined grid, inscribed on this surface. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

194 Computational Methods and Experimental Measurements XIV Three types of sampling point grids are applied; they are presented in Fig. 3. Excessively dense sampling is not necessary for accurate presentation of the electromagnetic field. In practice, a ca 10-20% redundancy coefficient is introduced. Redundant density of sample occurrence is mostly affected by the pattern in which the lines with sampling points converge into a single central point (3(b)), (3(c)). Such patterns of sampling point grids are often used in practice due to the advantageous kinematic systems of the scanner. ∆x

∆y

b

(m∆x, n∆y, zt)

a

(a) Rectangular grid

(b) Uniaxial planar grid Figure 3:

(c) Biaxial planar grid

Measurement grids.

The radiating area of the near-field expands between the distance equal to λ wave length from the antenna and the distance determined by formula (1). Beyond this distance we have the far-field, where angular distribution of energy does not oscillate with the distance and the radiating power disappears with the distance. Dimension of the measurement area is important, because we are considering accuracy of the planar measurement technique in the near-field. Sampling plane

Antenna aperture plane

Figure 4:

Clarification of the size of angular sector of the area of importance of the measured pattern.

The size and location of the measurement area define the value of the angular sector of the area of importance. The size of this angular sector depends, i.a., on the size of scanning surface of the electromagnetic field and on the distance of this surface from the aperture of the tested antenna [7]. The computed radiation pattern in the far-field will be precise in the ±ΘS area. a − D  Θ S = arctg   2zt  WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

(3)


195

Total angular coverage may be obtained in the spherical system by adding measurements in the near-field along the whole spherical surface of the nearfield Regardless of the chosen measurement method, equipment for carrying out specific measurements is similar. The differences are primarily caused by: arrangement of different measuring instruments in respect of the source of the measurement signal and the tested antenna, the type of measurements to be taken and the required level of automation of measurements. The equipment required for taking measurements of antenna patterns in the near-field consists of 4 major sub-systems which may be controlled from one, central control panel. These are: • positioning and control subsystem, • receiving subsystem, • signal source subsystem, • measurement data saving and processing subsystem. It should be emphasized that while taking measurements in the near-field the results obtained should be transformed, with the use of analytical methods, into data suitable for computing radiation patterns in the far-field. The computed radiation patterns are as accurate as measurement of range in the far-field. Depending on the required accuracy it is necessary to use more complex and expensive regulatory procedures and more complicated software and radiation patterns are not obtained in real time.

3

Theoretical basis for determination of antenna pattern based on the near-field measurement

Modern planar scanning techniques in measurements of the antenna near-field are based on representation of the field in form of planar wave spectrum. Electromagnetic waves with a given frequency may be represented as a superposition of elementary planar waves of the same frequency. Further considerations shall be based on a rectangular x, y, z coordinate system (Figure 5). In the passive and lossless area of free space, Maxwell equations describing the phenomenon of electromagnetic wave propagation, may be transformed into homogeneous second order Helmholtz equations [9],

G G ∇2E + k 2E = 0

G G ∇ H + k 2H = 0 G G ∇⋅E =∇⋅H =0 2

(4) (5) (6)

On assumption that observations of components of the vector of electric and magnetic field, for a wave sinusoidal variable in time, were carried out at the same moment t = tp in all studied points of space, the segments dependent on independent variable t representing time, were abandoned in the above equations. Due to linearity of the said operators and linearity of the medium in which the described phenomenon of electromagnetic wave propagation takes place, it is WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

196 Computational Methods and Experimental Measurements XIV fairly easy to prove that the equations below satisfy the set of equations (4), (5), (6) and the threshold requirements in plane z = 0, +∞ +∞ G GG G (7) E ( x, y , z ) = A(k , k )exp(− jk r )dk dk ,

∫∫

x

y

x

y

− ∞− ∞

(

)

+∞ +∞ G GG G G H (x, y, z ) = ∫ ∫ k × A(k x , k y )exp − jk r dk x dk y , − ∞− ∞

(8)

G G k ⋅ A(k x , k y ) = 0 .

(9) G G G G where: k = k x i x + k y i y + k z i z - wave vector, indicates the direction of propagation of the wave described by wave equations (4), (5), (6), G G - wave number (it is the length of the wave vector), k2 = k ⋅k G G G G r = xi x + yi y + zi z - vector indicating to observation point, G G G G A(k x , k y ) = Ax (k x , k y )i x + A y (k x , k y )i y + Az (k x , k y )i z - wave vector describing

the planar wave spectrum. z

(xt,y,z) r

X

(xt,o,o) Θ

ϕ Tested antenna

(0,0,0)

y

Sampling plane

Figure 5:

Location of the antenna in the reference arrangement adopted in the analysis G G The integrand A(k x , k y )exp − jk rG occurring in relationships (7), (9) represents the homogeneous planar wave propagating along the direction G determined by vector k therefore a monochrome wave emitted through the aperture may be recorded as superposition of planar waves with the same frequency, different amplitudes and expanding in different directions. Equation (9) in turn, which is a natural consequence of the Gauss law for a

(

)



197

passive area expressed in form of equation (6), allows distinguishing two independent components (here Ax (k x , k y ) i A y (k x , k y ) ) of vector AG (k x , k y ) Az (k x , k y ) = −

1 (Ax (k x , k y )k x + Ay (k x , k y )k y ). kz

(10)

In order to determine the value of the electric field for an aperture located in the far-field, the following relationship was obtained with the use of expression (7),  Ax (k x , k y ) , G j exp(− jkr ) G j exp(− jkr )  E (r , θ , ϕ ) = k z A(k x , k y ) = k z  A y (k x , k y ) 2π r 2π r  A (k , k )  z x y 

where:

(11)

k x = k sin θ cos ϕ , k y = k sin θ sin ϕ , k z = k cos θ ,

Az (k x , k y ) is expressed by (10).

The necessity to determine components Eθ = (r , θ , ϕ ) and Eϕ = (r , θ , ϕ ) of the far-field determined in spherical coordinates implies carrying out transformations which produce the relationship (11) in respective form: G jk exp(− jkr ) ((Ax (k x , k y )cos ϕ + Ay (k x , k y )sin ϕ )iθ + E (r , θ , ϕ ) = 2π r + cos θ (Ay cos ϕ − Ax sin ϕ )iϕ ) = Eθ (k x , k y )iθ + Eϕ (k x , k y )iϕ

(12)

In the next step, depending on the method of polarisation of the tested aperture, we determine co-polarisation E co (θ , ϕ ) and cross-polarisation

E cross (θ , ϕ ) patterns.

Polarization of antenna E x :

E co (θ , ϕ ) = Eθ (θ , ϕ ) cos ϕ − Eϕ (θ , ϕ ) sin ϕ =

(

)

= Ax (k x , k y ) cos 2 ϕ + sin 2 ϕ cos θ + Ay (k x , k y )sin ϕ cos ϕ (1 − cos θ ) E cross (θ , ϕ ) = Eθ (θ , ϕ ) sin ϕ − Eϕ (θ , ϕ ) cos ϕ =

(

= Ax (k x , k y )sin ϕ cos ϕ (1 − cos θ ) + Ay (k x , k y ) cos 2 ϕ + cos θ cos 2 ϕ

plane E (ϕ = 0) :

E co (θ , ϕ ) = Ax (k x , k y ) ,

)

(13) (14)

(15)

E cross (θ , ϕ ) = Ax (k x , k y )cos θ ,

(16)

E co (θ , ϕ ) = Ax (k x , k y )cos θ

(17)

plane H  ϕ = π  : 

2

E cross (θ , ϕ ) = Ax (k x , k y ) .

(18)

In order to determine the far-field pattern it is necessary to know the components G Ax (k x , k y ) and A y (k x , k y ) of the planar wave spectrum vector A(k x , k y ) . In the WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

198 Computational Methods and Experimental Measurements XIV case of observation of the electric field vector in the z = z t plane, the vector equation (7) adopts the following form: E x ( x, y , z = z t ) =

+∞ +∞

∫ ∫ [A (k x

x

, k y )exp(− jk z z t )]exp(− jk z x ) exp(− jk z y )dk x dk y , (19)

− ∞− ∞

E y ( x, y , z = z t ) =

+∞ +∞

∫ ∫ [A (k y

x

, k y )exp(− jk z z t )] exp(− jk z x ) exp(− jk z y )dk x dk y ,

(20)

x

, k y )exp(− jk z z t )]exp(− jk z x ) exp(− jk z y )dk x dk y ,

(21)

− ∞− ∞

E z ( x, y , z = z t ) =

+∞ +∞

∫ ∫ [A (k z

− ∞− ∞

where:

(22) Selection of sample spacing allows obtaining such equations for Fourier integrals in which the structure is actually a modified version of two-dimensional discrete Fourier transform or inverse. Because of the considerable number of sampling points acquired during scanning the measurement plane, the choice of effective numerical algorithms for data processing becomes essential. A good example here is offered by algorithms of Fourier fast transform (FFT) and inverse Fourier fast transform (IFFT), which may be applied to determine transforms based on line/column type decomposition. For a finite number of observation points (2N sampling points) and the rectilinear domain (s ∈ (− ∆s, ∆s )) the expression may be as follows:

~ F (s ) =

G sin w(s − n∆s ) ψ (s − n∆s )F (n∆s ) . w(s − n∆s ) n = − N +1 N

∑

(23)

The approximating function ψ (s ) is designed to ensure quick convergence of the approximation error with the growing value of the rate of oversampling π / w ( π / w is the maximum admissible spacing – Nyquist spacing – χ= ∆s

resulting from the sampling theorem) and minimisation of the so-called truncation error resulting from the finite size of the measurement grid. On the other hand, in the case of occurrence of approximation of angular variable domain ϕ (ϕ ∈ (− ∆ϕ , ∆ϕ )) in the task, the following rule should be applied: M G ~ (24) F (ϕ ) = ∑ DM n (ϕ − m∆ϕ )Ω M r (ϕ − m∆ϕ )F (m∆ϕ ) , m = − M +1

where:

ϕ  sin (2 M n + 1)  2  - Dirichlet function,  D M n (ϕ ) = (2M n + 1) sin (ϕ / 2)



199

Ω M r (ϕ ,0) plays the role of a function which reduces the truncation error

value. In view of the fact that this paper describes the application of planar scanning, errors for this case shall not be analysed.

4

Principles of field sampling in the near-field

In this paper the antenna was tested with the use of the planar method with a rectangular grid of sample points. This choice was based upon its major advantages, such as: low cost of scanning mechanism, the smallest amount of computations and stationary tested antenna. Data acquisition in a planar near-field is done over a rectangular x-y grid, Figure 3(a), with maximum sample spacing in the near-field (25) ∆x = ∆y = λ 2 The measurement procedure requires that the zt plane surface at a distance from the tested antenna be selected where the measurements are taken. The zt distance should be located at a distance of at least two or three wavelengths between the tested antenna and the near-field interaction limit. The plane in which the measurements are taken is split into a rectangular grid with M x N points spaced ∆x and ∆y apart and defined by coordinates (m∆x n∆y, zt), where: M M N N (26) − ≤m≤ −1 and − ≤ n ≤ −1 2

2

2

2

Values M and N are determined by linear dimensions of sampling plane divided by sampling spaces. Measurements are taken till the time when the signal at the plane edges reaches the level of -40dB below the highest level of the signal inside the measured plane. Defining a and b as the width and height for the measured plane, M and N are determined by the expressions: a b (27) M = +1 and N= +1 ∆x

∆y

The selected sampling spaces in the measurement grid should be smaller than half the wave length and should meet the Nyquist sampling criterion. If plane z = zt is located in the far-field of the source, sampling spaces may grow to its maximum value λ/2. Points of the rectangular grid are spaced by grid spacing, as: π π (28) ∆x =

k xo

and

∆y =

k yo

where k xo and k yo are real numbers and are the largest dimensions k x and k y respectively, so that f (k x , k y ) ≅ 0 for

k x > k xo or k y > k yo .



5

Verification of the experiment

Determination of the radiation pattern of the antenna based on data from measurements taken in the near-field requires application of an advanced mathematical tool. Because of complicated computing techniques, large volumes of input data, the requirement of graphical interpretation of computation results, the Matlab 6.5 programme was used. Two computer programmes were developed. The first of them determines theoretical distributions of the electric field on the surface of the antenna aperture, as well as distributions of intensity and phase of the electric field in a plane in parallel with the aperture plane at zt distance from it. The second programme determines the cross-section of the radiation pattern of the tested parabolic antenna based on data from measurements or theoretical data determined by the first programme. In order to check the correctness of the developed concept for measurement setup concept antenna measurements were carried out with a symmetrical dish reflector with diameter D = 0.6m. The radiant element used was in form of halfwave dipole, combined with a circular convergent mirror. In this case the far-field will occur at the distance of 7.2 m. The size of the available anechoic chamber allow measurements at a maximum distance of 5 m. The figures below present measurements of the amplitude and electromagnetic field phase pattern of the tested antenna, measured in the near-field. They were taken in an anechoic chamber. The amplitude and signal phase were measured with the use of vector analyzer HP HP8530 with accuracy to two decimal places. The results of measurements were standardized. The reference was adopted as the 0dB signal level. The phase difference was computed by a frequency converter based on the signal received from the antenna to the reference signal ratio. Figures 6 and 7 present the measured distributions of electric field amplitudes in the scanning plane for component E x . Ex [dB]

Kierunek próbkowania y sampling direction y

Ex [ dB ]

Kiey rusa n em kppli rn óbg kodir wec a n tio ia n y

tioxn

ec ia owan g kdir prób mplin runek Kie x sa

Kierunek próbkowania x x sampling direction

Figure 6:

Spatial standardized amplitude pattern of the tested antenna in the near-field.

Figure 7: Standardized amplitude of electric field intensity.

Whereas Figure 8 presents the measured distribution of the electric field phase in the scanning plane. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


201

The kierunkowa programme was used for determination of the following patterns: theoretical (based on data generated by the pole_bliskie programme) and that obtained from computations (based on formula (11). Both patterns are presented in one figure – Figure 9. The same figure also additionally presents radiation pattern obtained from measurements. In such diagram comparisons and verification of the experiment may be done. φ [°] 180

100

-180

Kie run ek

Figure 8:

p ró bko wa nia y

ne Kiru

x ania ko w rób kp

Spatial phase pattern of the tested antenna in the near-field.

The patterns have been marked with numbers and colours respectively:  Blue – cross-section of the antenna radiation pattern, determined from measurements – No 1;  Red – cross-section of the antenna radiation pattern computed by the kierunkowa programme based on data received from the antenna nearfield – No 2;  Green – theoretical cross-section of the antenna radiation pattern generated by the pole_bliskie and kierunkowa programmes – No 3. Angel

Measurments pattern Calculations pattern Theoretical pattern

Figure 9:

Radiation patterns of the tested antenna.


202 Computational Methods and Experimental Measurements XIV While comparing the patterns of Figure 9, one may observe contraction of patterns 2 and 3 as compared with 1. This is caused not only by phase disturbances but also by failure to meet the requirement of the far-field during measurements of pattern No 1. The difference between the far-field limit and the distance at which the measurements were taken and which amounts to 2.85m is so significant that it has a direct impact on the pattern form.

6

Conclusions

It was assumed in the to-date analysis that components tangent to the measurement plane of the electric field vector are measured precisely at point. In reality such probe does not exist and the antenna used for measurements has some definite geometrical dimensions. Therefore the values of the amplitude and phase are averaged on its surface. The impact of the probe radiation pattern is also significant. The pattern was further distorted by imprecise, non-automatic scanning setup. Precision of positioning the probe in vertical plane is not satisfactory, nor is the time necessary for completion of measurements. The measurement results obtained confirmed the correctness of adopted design assumptions and correctness of algorithms made. Radiation patterns transformed into the far-field are convergent with the theoretical patterns and the results of comparative measurements.

References [1] R. E. Collin; Prowadzenie fal elektromagnetycznych; WNT Warszawa 1966, [2] HP 8530A Microwave Receiver. Operating and Programming Manual, Edition 2, Hewlett-Packard Company, February 1994. [3] P. Kabacik: Reliable evaluation and property determination of modern-day advanced antennas; Oficyna Wydawnicza Politechniki Wrocławskiej, Wrocław 2004. [4] J. Modelski, E. Jajszczyk, H. Chaciński, P. Majchrzak: Pomiary parametrów anten Oficyna Wydawnicza Politechniki Warszawskiej, Warszawa 2004. [5] Y. Rahmat-Samii, L. I. Williams, and R.G. Yaccarino: „The ULCA bi-polar planar near-field antenna –measurement and diagnostic range”, IEEE. Antennas and Propagat., Magazine., vol. 37, pp. 16-34, December, 1995. [6] H. Trzaska; Pomiary pól elektromagnetycznych w polu bliskim. PWN 1998, [7] W Zieniutycz; Anteny. Podstawy polowe; WKŁ Warszawa 2001. [8] http://www.nearfield.com [9] M. Wnuk; Analiza struktur promieniujących położonych na wielowarstwowym dielektryku, Wojskowa Akademia Techniczna, Warszawa 1999



203

Sub-ppb NOx detection by a cavity enhanced absorption spectroscopy system with blue and infrared diode lasers Z. Bielecki1, M. Leszczynski2, K. Holz2, L. Marona2, J. Mikolajczyk1, M. Nowakowski1, P. Perlin2, B. Rutecka1, T. Stacewicz3 & J. Wojtas1 1

Military University of Technology, Poland Institute High Pressure Physics Unipress, Poland 3 Warsaw University, Poland 2

Abstract This paper presents opportunities of application of cavity enhanced absorption spectroscopy (CEAS) in nitrogen oxide (NOx) detection. The CEAS technique is based on the off-axis arrangement of an optical cavity. In this system, an absorbing gas concentration is determined by measurement of the decay time of a light pulse trapped in an optical cavity. Measurements are not sensitive to laser power fluctuation or photodetector sensitivity fluctuation. In this configuration, the setup includes the resonance optical cavity, build with spherical mirrors of high reflectance. Pulsed lasers are used as the light sources. NOx detection is carried out in the blue and far infrared range. The signal is registered with a newly developed low noise photoreceiver. The features of the designed system show that it is possible to build a portable trace gases sensor. Its sensitivity could be comparable with that of chemical detectors. Such a system has several advantages: relatively low price, small size and weight, low power consumption, and the possibility of the detection of other gases. Keywords: CEAS, NOx sensor.

1

Introduction

Cavity ringdown spectroscopy (CRDS) is a high-sensitivity absorptionmeasurement technique [1–4]. It is based on measurement of the changes in the WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090191

204 Computational Methods and Experimental Measurements XIV relaxation time of a high-finesse optical cavity. The changes depend on the absorbance of the species filling the cavity. In CRDS, intensity measurements are replaced with time measurements, and hence this method is not sensitive to laser intensity fluctuation. This technique is based on the phenomenon of light trapping inside an optical cavity composed of two mirrors characterized by the high reflectivity coefficient (R > 0,99995%). In this method, a pulse of the laser light is injected into an optical cavity (resonator) equipped with spherical and high reflectance mirrors (Figure 1). The pulse yields to multiple reflections in the resonator. After each reflection, a part of the laser light leaves a resonator because of lack of 100% mirrors reflectivity. The part of light leaving the cavity is registered with a photoreceiver. The amplitude of the optical signal decreases exponentially.

Figure 1:

Idea of the CRDS technique.

The speed of the decay intensity of the pulse of the laser light is dependent on the mirrors reflectivity coefficient R, the resonator length L, and extinction, which consists of absorption and scattering of light in the absorber filling the cavity. Therefore, by measuring the resonator quality, determination of the extinction coefficient  is possible [2]. The resonator quality can be determined with measurement of the radiation decay time constant τ τ=

L , c[( 1  R) + αL]

(1)

where c is the light speed [3]. The decay time τ is measured once when the cavity is empty ( = 0), and then when the cavity is filled with the absorber ( > 0). By comparison of the decay times for these two cases, and assuming that the absorption dominates, value of the absorber concentration N can be found N=

1 1 1    , cσ  τ τ 0 

(2)

where σ is the absorption cross section, τ0, τ are the time constants of the exponential decay of the output signal for the empty resonator and for the resonator filled with the absorber, respectively [4]. Assuming that the relative precision of τ determination is X=

τ0  τ , τ0


(3)


205

the detectable concentration limit NL is given by NL =

X . cσσ0

(4)

High sensitivity of the absorption measurement is achieved due to increasing the effective optical path length up to several kilometers in a very small volume of the optical cavity. Pulsed or AM-modulated cw lasers are used. However, in order to avoid multimode excitation of the cavity (and multiexponential changes of registered light) CRDS setup requires spatial mode filtering and matching of the laser light with the resonator. In order to much the laser wavelength to the mirror reflectivity spectrum and to avoid broadband pulses of fluorescence from the laser a spectral filtering must be applied. This demand causes some disadvantages. The optical system becomes complicated and requires rigorous vibration isolation as well as temperature stability to operate with high sensitivity. These limitations pose engineering challenges for field deployment. Subsequently, several modifications of CRDS have been developed [5–7]. One of them is cavity enhanced absorption spectroscopy (CEAS). It is based on off-axis arrangement of the laser and optical cavity. The light is repeatedly reflected by the mirrors, like in CRDS technique, but the light spots on mirror surfaces are spatially separated. It fills the whole volume of the cavity. Avoiding the spots overlapping eliminates the light interference and allows one to eliminate the sharp resonances of the cavity. Consequently the smooth and broad absorption spectrum might be observed. In practical implementation, the off-axis design in CEAS technique additionally eliminates optical feedback from the cavity to the laser, reduces the sensitivity to vibration. It is especially important when in the CEAS setup the diode lasers are used. However, some instabilities in the light source intensity may decrease the signal-to-noise ratio. In this work we report of an off-axis CEAS-based sensors applied to NO2, NO and N2O measurement. Such sensors are ideal for field measurements because of its advantages. Furthermore, simplicity of CEAS setup causes the costs reduction.

2

Analysis of NOx absorption

The described method is based on measurements of absorption changes of optical radiation. The changes are caused by existence of trace amount of the detected substances in the air. For this reason, both the familiarity with characteristic absorption spectrum of the substances and proper selection of diagnostic radiation source are very important. The research and experimental works were focused on detection of nitrogen dioxide so far. Note that the maximum of the absorption spectrum of nitrogen dioxide is in the range of 400–450 nm. There are no absorption interferences from other gases or other vapors normally existing in the air in the mentioned spectral range. For this reason, it could be assumed that the intensity changes of registered radiation WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

206 Computational Methods and Experimental Measurements XIV passed through the investigated air are caused by changes of nitrogen dioxide concentration. The assumption does not take into consideration an influence of scattering by some aerosols and smokes existing in the air. However, they can be eliminated from the investigated gas using special filters. For other nitrogen oxides, analysis of detection opportunities based on spectroscopy in the infrared spectral range were performed.

Figure 2:

Absorption cross-section of nitrogen dioxide and a spectrum of matched laser.

For this purpose, detailed analysis of absorption spectrum was done. The required information was received from database presented by United States Environmental Protection Agency EPA (www.epa.gov). The wavelength range of 2-10 m was considered. Since the NOx measurements might be interfered by other compounds commonly existing in the air. In the described analysis the influence of the water vapor (steam) and other selected compounds was also analyzed. In Figure 3 the absorption characteristic of nitrous oxide (N2O) in the range of 4.43 − 4.58 μm is presented. It is worth to notice, that in this spectral range also slightly contribute carbon monoxide (CO) as well as carbon dioxide (CO2). Due to an adulteration possibility of measurements by air gases (e.g. CO2), the N2O detection one should conduct in the first absorption band, i.e. 4.45 − 4.49 μm. For this spectral range the quantum cascade lasers seems to be the most suitable. In Figure 3 the possibility of matching of the Alpes #sb1840UP laser to nitrous oxide detection using cavity ring down spectroscopy method is also presented. In Figure 4 the detailed absorption characteristic of nitric oxide (NO) in the range 5.10 − 5.60 μm is shown. In this region H2O absorption is also observed. Due to this fact the NO detection should be realized in the first absorption band, i.e. at 5.20 − 5.30 μm. In this figure the possibility of matching of the quantum cascade laser to nitric oxide detection using cavity ring down spectroscopy method is presented. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


207

The absorption spectra presented in Figures 3 and 4 are deeply modulated, sometimes show split into single peak trains following from vibrational structure of the molecule energy levels. In contrary to this the experimental spectra are supposed to be smoother due to pressure broadening in the atmosphere. The most beneficial wavelengths to detect individual oxides of nitrogen occur in the regions that are listed in Table 1.

Figure 3:

Absorption cross-section for nitrous oxide and a spectrum of Alpes #sb1840UP laser [8, 9].

Figure 4:

Absorption cross-section for nitric oxide and a spectrum of Alpes #sb1770DN laser [8, 9].



Detection wavelength ranges of the NO2, NO and N2O.

Particle nitrogen dioxide (NO2) nitrogen oxide (NO)

Detection wavelength range 0.35-0.45 μm 5.2-5.3 μm

nitrous oxide (N2O)

4.45-4.49μm

Analysis of the parameters of the laser systems offered by Alpes Lasers [9] shows that in the market there is a wide range of laser radiation sources (semiconductor lasers), which can be used for spectral analysis within the broad spectrum range of infrared radiation. It is also possible to use the lasers as a radiation source in the presented wavelength ranges (complying with a maximum of nitrogen oxides absorption). A very important feature of these lasers is their tunability, to some extent, of the wavelength of emitted radiation. In addition, broad bands of absorption of investigated compounds, compared with the emission spectrum of lasers allow for the selection of the laser, with the best energetic parameters (energy pulse, peak power).

3

Lasers used in NO2 sensor

The pulsed laser diodes were constructed in the Institute of High Pressure Physics Unipress in collaboration with its spin-off TopGaN. In this proprietary technology, a number of technological steps are patented unique solutions, as: i) the application of high-pressure grown GaN crystals with a very low dislocation density, ii) plasma enhanced molecular beam epitaxy (MBE), iii) misorientation of the GaN substrates leading to special features of p-type GaN and electrical contacts. (a)

Figure 5:

(b)

Schematic drawing of the violet laser diode (a), and I-V characteristics for laser diodes grown on GaN substrates with different miscuts (b).



209

Figure 5(a) shows the laser diode structure: the substrate is a 100 micron thick GaN crystal grown at high hydrostatic pressure with dislocation density as low as 100 cm-2 (the world lowest value) and free electron concentration of above 1019 cm-3, what makes the lower electrical contact to N-face of GaN highly conductive. A special feature of the substrates used is the misorientation angle optimized to have low series resistance of the structure with electrical contacts. Figure 5(a) shows an advantage of such substrate preparation. The epitaxial structures were grown using plasma enhanced molecular beam epitaxy (MBE) or metalorganic chemical vapor phase epitaxy (MOVPE). The active layer consists of 2-5 quantum wells (3-4 nm) of InGaN with In content of about 10%, and GaN barriers of thickness (6-8 nm). The emitted light is confined by two claddings of AlGaN/GaN short period (2nm/2nm) superlattices to obtain high doping effectiveness and strain compensation. To have a good light confinement, the claddings are grown with thickness and Al-content close to the critical conditions for mismatch-related defect generation. The dislocation density in such laser-diode structure is of 105 cm-2 what means just a few dislocations per laser stripe. An important issue in the nitride-based laser diodes is a small contrast in refractive indices between GaN in the waveguide and AlGaN lower cladding layer resulting in a leakage of electromagnetic wave into the substrate. Application of thick cladding and with a high Al content would result in the creation of defects (misfit dislocations and cracks); therefore, we have developed new cladding that eliminates such leakage. Figure 6 shows the far-field patterns of the laser emission for standard and improved lower AlGaN claddings. (a)

Figure 6:

(b)

Far field patterns of the violet laser diodes for standard (a) and modified claddings (b).

For the first picture, a leakage of electromagnetic wave into the substrate can be seen. Figure 7(a) shows the L-I characteristics for those two lasers. The stripe is of width 5-10 microns what results in a multimode emission and peak width around 1 nm. The length of the lasers was of 500-1000 microns. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

210 Computational Methods and Experimental Measurements XIV (a)

Figure 7:

(b)

L-I characteristics for lasers with standard and modified cladding that eliminates a leakage of light into the substrate (a), and spectral dependence of light transmittance at the back mirror of the laser diode (b).

The back laser mirror consists of five-fold /4 pairs of SiO2/TiO2. The front mirror is covered with 50 Å Al2O3. Figure 7(b) shows the spectra dependence of transmission for the back mirror. It can be seen that at 400 nm, almost 95% of light is reflected by the back mirror and the ratio between the light intensity emitted through the front and back mirror is about 150. All the improvements mentioned above give the laser diode technical parameters close to those reported by Japanese manufacturers, as Nichia, Sony, or Sanyo. The table 2 shows those parameters. Table 2:

Laser diode technical parameters.

Density of threshold current

Voltage at threshold

Slope efficiency

Pulse duration

Power at peak [mW]

5-8 kA/cm2

7-8 V

0.3-0.5 W/A

50 ns

200-2500

The laser diodes are packaged in standard 5.6 mm cans and mounted in laser diode modules manufactured by TopGaN (Figure 8(a)). The highest power obtained was 2.5 W for 50 micron stripe, as shown in Figure 8(b). However, for such wide stripes, the filamentation is observed, what must be corrected with the special optics.

4

Detection system

The detection system was constructed in the Institute of Optoelectronics MUT. Light at 414 nm from the diode laser (TopGaN firm) was directed into a 50 cm long optical cavity consisting of two concave, highly reflective mirrors (R > WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


211

0.999976 at 414 nm, Los Gatos Research). The laser beam was directed to the cavity using the diffraction grating and the mirror. The radiation leaving the cavity was registered with a photomultiplier (R7518, Hamamatsu), which is characterized by high gain (1.1107), high speed (the pulse rise time is equal 6.4±0.1ns) and low dark current. The PMT was equipped with the interference filter, the bandpass of which was matched to the laser line. Signal from the PMT is fed to transimpedance preamplifier. In the preamplifier, the operational amplifier AD 8038 type was used. It is characterized by a wide dynamic range. (a)

Figure 8:

(b)

405 nm laser module (a), and optical power versus current for the 50 micron wide laser stripes (b).

In order to obtain signal-to-noise ratio (S/N) of the photoreceiver we analyzed its noise equivalent scheme [13]. The calculated value of (S/N) of the first stage of photoreceiver (PMT plus preamplifier) is equal 115. Next, the signal from the photoreceiver output was digitized (with 100 MS/s sampling rate) using a 12-bit USB oscilloscope CS 328 (Clever Scope). Signal-to-noise ratio of the detection system was additionally improved by the use of the coherent averaging. In the system, the S/N is directly proportional to a root of the number of the averaging samples (nsampl) S S nsampl ,     N  total  N  FSP

(5)

where S/NFSP is the signal-to-noise ratio of the first stage of photoreceiver. If nsampl = 104, thus S/N of our photoreceiver is equal 1.1104. Probing of ambient air was accomplished through a measurement cell constructed of aluminum coated with Teflon inside. The cavity mirrors were mounted on the two ends of the cell. The mirrors were adjusted using Q-ring mounts and three fine-pitched screws. The cavity is equipped with a gas inlet and outlet. In the investigation a mixing system supplied from a bottle with NO2, and additionally from the source of pure nitrogen, was applied. The system allows for precision gas mixing and preparing the assumed NO2 concentration. The measurement was performed under the steady flow of the gas through the cavity. Moreover the measurement with good detection limit requires also good WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

212 Computational Methods and Experimental Measurements XIV filtration of the investigated air. This is necessary in order to avoid the light scattering in the aerosol particles as well as the dust deposition on the mirrors surfaces.

5

Experimental results

In Figure 9 the photography of the CEAS system is presented. The diode laser generated the radiation pulses (414nm) which duration time was about 50 ns and a repetition rate -10 kHz. Their peak power was about 250mW. The laser beam was directed to the cavity using mirrors. The filter composed of diffraction grating and iris diaphragm was used to eliminate broadband fluorescence of laser diode. In order to determine the signal-to-noise ratio, the noise measurements were performed. They were determined with the spectrum analyzer (SR 770). The voltage of noise above 20 Hz was below 1V. The transimpedance preamplifier noise was dominant (about 90%) in the first stage of photoreceiver. S/N of the developed system with coherent averaging of 104 samples reached the value of 1400.

Figure 9:

Photography of portable NO2 optoelectronic sensor.

Table 3: No. 1. 2. 3. 4. 5. 6. 8.

NO2 optoelectronic sensor parameters.

Parameter Sensitivity (NO2) Measurement range (NO2) Resolution of measurements Uncertainty SNR Measurement time Interfaces

Value 2.8×10 cm-1 (0,2 ppb) 0,2 ppb – 43 ppm 0,2 ppb 0,3% about 1400 1 min. USB 2.0


-9


213

The software controlling the data acquisition was developed using LabVIEW 7.1. The software provides possibility of the signal decay time and concentration of the NO2 was determination. The main sensor parameters are presented in the Table 3. Thanks to the first stage of the photoreceiver optimization and coherent averaging of the measured signal, the detection limit achieves the value of 2.810-9cm-1. Therefore measurements of 0.2 ppb NO2 concentration with uncertainty of 0.3% are available. Detailed analyses are given in our earlier works [10-14].

6

Future work

Sensors based on optical-absorption spectroscopy determine the presence of chemical vapors by comparing light-absorption spectra from multiple air samples. The spectra can be measured remotely with lidar or passive spectrometers, or in air samples drawn into optical cavities or other chambers. The spectra of analytes and interferentes are fit to the measured spectrum of each air sample to determine the best fitting concentration combination. A sample is determined to contain a given analyte when the fit yields a concentration that is statistically greater than ambient. The sensitivity and selectivity of a sensor are therefore determined by the ability to separate spectral differences due to the analyte concentration from spectral differences due to instrument noise or fluctuation in ambient conditions. Choosing analyte spectral features with the best combination of absorption strength and spectral uniqueness is therefore key to sensor performance. Progress of CEAS sensors construction is strictly dependent on new radiation sources investigations and optical systems and photoreceivers. Elaboration of new quantum cascade lasers and hetero-system detectors provide possibility of the new generation CEAS sensors developing.

Figure 10:

Block diagram of a system used for vapor IM detection.

Figure 10 presents a block diagram of a system, which can be used for vapor improvised material (IM) detection. It will include a sampling and preconcentrator system and three sensors; NO2, NO, and N2O. The WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

214 Computational Methods and Experimental Measurements XIV preconcentrator will adsorb of an explosive molecules from an inlet supply of air flowing at a high rate and relatively large volume. Then the adsorbing medium will be heated quickly to desorb the explosive material on NO2, NO and N2O, which will be directed into sensors. The NO2 sensor was analyzed earlier. In the NO and N2O sensors we would like to use quantum cascade lasers Alpes Lasers firm, and detection system with a polish detector, Vigo System S.A. [15]. An active element of a detector is a multilayer Hg1-xCdxTe heterostructure, optimized for radiation detection at the wavelength of about 5 µm. The element of a photodetector is mounted at three-stage thermoelectric cooler situated in a hermetic housing. This cooler ensures receiving the work in temperature of a detector, of about 200 K at the ambient temperature of 20oC.

7

Conclusions

In the paper, a portable NOx optoelectronic sensor was described. In the sensor CEAS technique was applied. It is one of the most sensitive laser spectroscopy methods. Thanks to the theoretical analysis, the main parameters of the optical cavity, the signal processing system, and in particular the signal-to-noise ratio, the signal processing system was developed. This system ensures registration of low level signals, and the decay time measurements with the uncertainty below 0.3%. The system consists of PMT, low noise transimpedance preamplifier, and 12-bit digital signal processing circuit. Moreover coherent averaging technique was applied. The features of the designed sensor show that it is possible to build a portable NOx sensor with the sensitivity of subppb. Such a kind of system has several advantages such as: relatively low price, small size and weight, and possibility of detection of other gases.

References [1] OKeefe A., D.A. Deacon, “Cavity ringdown optical spectrometer for absorption measurements using pulsed laser sources”, Rev. Sci. Instrum., Vol. 59, No. 12, 2544-2551, 1988. [2] Engeln R., G. Berden, R. Peeters, G. Meier, “Cavity enhanced absorption and cavity enhanced magnetic rotation spectroscopy”, Review Of Scientific Instruments, Vol. 69, No. 11, 3763 – 3769, 1998. [3] Kasyutich V.L., C.E. Canosa-Mas, C. Pfrang, S. Vaughan, R.P. Wayne, “Off-axis continuous – wave cavity-enhanced absorption spectroscopy of narrow-band and broadband absorbers using red diode lasers”, Appl. Phys. B, Vol. 75, 755-761, 2002. [4] Merienne M.F., A Jenouvrier, B. Coquart, “The NO2 absorption spectrum. I: absorption cross-sections at ambient temperature in the 300-500 nm region”, J. Atmos. Chem., Vol. 20, No. 3, 281-297, 1995. [5] Andriew C., Papino R., Ultrasensitive surface spectroscopy with a miniature optical resonator. Phys. Rev. Letters, Vol. 83, No 15, 1999. [6] Andriew C., Papino R., Evanescent wave cavity ring-down spectroscopy for ultra-sensitive chemical detection, SPIE, Vol. 3535, 1998. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


215

[7] Nicola J. van Leeuwen, J. C. Diettrich, and A.C. Wilson. Periodically Locked Continuous-Wave cavity ring-down spectroscopy, Appl. Opt. Vol. 42, pp.3670-3677, 2003. [8] http://www.epa.gov/ttn/emc/ftir/aedcdat1.html [9] http://www.alpeslasers.ch/lasers-on-stock/index.html [10] J. Wojtas, A. Czyżewski, T. Stacewicz, Z. Bielecki. Sensitive detection of NO2 with cavity enhanced spectroscopy. Optica Applicata, vol. 36, No 4, pp. 461-467 (2006), [11] Czyżewski, J. Wojtas, T. Stacewicz, Z. Bielecki, M. Nowakowski. Study of optoelectronic NO2 detector using cavity enhanced spectroscopy. Proc. SPIE. Optics and Optoelectronics. Optical sensor. 6585-68 (2007), [12] J. Wojtas, T. Stacewicz, Z. Bielecki, A. Czyżewski, M. Nowakowski. NO2 monitoring setup applying cavity enhanced absorption spectroscopy. The International Conference on Computer as a Tool, EUROCON 2007, Warsaw, September 9-12. Conference Proceedings, pp. 1205-1207 (2007). [13] J. Wojtas, Z. Bielecki. Signal processing system in the cavity enhanced spectroscopy. Opto-Electron. Rev. 16, No 4, pp. 44-51, 2008. [14] J. Wojtas, Z. Bielecki, J. Mikołajczyk, M. Nowakowski. Signal processing system in portable NO2 optoelectronic sensor. 6-8 May. Sensor+Test 2008 Proceedings, Nurnberg, Germany pp. 105-108 (2008). [15] www.vigo.com.pl




217

Multispectral detection circuits in special applications Z. Bielecki1, W. Kolosowski1, E. Sedek2, M. Wnuk1 & J. Wojtas1 1 2

Military University of Technology, Poland Telecommunication Research Institute, Poland

Abstract In this paper, the first stages of receivers of optical radiation are described. Special attention was paid to the selection of a detector adequate for the range of radiation being detected. Detectors of optical radiation, including UV, visible, and IR detectors are characterised. The methods of broadening of a transmission bandwidth were also discussed. Moreover, the design criteria of low noise preamplifiers used for the mentioned detectors were presented. Keywords: photoreceiver, noise model, low noise electronics circuits.

1

Introduction

Receivers of optical radiation are used in many up-to-date fields of science and technology, determining the current level of technological progress. The most important fields of their applications are in industrial automation, robotics, space technology, medicine, and in military technology. The optical radiation range covers electromagnetic radiation longer than the gamma but shorter than millimetre waves. Optical radiation from an object (signal source) is detected with a photoreceiver. It consists of optics, a detector, a preamplifier and a signal processing system. All objects are continually emitting radiation at a rate with a wavelengths distribution that depends on the temperature of the object and its spectral emissivity. However, free space optical signal transmission is required for most of the mentioned applications. Then radiation is attenuated by the processes of scattering and absorption. Scattering changes direction of a radiation beam; it is caused by absorption and subsequent re-radiation of energy by suspended particles. For small particles, WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090201

218 Computational Methods and Experimental Measurements XIV compared with the radiation wavelength, the process is known as Rayleigh scattering and exhibits a -4 dependence. However, scattering by gas molecules is negligibly small for wavelengths longer than 2 m. Also, smoke and light mist particles are usually smaller than IR wavelengths. Therefore IR radiation can penetrate further through smoke and mists than visible radiation. However, rain, fog and aerosol particles are larger and consequently they scatter IR and visible radiation to a similar degree. Total radiation received from any object is the sum of the emitted, reflected and transmitted radiation. Objects that are not blackbodies emit only a fraction of blackbody radiation, and the remaining fraction is either transmitted or reflected, in the case of opaque objects. When the scene is composed of the objects and backgrounds of similar temperatures, the reflected radiation tends to reduce the available contrast. However, reflections of hotter or colder objects significantly affect the appearance of a thermal scene. The reflected sunlight is negligible for 8–14 m imaging, but it is important in UV, VIS and 3-5 m band. In general, the 8–14 m band is preferred for high performance thermal imaging because of its higher sensitivity to ambient temperature object and its better transmission through mist and smoke [1, 2]. However, the 3-5 m band may be more appropriate for hotter objects, or when sensitivity is less important than contrast. Also, additional differences occur, e.g. a virtue of MWIR (Medium Wave Infrared) band. There, smaller diameter optics are required to obtain a certain resolution. Furthermore some detectors may operate at higher temperatures (thermoelectric cooling) than in the LWIR (Long Wave Infrared) band. Detectors operating in this region require cryogenic cooling (about 77 K). The ultraviolet detectors have been accomplished by two different devices: the photomultiplier tube (PMT) and the solid state detectors. The PMT is a vacuum tube in which radiation falls on a photocathode, causing electrons to be emitted. Photocathode materials such as SbKCs and CsTe can be used which exhibit maximum sensitivity in the range from 400 nm to 235 nm, respectively. The photon in photoemissive devices must have sufficient energy to eject electron from a photocathode material. The emitted electrons are accelerated to strike another plate, the dynode, causing the emission of a number of secondary electrons. The process is repeated several times, leading to gains of a 106 or more. The resulting current is amplified by an external circuit. PMTs are relatively high-cost, require high-voltage power supplies, and are susceptible to magnetic fields that distort electron trajectories. Photoemissive devices produce negligible dark backgrounds at room temperatures and can be constructed to be inherently solar blind. The alternative approach has been to use semiconductor ultraviolet detectors. In most semiconductor devices, the photon causes an electron to be transition into the conduction band. Currently, the most common devices are silicon-based CCDs where the detection process requires energies of approximately an electron volt. The detectors should have excellent sensitivity in the visible and near-IR range. These devices should be sensitive to thermally induced backgrounds at room temperatures. The semiconductor detectors made of GaN or other highband-gap materials, such as SiC, AlN, diamond, have the activation energy WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


219

higher than 3.2 eV, making these devices inherently solar blind and UV sensitive without thermally induced background. While there is an increased interest in UV monitoring, the detection methods have not in general kept pace with the demand. However, looking for next generation semiconductor materials, have now led to the development of new kinds of UV detector. In particular, UV detectors based on a new material, Gallium Nitride, show very competitive properties, such as high responsivity, low dark current, very low input power requirements and good rejection of visible light. The latter is particularly important, since optical filters used to reject visible light cause complications in the design of optical systems and make it more difficult to ensure reproducible measurements. The visible rejections are 750:1 and 0.85:1 for GaN and Si detectors respectively (visible rejection 325nm:400nm). Another very important benefit of GaN based detectors is that it is possible to tailor this wide band gap semiconductor to fit a popular UV response range. When GaN and AlN form an alloy AlGaAn, the band gap can change from 3,4 to 6,2 eV, which corresponds to the change of the optical cut-off from 365 nm 200 nm. Within this range a particular cut-off can be achieved by controlling the composition of the alloy. The band gap also determines the maximum operation temperature. At high temperatures the band gap starts to collapse so that the material can no longer act as a semiconductor. The nitride based semiconductors have a band gap which is 3 to 5 times longer and in consequence, are more stable at high temperatures. Current detectors can operate well at up to 85C. Moreover, if material quality improves the detectors will operate at even higher temperatures. The natural rejection of visible light, high temperatures operation, low operating voltage and flexibility of the optical cut-off will allow this material to be used as a next generation tool in the field UV detection. Typical applications UV detectors: UV curing and drying, combustion monitoring, arc detection, phototherapy, sterilization control, spectroscopy, biological agent detection, solar irradiance measurement, industrial process monitoring, missile and artillery fire detection, solar irradiance measurement, climatological and biological studies.

2

Detection theory

The common problem of any type of photon detector is how to terminate the photodetector with a suitable load resistor, and to trade off the performance between bandwidth and signal-to-noise ratio. This is necessary for a wide family of detectors, including phototubes, photoconductors, photodiodes, etc., all of which are described by a current generator Iph with a stray capacitance C across it. Let us consider the equivalent circuit of a photodetector ending on a load resistor RL, as shown in Fig. 1. This is the basic circuit for detection. The output signal in voltage V = IR or in current I. Two noise contributions are added to the signal. One is the Johnson (or thermal) noise of the resistance RL, and the second is the quantum noise.



Figure 1:

A general noise equivalent circuit for a photodetector.

The total current I = Iph + Id is the sum of the signal current (Iph) and the dark current (Id). With this current is associated the quantum (or shot) noise arising from the discrete nature of electrons and photoelectrons. Its quadratic mean value is given by [3] I n2  2q I ph  I d f , (1)





where q is the electron charge and f is the observation bandwidth. A general noise equivalent circuit for a photodetector is shown in Fig. 1. The above two fluctuations are added to the useful signal and the corresponding noise generators are placed across the device terminals. Since the two noises are statistically independent, it is necessary to combine their quadratic mean values to give the total fluctuation as I n2  2q I ph  I d  4kTf / RL . (2)





The bandwidth and noise optimisation impose opposite requirements on the value of RL. To maximise f one should use the smallest possible RL, whilst to minimise I n2 the largest possible RL is required. A photodetector can have a good sensitivity using very high load resistances, but then only modest bandwidths ( kHz or less), or it can be made fast by using low load resistances (e.g. RL = 50  but at the expense of sensitivity. Using Eq. (2), we can evaluate the relative weight of the two terms in total noise current. In general, the best possible sensitivity performance is achieved when the shot noise is dominant compared to the Johnson noise, i.e.





2q I ph  I d f  4kTf / RL ,

(3)

which implies the condition

RL min 

2kT q . I ph  I d

(4)

At low signal levels, very high values of resistance are required; for example, for Id = 5 pA, not an unusually low dark current, then RLmin = 10 G. However, if we are using a resistor termination value RL < RLmin then, the total noise can be written as (5) I n2  2q I ph  I d f 1  RL min / RL  . This means that the noise performance is degraded by factor RLmin/RL, compared to the intrinsic limit allowed by the dark current level. Thus, using RL WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


221

< RLmin means that the shot-noise performance is reached at a level of current not less than (6) I ph  I d  2kT / qRL . From this equation, we see that, for a fast photodiode with a 50  load at T = 300K, this current has very large value of 1 mA. The above considerations are valid for photodetectors without any internal gain, G. Let extend the calculation of the signal-to-noise ratio (S/N) to photodetectors having internal gain. In this case, the shot noise can be expressed in the form (7) I n2  2q I ph  I d fG 2 F , where F is the excess noise factor to account for the extra noise introduced by the amplification process. Of course, in a non-amplified detector, F = 1 and G = 1. The total noise is a sum of shot noise and thermal noise, then we obtain a S/N ratio I ph S  N 2q I ph  I d fF  4kTf RL G 2

 

3







1/ 2

.

(8)

Response time of the first stage of a photoreceiver

A response time of the first stage of photoreceiver can be shortened in many ways. Here, the possibility of such shortening in photoreceivers with p-i-n photodiodes is described. High-frequency properties of p-i-n photodiodes depend on lifetime of minority carriers. In photodiodes with the Schottky barrier, lifetime of carriers is negligible (10–14 s) when compared with minority carriers lifetime in p-i-n photodiodes [4]. Response time of p-i-n photodiode depends on:  time of carriers drift through the depletion region,  time of carriers diffusion to depletion region,  RC time constant of a load circuit of a detector. Influences of the time (d) of carriers diffusion to depletion region can be neglected assuming that majority of carriers are generated in a depletion region. A drift time of minority carriers through the depletion region depends on its width and carriers velocity (voltage of reverse bias). The drift time can be reduced by narrowing the depletion region. However, it causes increase in junction capacity. A detector capacity depends on the photodiode area (A), its resistivity (), and a voltage of reverse bias. A . (9) Cd 

Vb  0.51/ 2 

Decrease in RC time constant of the first stage of a photoreceiver, the same broadening of a transmission band can be obtained by using the lower capacity photodiode and increase in a value of reverse bias voltage. However, the higher


222 Computational Methods and Experimental Measurements XIV bias voltage can cause increase in dark current of a photodiode, i.e., the higher noise. Next factor affecting the photodiode response time (tr) depends mainly on its capacity and input resistance of a preamplifier. A time constant of this circuit is  RC 

R L  Rs Rsh C  Req C , R L  Rs  Rsh

(10)

where Rs is the series resistance of a photodiode, C is the sum of photodiode capacity and input capacity of a preamplifier, Req is the resultant resistance which is a parallel connection of the resistance Rsh of photodiode and the load resistance RL. A response time and transmission bandwidth can be shaped by selection of a capacity of the first stage of a photoreceiver. Decrease in a width of a depletion region causes reduction of photodiode sensitivity, the same possibility of detection of optical signals of low amplitudes. Thus, the compromise between a response time of a photoreceiver and its sensitivity is necessary.

Figure 2:

Factors affecting the response time of the first stage of a photoreceiver.

The above considerations show, that the detector response time is



tT  tr2  tt2  td2 and 3 dB limit frequency is given as





f  3dB  2 Req C

1/ 2

,

1 .

(11) (12)

Usually, a designer can obtain the broadening of a transmission bandwidth of a photoreceiver when a detector of low capacity and preamplifier of low input resistance are used. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


223

Figure 2 presents the factors influencing response time of the first stage of a photoreceiver. Of course broadening of a transmission bandwidth of the first stage of a photoreceiver will cause noise increase, i.e., decrease in its sensitivity.

4

Selection of the first stage of a photoreceiver

As an active element in a preamplifier, the bipolar or FET transistor or integrated system, with an input at a bipolar transistor, FET transistor or MOSFET transistor, can be used. The main criteria of selection are both the value of detector resistance and the range of transmitted frequencies [5–7]. Because low intensity signals reach a photoreceiver, a very important task is to optimize the noise of a photodetector-preamplifier, i.e., to obtain maximum of S/N ratio. The first stages of complex electronic devices significantly affect the level of a total noise of a system, so preamplifiers have to fulfil very special requirements. Optimum parameters of a preamplifier can be determined by the basis of analysis of particular sources of the noise in a total equivalent input noise of a photodetector-preamplifier system and on the basis of calculation of an equivalent input noise. A level of an equivalent noise at the input of a photodetector-preamplifier is univocally determined by the detector noise Vnd, background noise, and equivalent noise sources Vn and In. For non correlated components of noise, the total equivalent noise at the input of a photodetectorpreamplifier is defined as 2 2 (13) Vni2  Vnd  Vnb  Vn2  I n2 Rd2 , where Rd is the detector resistance. It can be generally stated that if high level of current noise is in a transistor of the first stage of a preamplifier, this transistor cannot operate with highresistance detector. High level of the current noise of input transistor causes high equivalent input noise at high resistance of a detector. On the contrary, lowresistance detector can well operate with low voltage noise preamplifier. In [1, 3], detailed analyses have been performed on selection of preamplifiers for various detectors of optical radiation. For optical radiation detectors, the voltage preamplifiers, transimpedance preamplifiers, and charge preamplifiers are used. High signal-to-noise ratio in a voltage preamplifier causes narrowing of a transmission bandwidth of a system. Due to application of a transimpedance preamplifier for the determined detector resistance, the higher bandwidth can be obtained. Charge preamplifiers form a separate group. Figure 3 presents a simplified scheme of the first stage of a photoreceiver with transimpedance preamplifier and noise sources. Such a system is commonly used for UV, VIS and IR detectors, both photoconductive detectors and photodiodes. Improvement in this S/N ratio can be obtained by: - decrease in influence of background radiation (narrowed field of view of a detector due to applied cooled diaphragms and optical filters), - decrease in generation-recombination noise produced from thermally induced carriers in semiconductor (lower temperature of detector operation), - decrease in thermal noise of a detector (cooling), WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


Figure 3:

Simplified equivalent scheme of the first stage of a photoreceiver with noise sources.

- decrease in thermal noise of load resistance (using high RL values and in some applications additional decrease in detector operation temperature), - use of the optimised ultra low-noise preamplifier [8–11].

5

Experimental results

For amplification of the signals from UV detectors, the most frequently voltage and transimpedance amplifiers are used. The basic idea of increase in input impedance of preamplifier is reduction of thermal noises. However, the high resistances RL cause narrowing of a band of the input stage of a photoreceiver. A preamplifier of high input impedance is significant load resistance for a detector, so it does not ensure wide range of signal changes. The problem of serious changes of a signal has been solved in transimpedance preamplifiers. For amplification of the signals from UV detectors, we used transimpedance preamplifier. In the preamplifier the integrated circuit of AD 548, AD 549, AD 795, and OPA 129 type were used. The noise current was obtained 1.8 fA/Hz1/2, 0.5 fA/Hz1/2, 0.6 fA/Hz1/2, 0.1 fA/Hz1/2 respectively, for f=1 kHz. In the case of IR detectors (e.g. HgCdTe), we used transimpedance preamplifiers too, but we optimized it on voltage noise. Integrated circuits LT 1028 and AD 797 type were used at the input stage of a preamplifier. These preamplifiers can be applied both for photodiodes and photoresistors. When a photoresistor is connected, the biasing resistor RL is required. For this system the noise voltage was below 1 nV/ Hz1/2, for f=1kHz.



225

Many problems have to be overcome during construction of low-noise preamplifier for low-resistance detectors (e.g., HgCdTe of resistance below 100 ). It is due to the fact that the detectors of resistance of the order of 20  produce noise voltage lower than 0.6 nV/Hz1/2, i.e., below the noise voltage generated in the best (available) amplifying elements. So, the question arises; it is possible to build a preamplifier of noise voltage below a detector noise? It appears that it can be achieved when several identical preamplifiers are connected in parallel and next the output signals are added. Such a system of signal processing ensures reduction of final input signal according to the relationship

Vn total  Vn  n 1 / 2 ,

(14)

where Vn is the noise voltage of a single preamplifier and n is the number of amplifying stages. For preamplifier with AD 797 type, the noise voltage value of 0.3 nV/Hz1/2, for n=9 and f=10 kHz was obtained.

6

Example of low noise photoreceivers in military applications

In recent years an increase of terrorism threats has occurred. It is supported by a large number of infrared guided missiles; limitation of possibilities of the airport protection, access to the airplane schedules and airplanes are an easy target (especially passenger planes). To prevent such occurrences the following security mechanisms are used: decreasing target’s signature, using camouflage smoke, using pyrotechnics sources, radiation generators, multi-spectral detection systems as well as blinding systems [13,14]. Some of these security mechanisms have limited range of use e.g. flare shouldn’t be use in the urbanized area. Flares do not meet the functions in close to attacking missiles and also do not protect airplane against the armour-piercing missiles attack. The multi-spectral systems are free from these drawbacks e.g. active radio systems, passive infrared systems, passive ultraviolet systems. Progress in the aviation technique caused minimal altitude limit reduction up to 20-50m, similar causes can occurs using the rocket technique. These objects are hard to detect because of physics phenomena of the microwaves propagation in the ground zone. Based on radiolocation equation, maximal detection distance depends on direction toward the target and is written as rmax  ,    rmo F  ,   .

(15)

This formula describes the surface equation with following parameters: r,,, with the coordinates beginning from the radiolocation station as a reference point. This surface divides the area around radiolocation station into the two areas: area I, where the target is detectable and area II, where target is invisible (Fig. 4a). WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

226 Computational Methods and Experimental Measurements XIV Taking into consideration the ground impact it is necessary to add in radiolocation station’s range in the free space equation interferometric multiplier. The radiolocation station’s range is calculated using the following formula rmax     rmo F        , (16) where F() – characteristics multiplier, () – interferometric multiplier. In Fig. 4b radiolocation station visibility is shown.

Figure 4:

Radiolocation station visibility zone in the free space (a) and radiolocation station visibility (b)

Figure 4 appears a dead area. To detect the target in this area the optoelectronics systems are used. The systems operate in UV and IR spectrum region. They can be installed on the shared mast together with radiolocation antenna. The range of this system is determined using following formula d km  3,56  ha ( m )  hc ( m )  , (17) where: ha- hang height of the optoelectronics device in meters, hc - target height in meters. In the Institute of Optoelectronics MUT original devices were elaborated. One of them is laboratory model of passive locator of flying objects. This device includes thermodetection modules with PV HgCdTe detector, polish Vigo System Ltd., optimized for the spectral range of 3-4.2µm. The passive locator can detect thermal objects from several kilometres (on the right of Fig. 5). On the left of Fig. 5 we can see these same thermal object, but in UV spectral range. In such detection systems we used photoreceiver with AlGaN detectors. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


Figure 5:

7

227

View of the thermal object in UV and IR wave ranges [12].

Conclusions

Increase in a signal-to-noise ratio in the first stages of photoreceivers can be obtained as a result of decrease in the noise originating from background, minimisation of a detector’s noise, selection of an optimal point of detector operation, and application of low-noise preamplifier matched to detector. Decrease in background noise can be achieved by reduction of an angle of view of a detector, application of a selective filter matched to a spectrum of a signal noise, and application of cooled choppers for long wavelength detectors. Reduction of detector’s noise is mainly due to lower operation temperature, narrower noise bandwidth of a system, and selected optimal working point. The noise of a preamplifier in the total equivalent noise of the first stage of a photoreceiver is the lowest one in the conditions of noise matching. A value of detector resistance is a main criterion for selection of low-noise preamplifier. For low-resistance detectors, preamplifiers optimised with respect to noise voltage should be used but for high-resistance detectors with respect to noise current. Further increase in signal-to-noise ratio at the receiver output is possible due to application of adequate techniques of signal modulation and demodulation. The results of the above activities were used for experimental investigation, i.e., several optical detection devices were designed, performed and tested. The results presented in this work do not comprise all the problems related to maximisation of S/N ratio in optical receivers.

References [1] Rogalski, Z. Bielecki. Chapter entitled “Detection of optical radiation”, in Handbook of optoelectronics. Taylor & Francis, New York, London pp. 73117 (2006).


228 Computational Methods and Experimental Measurements XIV [2] Z. Bielecki, J. Mikolajczyk. Chapter entitled. “Passive infrared detection” in Photonics Handbook, Laurin Publishing, USA H123-H125, (2005). [3] A. Rogalski, Z. Bielecki. Detection of optical radiation. Bulletin of Polish Academy of Science, vol. 52, no. 1, pp. 43-66 (2004). [4] Z. Bielecki: “Photoreceiver with avalanche C-30645 photodiode” IEE Proceedings Optoelectronics, Vol. 147, pp. 234-236, 2000. [5] C.D. Mothenbacher, J.A. Connelly: Low-noise electronic system design. New York, Willey, 1995. [6] Z. Bielecki: “Maximisation of signal to noise ratio in infrared radiation receivers,” Opto-electron. Rev. 10, 209–216 (2002). [7] Z. Bielecki. Some problems of optimization of signal-to-noise ratio in infrared radiation receivers. Proc. SPIE, Vol. 5125, pp. 238-245 (2002). [8] Z. Bielecki, M. Brudnowski. Method of popcorn-noise reduction. OptoElectron. Rev., Vol. 11, no. 1, pp. 45-50 (2003). [9] Z. Bielecki. Readout electronics for optical detectors. Opto-Electron. Rev., Vol. 12, no. 1, pp.129-137 (2004). [10] Z. Bielecki, W. Kolosowski, R. Dufrene, E. Sedek, J. Wojtas. Photoreceiver for BLU/UV detection. Proc. of SPIE, Vol. 5472, pp. 383-390, (2004), [11] R. Cwirko, Z. Bielecki, J. Cwirko, L. Dobrzanski. Low-frequency noises as a tool for UV detectors characterisation. Opto-Electron. Rev., vol. 14, no 2. pp.155-160 (2006). [12] Z. Bielecki, K. Kopczynski, M. Kwasny, Z. Mierczyk. In polish. Monitoring zagrozen bezpieczenstwa. III Międzynarodowa Konferencja Naukowa Zarzadzanie kryzysowe” Szczecin, Materiały konferencyjne, s. 310-320 (2005). [13] Z. Bielecki, W. Kołosowski, M. Muszkowski, E. Sędek. Chapter entitled. “Phase shifters or optoelectronic delay lines application to sequential analysis of space adaptive phased arrays antennas”, in Computational Methods and Experimental Measurements, WIT Press, Southampton, Boston, UK, pp.241-249 (2005). [14] E. Sędek, Z. Bielecki, M. Muszkowski, W. Kołosowski, G. Różański, M. Wnuk. „Optoelectronic system for phase array antenna beam steering” in Computational Methods and Experimental Measurements, WIT Press, Southampton, Boston, UK, pp. 801-808 (2007).



229

Modification of raised cosine weighting functions family C. Lesnik, A. Kawalec & J. Pietrasinski Department of Electronics, Military University of Technology, Poland

Abstract Modification of the known family of raised cosine weighting functions with the power of n by using its convolution operation with an auxiliary rectangular window having variable duration time is considered in the paper. Such a modified functions family was derived in a general and exact form for the defined weighting function in a time and frequency domain as well. It was done with the help of the convolution windows constant – length type preparation technique. The derived weighting function was applied for radar chirp signals synthesis with nonlinear frequency modulation (NLFM). Chosen examples of the simulation research of the discussed modified weighting functions family features and radar signals of NLFM type synthesised thanks to it are presented in the paper. Keywords: weighting functions, radar signal synthesis, nonlinear frequency modulation signal, mainlobe, sidelobe, matched filtration.

1

Introduction

Radar signals synthesis is one of the most important problems of modern radiolocation. These signals transmitted to observed space should have very specific features. The so-called matched filter is a very specific part of a radar receiver. Its main task is to maximize SNR (Signal to Noise Ratio) at its output. In this way radar range may be maximized too. In the case of radar chirp signals transmission an echo signal observed at the receiver output has a shape of the very short pulse. Thanks to it radar range resolution may be good. A product of a received signal matched filtration is an output signal having mainlobe and sidelobes. Presence of sidelobes is of course an unwanted effect because it causes weak echo signals detection difficulty. WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line) doi:10.2495/CMEM090211

230 Computational Methods and Experimental Measurements XIV An effective technique of sidelobes attenuation is suitable weighting function application. These are so called weighting windows that modify the matched filter transmittance. A resulted filter mismatch is an additional effect of the weighting procedure. It leads to the worse value of the output SNR and in this way it causes radar range shortening. Application of radar signals of NLFM type is a lossless technique of sidelobes attenuation. The principle of stationary phase is used for the NLFM signals synthesis, detailed elsewhere in [1–3]. In such an approach suitable weighting functions are applied too. As a result of principle of stationary phase using unwanted accumulation of numerical errors can be observed. In order to its decrease the knowledge concerning exact, analytical form of the weighting function applied for the signal synthesis is desired. There are several known techniques used for weighting functions synthesis having expected features important from the point of view of application. One of them is based on convolution operation. To be more precise, there are known methods, such as: time convolution of parent windows, described by Harris [4] and Nuttall [5], multiple time convolution of rectangular windows, described by Wen [6] and Dai and Gretsch [7] or multiple time auto and cross convolution of other well known windows, described by Reljin et al. [8, 9]. These so called convolution windows have greater sidelobe attenuation and their fast decay. Unfortunately the output signal mainlobe width increases. Modification of the known family of the raised cosine weighting functions by using convolution operation of this functions with an auxiliary rectangular window having variable duration time is considered in the paper. It enables finetuning of the weighting function properties. The exact analytical formula of the modified function was derived in time and frequency domain. The paper is organized as follows. The known idea of convolution weighting function preparation is presented in section 2 as an introductory remark. The exact closed formula of the modified raised cosine weighting functions family in the continuous domain of time and frequency is derived in section 3. Examples of chosen features of the modified raised cosine weighting functions family are discussed in section 4. There is an example of its application for radar signal NLFM type synthesis too. A few sentences of conclusion are written in section 5.

2 Convolution weighting function design General form of weighting function with finite time duration wT (t ) can be described as a product as follows wT (t ) = w(t )rT (t ) ,

where:

w(t ) - considered unlimited time function,

rT (t ) - unitary rectangular window as follows:


(1)


231

T  1 for t ≤   2 , T > 0, (2) =  0 for elsewhere t  T - rectangular window time duration. Equation (1) is the equivalent of the convolution operation in the frequency domain 1 [W (ω) ∗ RT (ω)] , WT (ω) = (3) 2π t rT (t ) = rect  T

where: W (ω) and RT (ω) are Fourier transforms of the w(t ) and rT (t ) respectively. Let us consider weighting function modification wT (t ) described by eqn (3) that is based on its convolution with auxiliary rectangular window rτ (t ) having variable duration time τ wτ (t ) = wT (t ) ∗ rτ (t ) = [w(t )rT (t )] ∗ rτ (t ) ,

(4)

where wτ (t ) - modified weighting function. The auxiliary rectangular window rτ (t ) is defined as follows 1  t  τ rτ (t ) = rect   =   τ  0 

for t ≤

τ 2

, τ>0,

(5)

for elsewhere t

where τ - variable duration time of the auxiliary rectangular window. A factor 1 τ causes that the weighting function max value described by eqn (4) is independent of the auxiliary rectangular window duration time. It results from the fact that the area under rτ (t ) is unitary, described by Brandwood [10]. The convolution operation described by eqn (4) equals formula as follows in the frequency domain Wτ (ω) =

1 [W (ω) ∗ RT (ω)]Rτ (ω) , 2π

(6)

where: Wτ (ω) and Rτ (ω) are the Fourier transforms of the wτ (t ) and rτ (t ) respectively. As a result of convolution operation modified weighting function wτ (t ) is different from zero within an interval from − (T + τ ) 2 up to (T + τ ) 2 and its duration time is (T + τ ) . In order to keep constant the final weighting function duration time (it equals T ) there is necessity for suitable scaling of eqns (4) and (6) in time and frequency according to the Fourier transformation scaling property WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

232 Computational Methods and Experimental Measurements XIV t x   ↔ a X (aω) , a

where factor of scale change is defined as follows a=

T . T +τ

As a result of eqns (4) and (6) scaling the general form of the modified weighting functions in the time and frequency domain are as follows:   t   t  t wτ (t ) =  w  rT   ∗ rτ   , a a a      Wτ (ω) =

(7)

a [W (aω) ∗ RT (aω)]Rτ (aω) . 2π

(8)

Scaled modified weighting function described by eqn (7) is different from zero within the interval from −T 2 up to −T 2 and their duration time is T . The window function received in this way is called constant-length CON window and it is described by Reljin et al. [9]. The presented method will be applied for known and very useful (among others for synthesis of the radar and sonar signals with nonlinear frequency modulation, NLFM) family of cosine windows.

3 Modification of raised cosine weighting functions family – general solution Let us consider cosine window family with power of n (raised cosine with power of n ) having a general form as follows   t wT(n ) (t ) = k + (1 − k )cos n  π  T 

where: k

t   rect   , T  

(9)

- real parameter of the function, k ∈ 0, 1 ,

n - integer parameter of the function, n ≥ 1 , T - weighting function duration time. As a result of known trigonometric identities application the general description of the function described by eqn (9) with the n -th power was achieved s −1 n   t t wT(n ) (t ) =  A + B ∑   cos ni π  rect  i T T i =0    


 , 

(10)


233

where:  n + 1 s= , x  the biggest integer not higher than x ,  2  for odd n, n = 1, 3, 5, ... k  1− k  n  A=  , k + n   for even n, n = 2, 4, 6, ...  2  n 2 B =

1− k

, 2n −1 ni = (n − 2i ) ,  x  , x ≥ y - binomial coefficients.  y As a result in the general case the weighting function described by eqn (10) in the time domain is a sum of a constant component and cosine functions limited in time and scaled in amplitude. Their period is a total submultiple of the fundamental period that equals 2T . Considering the weighting function general description presented in the eqn (10) in the frequency domain and based on the eqn (6) one can obtain WT(n ) (ω) =

[

]

1 W (n ) (ω) ∗ RT (ω) . 2π

(11)

Using Fourier transforms pairs 1 ↔ 2πδ(ω) , cos ω0 t ↔ πδ(ω − ω0 ) + πδ(ω + ω0 )

one can achieve s −1 n   π  π   W (n ) (ω) = A2πδ(ω) + B 2π ∑   δ ω − ni  + δ ω + ni  . i T T    i =0    

(12)

After Fourier transform of the rectangular window application ωT t , rect   ↔ T Sa 2 T 

where Sa (.) denotes a function in a shape of Sa (x ) = (sin x ) x , and using delta distribution properties it is possible to achieve the final, general form of the cosine functions family with power of n in the frequency domain WT(n ) (ω) = AT Sa

 π T  π  T  ωT B s −1 n    + T ∑   Sa  ω − ni   + Sa  ω + ni    . (13) 2 2 i = 0  i    T  2  T  2 

This function is a sum of functions of the (sin x ) x type that parameters and location on the radian frequency axis depend on the power n and the window duration time T . WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

234 Computational Methods and Experimental Measurements XIV The general form of the modified weighting function in the time domain, defined for separate time intervals was solved thanks to convolution operation of the function presented in the eqn (10) with the auxiliary rectangular window described by eqn (5): for −

T τ T τ − ≤tT 1

CHIRP GENERATOR

CHIRP GENERATOR

exp(-jµt 2)

exp(-jµ W 2)

SHORT MULTIPLIER

Figure 1:

B 1 ,T 1

Chirp algorithm arrangement.

Figure 2:


CR performance for CW signals, (2µs/div).


Figure 3:

CR performance for pulsed LFM signal, (2µs/div).

Figure 4:

CR output signals for two input signals: pulsed LFM signal and CW signal, 0,5µs/div.

Figure 5:

CR output signals for two input signals: pulsed LFM signal and CW signal, 0,5µs/div.

Figure 6:

CR input and output signals pulsed unmodulated signal, 5µs/div.

The CR output signal reflects instantaneous spectrum of the analysed signal. It performs signal’s analysis in particular time samples refer to us as duty cycles. This results signal spectrum averaging over reference signal time and bandwidth. In other words, the CR output signal expresses signal energy being under analysis. Thus, with respect to the time-frequency distribution terms, to collect a total signal energy it is necessary to gather signals existing in several duty cycles of the CR. Therefore the time-frequency signal representation will be referred to the plane (duty cycle number, frequency expressed by time) over which the CR output signal will be evaluated. As resulted from the registrations, the CR operation consists of several duty cycles in which instantaneous spectrum of a signal is yielded. For particular cycle, signal spectrum is represented by video pulse position relatively to the beginning of a reference signal. Pulse position is determined by the input signal frequency value. Thus to evaluate spectrum it is sufficient to measure the time intervals between CR output video pulse and the WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


263

pulse triggering the reference signal. Another factor that should be taken under attention is CR output pulse amplitude variability.

3

Compressive receiver output signal processing

The CR output is followed by many short pulses with positions determined by the input signal frequency value. There exist also side lobes associated with these pulses. Thus to measure the frequency of the input signal the measuring of the main pulse centre with simultaneous neglecting of side lobes is required. To determine the signal frequency, the time parameters measurements should be performed. The simplest approach is to compare the CR output signal with fixed thresholds. When the processed pulse breaks these thresholds, it will be declared a legible output. This approach includes two main shortcomings. Firstly, the amplitude of the output pulse changes with the input signal level. Secondly the fixed thresholds detection scheme does not distinguish between main and side lobes in general. Thus to overcome problems of the CR output signal processing digitizing technique is recommended. To proceed examinations the CR practical model has been used [3, 5–7]. VIDEO MICROWAVE RECEIVER

IF

COMPRESSIVE RECEIVER

DIGITIZER

RECEIVING ANTENNA

COMPUTER

Figure 7:

Measurement system lock diagram.

Figure 8:

CR and receiver video outputs, (pulsed LFM signal, pulse width 10µs).

The measurement system block diagram is presented in Fig.7. Some results of the examinations are presented in Fig.9-12 (in Fig.10 upper trace represents CR output, lower trace represent receiver video output). Fig.8 combines instantaneous video of the intercepted signal and its instantaneous spectrum represented by the CR output video for the single pulse. Fig.9-12 present spectrum two types of radar signals - pulsed linear frequency modulated and narrowband (unmodulated) (in Fig.10, 12 record number denotes time, sample number denotes frequency). Number of pulses under analysis depends on digitizer threshold. As it is seen in the case of pulse train it is possible to estimate frequency deviation on pulse (on the base of pulse train), however it is not possible to determine modulation type.



Figure 9:

CR output signals, pulsed unmodulated signal (pulse width 5µs).

Figure 10:

CR output signals pulsed unmodulated signal.

Figure 11:

CR output signals, pulsed LFM signal (pulse width 10µs).

Figure 12:

CR output signals, pulsed LFM signal (pulse width 10µs).

4

Wigner-Hough transform

Methods of time-frequency analysis enable to create radar signal image representing its instantaneous frequency [11, 12]. Processing such an image one can determine intrapulse parameters of radar signal. Hough transform is a technique used in image processing, that could also be implemented here [15]. This technique enables to analyze global features of an image of the base of local ones (point is the favourite one). The method relay on lines detection that could be represented in parametric form as straight lines circles, polynomials [9, 10, 15]. The problem analyzed in the paper may be considered as a separate application of two stages: firstly the Wigner-Ville transform (WV) is calculated [11, 12] and secondly the Hough transform is used. It is possible to apply here joint Wigner-Hough transform. Such method relays on Hough transform application to Wigner-Ville transformation of a signal. Particularly the method will be used to analyze multi-component signals with linear intra-pulse frequency modulation. The aim of the paper is to present WHT performance with application to multi-component radar signals processing. Wigner-Hough WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)


265

transform of complex signal s(t) is defined as a transformation from time domain to parameters (f,g) domain accordingly to the expression [10]:

WH s ( f , g ) = ∫

∞

∞

∫ s(t + τ 2)s (t − τ 2)⋅ e *

− j 2π ( f + gt )τ

−∞ −∞

dτdt

(4)

The transform defined by the equation (4) one can consider as integral of Wigner-Ville transform:

WH s ( f , g ) = ∫

∞

∫

∞

∞

Ws (t , v)δ (v − f − gt )dtdv = ∫ Ws (t , f + gt )dt

−∞ −∞ −∞ (5) where δ(ν-f-gt) is delta function and Wigner-Ville transform is defined accordingly [11]:

∞

Ws (t , f ) = ∫ s (t + τ 2 )s * (t − τ 2 )e − j 2πfτ dτ −∞

(6)

For practical applications discrete Wigner-Hough transform is needed. The discrete WHT for sequence x(n), n=0,1,...,N-1 (where N is even) is given by the equation [10]: WH x ( f , g ) = +

N −1

∑

N / 2 −1

n

∑ ∑ x ( n + k ) x ( n − k )e *

− j 4πk ( f + gn )

n =0 k = − n

N −1− n

∑ x ( n + k ) x ( n − k )e *

+

(7)

− j 4πk ( f + gn )

n = N / 2 k = − ( N −1− n )

Figure 13:

WV of two LFM signals embedded in noise.

Figure 14:

WHT of two LFM signals.

As it results from (7), WHT for signals with linear frequency modulation (LFM) reaches the maximum value in the points of coordinates (f0,g0). It denotes that detection and parameter estimation of LFM signals embedded in noise one can consider as a peak search in parameter domain (f,g). An example of WHT application in the case of multi-component signals interfered by additive Gaussian noise is presented on Fig.13, 14. Fig.13 presents WV of two LFM signals for signal-to-noise ratio (SNR) equal zero dB. Both signals are LFM signals and have the same amplitude, the same frequency-to-time slope but different frequency carriers. Signals were simulated, however they could reflect possible real values. On the base of WV image it is difficult to identify the signal type. When WHT is used then in transformed WV image two peaks, representing analyzed signals, are observed. Fig.14 shows two important advantages of WHT: WIT Transactions on Modelling and Simulation, Vol 48, © 2009 WIT Press www.witpress.com, ISSN 1743-355X (on-line)

266 Computational Methods and Experimental Measurements XIV first - in parameters domain peaks of signals are observed because cross terms were canceled, second – WHT gives integration gain with relation to noise as a result of integration. Parameters estimation of multi-component LFM signals is performed simultaneously for all elements in (f,g) plane represented here by r and theta (Fig.14). When peak exceeds predetermined threshold, it is declared that LFM signal is present and its parameters f0 and g0 (r,theta) are coordinates of the peak.

5

Noise influence on WHT transform

When signal is embedded in noise the WHT is becoming random variable WHTs+v(f0,g0). The peak of WHT moves to the point of coordinates (f0+δf, g0+δg). In this case the WHT properties are described by output SNR defined accordingly to [10]: WHTs ( f 0 , g 0 ) (8) var{WHTs + v ( f 0 , g 0 )} As it results from expression (8) output SNR is defined by WHTs(f,g), (WHT of signal) and by WHTs+v(f,g), (WHT of signal embedded in noise). Signal parameters accuracy estimation is determined by variance of random variables δf and δg. Assuming that noise is zero mean, white Gaussian process the following expression for SNROUT is received [10]: 2

SNROUT =

SNROUT =

N 4 A4 4

N 3 A 2σ n2 N 2σ n4 + 2 2

N2 2 SNR IN 2 = N ⋅ SNR IN + 1

(9) where input SNRIN is defined as A2/σ2n, A – signal amplitude, σ2n – noise variance, N – number of samples. Last expression reveals the threshold effect. When input SNR is high (SNRIN>>1), the expression (5) is approximated by SNROUT=N SNRIN/2. The integration gain proportional to number of integrated samples is present. In the case when input SNR is low (SNRIN

Computational Methods and Experimental Measurements XIV (WIT Transactions on Modelling and Simulation) (Wit Transactions on Modeling and Simulation; Fourteenth ... Methods and Experimental Measurements)

Computational Methods and Experimental Measurements XV (Wit Transactions on Modelling and Simulation)

Computational Methods and Experimental Measurements XIII

Boundary Elements and Other Mesh Reduction Methods XXIX (Wit Transactions on Modelling and Simulation)

Measurements, Modelling and Simulation of Dynamic Systems

Measurements, Modelling and Simulation of Dynamic Systems

Measurements, Modelling and Simulation of Dynamic Systems

Computational Methods in Multiphase Flow V (WIT Transactions on the Engineering Sciences) (Wit Transactions on Engineering Sciences)

Air Pollution XIX (Transactions on Ecology and the Environment) (Wit Transactions on Ecology and the Environment)

Monitoring, Simulation, Prevention and Remediation of Dense and Debris Flow III (Wit Transactions on Engineering Sciences)

Materials Characterisation V: Computational Methods and Experiments (Wit Transactions on Engineering Sciences)

Materials Characterisation IV: Computational Methods and Experiments (Wit Transactions on Engineering Sciences)

Environmental Toxicology III (Wit Transactions on Ecology and the Enviromnment)

Air Pollution XVII (Wit Transactions on Ecology and the Environment)

Air Pollution XVI (Wit Transactions on Ecology and the Environment)

Sustainable Chemistry (Wit Transactions on Ecology and the Environment)

Island Sustainability (Wit Transactions on Ecology and the Environment)

Modelling in Medicine and Biology VIII (Wit Transactions on Biomedicine and Health)

Transactions on Rough Sets XIV

Computational Methods in Multiphase Flow VI (Wit Transactions on Engineering Sciences)

Biomathematics: Modelling and Simulation

Biomathematics: Modelling and Simulation

Multiscale modelling and simulation

Cancer Modelling and Simulation

Biomathematics: Modelling and Simulation

Cancer Modelling and Simulation

Multiscale Modelling and Simulation

Biomathematics: Modelling and Simulation

Computational Methods and Experimental Measurements XIV (WIT Transactions on Modelling and Simulation) (Wit Transactions on Modeling and Simulation; Fourteenth ... Methods and Experimental Measurements)

Computational Methods and Experimental Measurements XV (Wit Transactions on Modelling and Simulation)

Computational Methods and Experimental Measurements XIII

Boundary Elements and Other Mesh Reduction Methods XXIX (Wit Transactions on Modelling and Simulation)

Measurements, Modelling and Simulation of Dynamic Systems

Measurements, Modelling and Simulation of Dynamic Systems

Measurements, Modelling and Simulation of Dynamic Systems

Computational Methods in Multiphase Flow V (WIT Transactions on the Engineering Sciences) (Wit Transactions on Engineering Sciences)

Air Pollution XIX (Transactions on Ecology and the Environment) (Wit Transactions on Ecology and the Environment)

Monitoring, Simulation, Prevention and Remediation of Dense and Debris Flow III (Wit Transactions on Engineering Sciences)

Materials Characterisation V: Computational Methods and Experiments (Wit Transactions on Engineering Sciences)

Materials Characterisation IV: Computational Methods and Experiments (Wit Transactions on Engineering Sciences)

Environmental Toxicology III (Wit Transactions on Ecology and the Enviromnment)

Air Pollution XVII (Wit Transactions on Ecology and the Environment)

Air Pollution XVI (Wit Transactions on Ecology and the Environment)

Sustainable Chemistry (Wit Transactions on Ecology and the Environment)

Island Sustainability (Wit Transactions on Ecology and the Environment)

Modelling in Medicine and Biology VIII (Wit Transactions on Biomedicine and Health)

Transactions on Rough Sets XIV

Computational Methods in Multiphase Flow VI (Wit Transactions on Engineering Sciences)

Biomathematics: Modelling and Simulation

Biomathematics: Modelling and Simulation

Multiscale modelling and simulation

Cancer Modelling and Simulation

Biomathematics: Modelling and Simulation

Cancer Modelling and Simulation

Multiscale Modelling and Simulation

Biomathematics: Modelling and Simulation

Recommend Documents