10.1007/978 3 642 30668 6

Topics in Intelligent Engineering
and Informatics 2
Editorial Board
Editors-in-Chief
János Fodor
Imre J. Rudas
Editorial Advisory Board

Ildar Batyrshin (Mexico)
József Bokor (Hungary)
Bernard De Baets (Belgium)
Hamido Fujita (Japan)
Toshio Fukuda (Japan)
Fumio Harashima (Japan)
Kaoru Hirota (Japan)
Endre Pap (Serbia)
Bogdan M. Wilamowski (USA)
Review Board
P. Baranyi (Hungary) E. Petriu (Canada)
U. Bodenhofer (Austria) R.-E. Precup (Romania)
G. Fichtinger (Canada) S. Preitl (Romania)
R. Fullér (Finland) O. Prostean (Romania)
A. Galántai (Hungary) V. Puri (Italy)
L. Hluchý (Slovakia) GY. Sallai (Hungary)
MO Jamshidi (USA) J. Somló (Hungary)
J. Kelemen (Czech Republic) M. Takács (Hungary)
D. Kocur (Slovakia) J. Tar (Hungary)
P. Korondi (Hungary) L. Ungvari (Germany)
G. Kovács (Hungary) A.R. Várkonyi-Kóczy (Hungary)
L.T. Kóczy (Hungary) P. Várlaki (Hungary)
L. Madarász (Slovakia) L. Vokorokos (Slovakia)
CH.C. Nguyen (USA)
For further volumes:

http://www.springer.com/series/10188
Aims and Scope
This book series is devoted to the publication of high-level books that contribute
to topic areas related to intelligent engineering and informatics. This includes
advanced textbooks, monographs, state-of-the-art research surveys, as well as edited
volumes with coherently integrated and well-balanced contributions within the main
subject. The main aim is to provide a unique forum to publish books on math-
ematical models and computing methods for complex engineering problems that
require some aspects of intelligence that include learning, adaptability, improving
efficiency, and management of uncertain and imprecise information.
Intelligent engineering systems try to replicate fundamental abilities of humans
and nature in order to achieve sufficient progress in solving complex problems.
In an ideal case multi-disciplinary applications of different modern engineering
fields can result in synergistic effects. Informatics and computer modeling are the
underlying tools that play a major role at any stages of developing intelligent sys-
tems. Soft computing, as a collection of techniques exploiting approximation and
tolerance for imprecision and uncertainty in traditionally intractable problems, has
become very effective and popular especially because of the synergy derived from
its components. The integration of constituent technologies provides complemen-
tary methods that allow developing flexible computing tools and solving complex
engineering problems in intelligent ways.
Ladislav Madarász and Jozef Živčák (Eds.)
Aspects of Computational
Intelligence: Theory and
Applications
Revised and Selected Papers of the 15th
IEEE International Conference on Intelligent
Engineering Systems 2011, INES 2011
123
Editors
Prof. Ladislav Madarász Prof. Jozef Živčák
Dept. of Cybernetics and Artificial Dept. of Biomedical Engineering and
Intelligence Measurement
Faculty of Electrical Engineering and Faculty of Mechanical Engineering
Informatics Technical University of Košice
Technical University of Košice Košice
Košice Slovakia
Slovakia
ISSN 2193-9411 e-ISSN 2193-942X

ISBN 978-3-642-30667-9 e-ISBN 978-3-642-30668-6
DOI 10.1007/978-3-642-30668-6
Springer Heidelberg New York Dordrecht London
Library of Congress Control Number: 2012939521
c Springer-Verlag Berlin Heidelberg 2013

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of pub-
lication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any
errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect
to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Preface
The International Conference on Intelligent Engineering Systems (INES) took place

in Slovakia in AquaCity Poprad (15th IEEE International Conference on Intelligent
Engineering Systems). This is the second time that Slovakia through the interme-
diary of the Technical University of Košice has been involved in its preparation
and hosted this prestigious international event. The first IEEE INES event in Slo-
vakia was organized as the third Annual Conference (3rd IEEE INES’ 1999, Stará
Lesná, Vysoké Tatry, November 1–3, 1999) attended by participants from 24 coun-
tries where 114 full papers were presented.
This volume put together by the editors is a compilation of carefully selected
extended papers written on the basis of original contributions corresponding to the
papers presented at the 15th Annual International Conference INES 2011.
Over the past few years, the Annual Conference as a continuation of successful
international meetings has become a major, well-established scientific event. It was
originally initiated by Prof. I.J. Rudas from Óbuda University Budapest (the former
Bánki Donát Polytechnic, Budapest Tech) in 1997 with the aim to create a scientific
platform for regular meetings of leading specialists in the field of Intelligent Engi-
neering Systems from all around the world, with particular focus on the region of
Central Europe.
At the 15th International Conference, 80 peer-reviewed papers were presented.
The increased interest of young researchers and enthusiasm of the participants of the
International Conference has clearly proved the value of this annual event as a place
to share inspiration and ideas. In terms of the age range of authors and co-authors, it
was very pleasing to note a much higher level of participation of young researchers
and scientists in comparison with the previous annual conferences. The book, which
is structurally divided into five chapters, contains a selection of papers targeted at
interested readers, rather than a full coverage of the problem area discussed in the
framework of the International Symposium.
The papers included in the first chapter deal with the following issues: “Diag-
nostics, performance monitoring and condition - based maintenance”: sufficient de-
sign conditions for fuzzy-observer-based residual generator design for a class of
VI Preface
continuous-time nonlinear systems represented by Takagi-Sugeno models; applica-

tion of two Iterative Learning Control (ILC) structures to the position control of 3D
crane systems; description of algorithms used for the determination of pitch dis-
tances in the calibration of optical scales on a calibration machine.
In the second chapter called “New Approaches for Mechatronics, Modeling and
Human Knowledge” theoretical knowledge of mathematical modeling is used in
practical applications in the fields of mechatronics, mechanics, aeronautics, various
special structures, particular robotic systems and knowledge. The chapter contains
the design of an intelligent supervisory system usable in the area of engineering
applications of small turbojet engines, model representation that applies to human
requests based active knowledge and assists handling information for the lifecycle
of a product as a contribution to Product Lifecycle Management (PLM) technol-
ogy, proposal of an engineering model for automation related topics, design of a
rectilinear motion model of a snake-like robot.
The third chapter, “Computational Intelligence, Web Technologies and Data Min-
ing”, discusses software solutions based on various methods of computational in-
telligence also by the use of progressive web technologies and data mining. The
chapter contains analysis of weights aging in social networks.
The penultimate part of the book, “Methodologies, methods and algorithms”,
contains new methods and brings new knowledge about modification of existing
procedures and algorithms of computational intelligence, particularly in the follow-
ing areas: generalized neural networks, higher order single value decomposition,
fuzzy mathematics, fuzzy inference systems, p-graphs, differential evolution algo-
rithms and anytime systems.
Articles in the chapter also describe applications of the mentioned methodologies
in the fields like microstrip filters design, pipeline computer data flow systems and
logistics. New approaches in all these areas offer new and progressive insights to
solution of particular real-world engineering problems.
As in many other fields, biomedical engineers benefit from the use of compu-
tational intelligence (CI) tools to solve a wide range of nonlinear and stochastic
problems. In the 5th chapter, the papers selected present current BMI research and
development, where BMI technologies and systems are applied using CI techniques
implemented by modern hardware and software solutions. The chapter is concep-
tually based on research and development in medical or biomedical diagnostics;
application of thermography in evaluation of carpal tunnel syndrome (CTS), and for
architecture of referential thermographic atlas of the human body, and metrotomog-
raphy as a non-destructive tool for biomedical testing and measuring of materials
and products. The second part of the chapter deals with intelligent titanium im-
plants, where magnetic microwires were applied to allow wireless and non-invasive
screening of the selected parameters of the human body, and surgical instrument
with novel bipolar multi-electrode radiofrequency ablation designed to protect a
significant part of healthy tissue and reduce the surgical time as well.
The editors wish to express their sincere gratitude to all the authors of the contri-
butions for their dedication, commitment and publishing the results of their scientific
observations. Much appreciation and thanks also go to Mr. MEng. Vladimı́r Gašpar,
Preface VII
a full-time PhD student of the Department of Cybernetics and Artificial Intelligence

FEEaI TU of Košice for his editorial assistance in preparing this collection of papers
and making significant effort to ensure the volume is published on time.
We would also like to tender our grateful thanks and acknowledgement to the
Scientific Grant Agency of the Ministry of Education, Science, Research and Sports
of the Slovak Republic (MESRS) (VEGA) and the Culture and Education Grant
Agency of the MESRS SR (KEGA) whose financial support of the projects has
made the publication of this book possible:
• VEGA č. 1/0298/12 “Digital control of complex systems with two degrees of
freedom”,
• KEGA č. 018TUKE-4/2012 “Progressive methods of education in the area of
control and modeling of complex systems object oriented on aircraft turbo-
compressor engines”,
• ITMS 26220120060 “Center for research of control of technical, environmen-
tal and human risks for permanent development of production amd products in
mechanical engineering”,
• STIMULY 0543/2010 “Research of new diagnostic methods in invasive implan-
tology”.
April 12, 2012 Ladislav Madarász

Jozef Živčák
Technical University of Košice
Slovakia
Contents
Part I: Diagnostics, Performance Monitoring and

Conditional – Based Maintenance
Residual Generator for a Class of Nonlinear Systems Described by
Takagi-Sugeno Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Dušan Krokavec, Anna Filasová, Vratislav Hladký
Lead-Lag Controller-Based Iterative Learning Control Algorithms
for 3D Crane Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Radu-Emil Precup, Florin-Cristian Enache, Mircea-Bogdan Rădac,
Emil M. Petriu, Stefan Preitl, Claudia-Adina Dragoş
Algorithms for Pitch Distance Determination . . . . . . . . . . . . . . . . . . . . . . . 39
Gyula Hermann
Objects for Visualization of Process Data in Supervisory Control . . . . . . 51
Iveta Zolotová, Roman Mihal’, Rastislav Hošák
Mesurement of Energy Flows and CO2 Emissions Balance of the
Low-Potential Heat Source in Combination with a Cogeneration
Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Natália Jasminská
Part II: New Approaches in Mechatronics, Modeling and

Human Knowledge
Intelligent Supervisory System for Small Turbojet Engines . . . . . . . . . . . . 85
Rudolf Andoga, Ladislav Madarász, Tomáš Karol’, Ladislav Főző,
Vladimı́r Gašpar
Product Definition Using Knowledge in Human Influence
Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
László Horváth, Imre J. Rudas
X Contents
Usage Dependent Rehabilitation and Maintenance Model for Special

Engineering Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Kornélia Ambrus-Somogyi, András Bakó
Course and Engineering Model Integration for Teaching Automation . . . 135
József Gáti, Gyula Kártyás, Marián Bučko
Dynamics and Control of a Snake Robot Rectilinear Motion . . . . . . . . . . 151
Alexander Gmiterko, Michal Kelemen, Ivan Virgala, Róbert Surovec,
Martina Vacková, Erik Prada
Part III: Computational Intelligence, Web Technologies and

Data Mining
Smart and Cooperative Neighbourhood for Spatial Routing in
Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Jan Nikodem, Zenon Chaczko, Maciej Nikodem, Ryszard Klempous
Elimination of Dynamic Errors of Thermocouples in Aircraft Engines
Using Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
František Adamčı́k, Rudolf Andoga, Ladislav Madarász, Peter Krajňák
Special Aspects of Teaching in Virtual Learning Environments . . . . . . . . 195
Andrea Tick
Weights Aging in Social Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Gabriel Tutoky, Ján Paralič
Part IV: Methodologies, Methods and Algorithms

The HOSVD Based Canonical Form of Functions and Its
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
András Rövid, László Szeidl, Péter Várlaki
Anytime Extension of Generalized Fuzzy Neural Network Models
with Non-singleton Consequents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Annamária R. Várkonyi-Kóczy
On Probabilistic Correlation Coefficients for Fuzzy Numbers . . . . . . . . . 249
Robert Fullér, István Á. Harmati, Péter Várlaki
Differential Evolution Algorithm in Optimal Design of Microstrip
Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Juraj Paleček, Peter Vestenický, Martin Vestenický, Juraj Spalek
P-Single Operators in Pipeline System of DF-KPI Architecture . . . . . . . . 277
Liberios Vokorokos, Norbert Ádám, Branislav Madoš
Contents XI
Potential Application of P-Graph-Based Workflow in Logistics . . . . . . . . 293

József Tick
Part V: Softcomputing in Biomedical Engineering

Linear Surgical Instrument for Bipolar Multielectrode
Radiofrequency Ablation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
Marek Penhaker, Petr Vavra, Milada Hlavackova, Jan Kracmar,
Tadeusz Sikora, Lukas Prokop
Biomedical Applications of Diagnostics and Measurements by
Industrial Computer Tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
Radovan Hudák, Jozef Živčák, Peter Kat’uch, Bruno Goban
Infrared Thermography in the Evaluation of Carpal Tunnel
Syndrome . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
Mária Tkáčová, Radovan Hudák, Jozef Živčák, Ivana Živčáková
Determination of Conditions and Positions for Obtaining
Thermographic Atlas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
Mária Tkáčová, Radovan Hudák, Jozef Živčák, Patrı́cia Čekanová,
Štefan Medvecký
Determining the Calibration Element Location in Planning of

Surgical Interventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
Monika Michalı́ková, Teodor Tóth, Lucia Bednarčı́ková, Jozef Živčák
Application of Magnetic Microwires in Titanium
Implants – Conception of Intelligent Sensoric Implant . . . . . . . . . . . . . . . 413
Radovan Hudák, Rastislav Varga, Jozef Živčák, Jozef Hudák, Josef Blažek,
Dušan Praslička
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435

Part I
Diagnostics, Performance
Monitoring and Conditional – Based
Maintenance
Residual Generator for a Class of Nonlinear
Systems Described by Takagi-Sugeno Models
Dušan Krokavec, Anna Filasová, and Vratislav Hladký
Technical University of Košice, Faculty of Electrical Engineering and Informatics,

Department of Cybernetics and Artificial Intelligence, Letná 9, 042 00 Košice, Slovakia
{dusan.krokavec,anna.filasova,vratislav.hladky}@tuke.sk;
http://web.tuke.sk/kkui
Abstract. The paper presents the principle of fuzzy-observer-based residual gene-

rator design for a class of continuous-time nonlinear systems represented by Ta-
kagi-Sugeno models, and measurable, as well as un-measurable premise variables.
The problem addressed can be indicated as approach giving sufficient design con-
ditions outlined in the terms of linear matrix inequalities, and possess a global sta-
ble structure. Simulation results illustrate the design procedures and demonstrate
the basic performances of the proposed residual generator structures.
Keywords: Fault detection, residual generators, fuzzy observers, convex optimi-

zation, linear matrix inequalities.
1 Introduction
The essential aspect for designing fault-tolerant control requires the conception of
diagnosis procedure that can solve the fault detection and isolation problem. This
procedure composes residual signal generation (signals that contain information
about the failures or defects) followed by their evaluation within decision func-
tions. Residuals are derived from implicit information in functional relationships,
which exist between measurements taken from the process, and information ob-
tained from a process model. Consequently, model-based diagnosis use models to
obtain residual signals that are as a rule zero in the fault free case, and non-zero
otherwise, and faults are detected usually by setting a threshold on a residual sig-
nal. Research in fault-tolerant control has attracted many investigators, and is now
the subject of widely scattered publications (see e.g. [2], [3], [4], [9]).
As it is well known, observer design is a hot research field owing to its particular
importance in observer-based fault detection and isolation. The nonlinear system
theory principles using Lipchitz condition has emerged as a method capable of use in
state estimation [8], [17], although Lipschitz condition is a strong restrictive condition
which many classes of systems may not satisfy. An alternative to design an observer
for nonlinear systems is fuzzy modeling approach which benefits from the advantages
of the approximation techniques approximating nonlinear system model equations.
Using the Takagi–Sugeno (TS) fuzzy model [15] the nonlinear system is
represented as a collection of fuzzy rules, where each rule utilizes the local
L. Madarász et al. (Eds.): Aspects of Computational Intelligence, TIEI 2, pp. 3–23.

springerlink.com © Springer-Verlag Berlin Heidelberg 2013
4 D. Krokavec, A. Filasová, and V. Hladký
dynamics by a linear system model. Since TS fuzzy models can well approximate
a large class of nonlinear systems, the TS model based approach can apprehend
the nonlinear behavior of a system while, keeping the simplicity of affine linear
models. However, the advantage of the TS type fuzzy models is property of utili-
zation of the state representation, and exploit the maximum of the potential rela-
tive to this methodology (see e.g. [12], [13], [16], [18]).
The state observers, based on fuzzy TS systems model, are usually realized in
terms of the parallel distributed compensation scheme, incorporating interpolations
between local Luenberger observers [1]. Once constructing the fuzzy models to de-
scribe the behavior of the nonlinear dynamic system, the same set of normalized
membership (activating) functions is used in nonlinear observer multi-model
structures. This type of nonlinear observers so gives multi-model structure, which
stability conditions rely on the feasibility of an associated system of linear matrix in-
equalities (LMI). Since large number of local models of the nonlinear system im-
plies a large number of LMIs in the design, it may limit the use of these structures.
Using the state observer based on fuzzy TS systems model, different residual
generator structures can be designed, and most of design principles (see e.g. [5],
[7]) also apply the linear matrix inequality technique. To achieve robustness, TS
fuzzy observers can be combined with classical sliding mode technique, where the
fault reconstruction has been achieved.
The main contribution of the paper is to present one principle for designing ob-
server-based residual generators for continuous-time nonlinear MIMO systems
approximated by a TS model. Constructing the fuzzy models to describe the beha-
vior of the nonlinear dynamic system, a set of observers is combined by fuzzy
rules to estimate the system state vector. Based on the observers, a residual gene-
rator is developed for guaranteeing the fault detection performance. Used structure
is motivated by the need for robustness in residual generators, and the stability of
the observer is established while being based on the work presented in [10], [11].
By using a quadratic Lyapunov function, the convergence conditions are ex-
pressed in the form of a set of linear matrix inequalities (LMI).
The remainder of this paper is organized as follows. In Section 2 the general
structure of TS models is briefly described, and in Section 3 the system stabilizing
principle is presented. The nonlinear observers for systems with measurable pre-
mise variables is given in Section 4, where TS fuzzy observer algorithm for com-
puting the state estimation, as well design conditions are derived. In Section 5, the
conditions for designing a globally asymptotically stable estimator based on TS
are proven, to be acceptable if the premise variables are un-measurable. Section 6
gives a numerical example to illustrate the effectiveness of the proposed approach,
and to confirm the validity of the fault residual generator scheme. The last section
draws conclusion remarks and some future directions.
2 System Model
The systems under consideration can be understood as one class of multi-input and
multi-output nonlinear (MIMO) dynamic systems. In state-space form this class of
dynamic systems is represented as
Residual Generator for a Class of Nonlinear Systems 5
q (t ) = a(q (t )) + Bu(t ) , (1)
y (t ) = Cq (t ) (2)
z (t ) = C z q (t ) (3)
where q(t ) ∈  , u(t ) ∈  , y(t ) ∈  , and z(t ) ∈  are vectors of the state, in-
n r m r
put, output, and objective variables, respectively, and B ∈  n× r , C ∈  m× n , and

Cz ∈ r×n are real matrices.
It is assumed that a(q(t)) is a vector function of the vector variable q(t), is
bounded in associated sectors, i.e. in the fuzzy regions where it is assumed the
system will operate within, and takes the value a(0) = 0. It is considered that the
number of the nonlinear terms in a(q(t)) is p, and that there exists a set of nonli-
near sector (partition) functions of these properties
 w lj (θ (t )), j = 1, 2, , k , l = 1, 2, , p,


 k
 w l1 (θ (t )) = 1 −  w lj (θ (t )), (4)
 j=2
 w (θ (t )) = w (θ (t )),
 lj lj j
where k is the number of sector functions, and
θ (t ) = θ1 (t ) θ 2 (t )  θ q (t )  (5)
is the vector of known premise variables. It is supposed that the premise variables
depend on the state variables that may not be available for measurement (un-
measurable premise variables).
Using a TS model, the conclusion part of a single rule consists no longer of a
fuzzy set [13], but determines a function with state variables as arguments, and the
corresponding function is a local function for the fuzzy region that is described by
the premise part of the rule. Thus, using linear functions, a system state is de-
scribed in fuzzy regions by linear models, and at the region boundaries a suitable
interpolation is used between the corresponding local models.
Combining the set of sector functions, the system state equation is inferred as
follows
k k
  w
h =1 j =1
1h (θi (t ))  w pj (θ j (t )) Ωi j
q (t ) = k k
, (6)
 h =1
  w1h (θi (t ))  w pj (θ j (t ))
j =1
Ωh j = Aq(t ) + Ah j q(t ) + Bu(t ) (7)
where Ωh j is the linear model associated with the (h...j) combination of sector
function indexes. It is evident that the fuzzy model is achieved by fuzzy amalga-
mation of the linear systems models.
The aggregated functions set wi (θ (t )), i = 1,2,, s, s = 2 can be calculated
k
from all combinations of the sector functions (4), e.g. ordered as follows
w1 (θ (t )) = w11 (θ1 (t )) w p1 (θ1 (t )) ,

 (8)
w s (θ (t )) = w1k (θ k (t )) w pk (θ k (t )) .
Constructing the function set {w (θ (t )), i = 1, 2, s, s = 2 }

i
k
from all combina-
tions of the sector functions gives
 w (θ (t )) Ω (t )
i i s
q (t ) = i =1
s
=  h i (θ (t )) Ωi (t ) , (9)
 w i (θ (t ))
i =1
i =1
Ωi = ( A + Ai ) q(t ) + Bu(t ) (10)
where
w i (θ (t ))
h i (θ (t )) = s
(11)
 w (θ (t ))
i =1
i
is the i-th aggregated normalized membership function satisfying conditions

s
0 ≤ h i (θ (t )) ≤ 1,  h (θ (t )) = 1 ∀i ∈ {1, , s} .
i =1
i (12)
n×n
Therefore, the fuzzy approximation of (1) leads to (9), (10) where A + Ai ∈ 
is the Jacobian matrix of a(q(t)) with respect to q (t ) = qi , and qi is the center of
the i-th fuzzy region described by the set of sector functions (4).
Thus, the TS fuzzy model of (1), (2) is considered as
s
q (t ) =  h i (θ (t )) ( ( A + Ai ) q (t ) + Bu(t ) ) , (13)
i =1
y (t ) = Cq (t ) , (14)
and can be interpreted as a combination of s linear sub-models through the set of

normalized membership (activating) functions {hi (θ (t )), i = 1, 2,, s} .
Remark 1. The matrices A, B, C , C z are the same for all local models.
Remark 2. The nonlinear part a(q(t)) is locally stable, the pair (a(q(t)), C) is local-
ly observable, C , C z ( B ) are of full row (column) rank, and
s
a(q (t )) =  h i (θ (t )) ( A + Ai ) q (t ) . (15)
i =1
3 System Stabilization
Definition 1. Considering (13), and using the same set of membership function,
the nonlinear fuzzy state controller is defined as
s
u(t ) = −  h j (θ (t )) K j q (t ) (16)
j =1
Theorem 1. The equilibrium of the fuzzy system (13), (14) controlled by the fuzzy
controller (16) is global quadratically stable if there exist symmetric positive
definite matrices U , V ∈  n× n , a positive definite matrix T ∈  n× n , symmetric ma-
trices Yij = Y ji ∈ n×n , and matrices W j ∈ r×n such that
T > 0, U = U T > 0, V = V T > 0 (17)
( A + Ai )V + V ( A + Ai )T − BW j − W jT BT − Yij ∗ 
 <0 (18)
 T − U + ( A + Ai )V − BW j −2U 
Y11 Y12  Y1s 

Y  Y2 s 
 12 Y22 >0 (19)
     
 
Y1s Y2 s  Yss 
for h i (θ(t ))h j (θ(t )) ≠ 0, i, j = 1,2,, s . The set of gain matrices is given as
follows
K j = W jV −1 , j = 1, 2,, s (20)
Hereafter, * denotes the symmetric item in a symmetric matrix.

Proof. Considering (13), (14), and (16) then (13) with Ai = A + Ai implies

q (t ) =  h i (θ (t ))h j (θ (t )) ( Ai − BK j ) q (t )
s s
(21)
i =1 j =1
Writing (13) in the form
q (t ) −  h i (θ (t ))h j (θ (t )) ( Ai − BK j ) q (t ) = 0
s s
(22)
i =1 j =1
n×n
then with arbitrary symmetric regular matrices S1 , S2 ∈  it yields
 
(q (t ) S1 + q T (t ) S 2 )  q (t ) −   h i (θ (t ))h j (θ (t )) ( Ai − BK j ) q (t )  = 0
s s
T
(23)
 i =1 j =1 
Defining the quadratic positive definite Lyapunov function
v(q(t )) = qT (t ) Pq(t ) > 0 (24)
where P ∈  n× n is a symmetric positive definite matrix, then it yields
v(q(t )) = qT (t ) Pq(t ) + qT (t ) Pq(t ) < 0 (25)

Thus, adding (23) as well as the transposition of (23) to (25) it can be obtained
v(q(t )) = q T (t ) Pq (t ) + qT (t ) Pq (t ) +
 T 
+  q T (t ) − qT (t ) h i (θ (t ))h j (θ (t )) ( Ai − BK j )  ( S1q (t ) + S 2 q (t ) ) +
s s
 i =1 j =1 
  (26)
+ ( qT (t ) S1 + q T (t ) S 2 )  q (t ) −  h i (θ (t ))h j (θ (t )) ( Ai − BK j ) q (t )  −
s s
 i =1 j =1 
s s s s
−qT (t ) h i (θ (t ))h j (θ (t )) X ij q(t ) < −qT (t ) h i (θ (t ))h j (θ (t )) X ij q(t ) < 0
i =1 j =1 i =1 j =1
where {X ij = X ijT ∈  n× n , i, j = 1, 2, , s} is the set of symmetric matrices. Now

(26) can be compactly rewritten as
s s
v(q (t )) =  h i (θ (t ))h j (θ (t ))q •T (t )Pij• q • (t ) < 0 (27)
i =1 j =1
q •T (t ) = qT (t ) q T (t )  (28)
 − S ( A − BK ) − ( A − BK )T S − X ∗ 
P = • <0
1 i j i j 1 ij
(29)
 P + S1 − S2 ( Ai − BK j ) 2 S2 
ij
T 

Since S1, S2 are regular, defining the congruence transform matrix
T1 =  S1−1 S2−1  (30)
then pre-multiplying left-hand side and right-hand side of (29) by T1 gives
 − ( A − BK ) S −1 − S −1 ( A − BK )T − S −1 X S −1 ∗ 
 i j 1 1 i j 1 ij 1
<0 (31)

 S2−1 PS1−1 + S2−1 − ( Ai − BK j ) S1−1 −1 
2 S2 
and using the substitutions

T = S2−1 PS1−1 , V = − S1−1 , U = − S2−1 , Yij = VX ijV , W j = K jV (32)
(31) implies (18). Since evidently (26) is conditioned by

s s
Z ( θ (t ) ) =  h i (θ (t ))h j (θ (t )) X ij > 0 (33)
i =1 j =1
Z v ( θ (t ) ) = VZ ( θ (t ) )V =
Y11 Y12  Y1s   h1 (θ (t )) 
Y Y22  Y2 s   h 2 (θ (t ))  (34)
= [ h1 (θ (t )) h 2 (θ (t ))  h s (θ (t )) ]  12 >0
       
  
Y1s Y2 s  Yss   h s (θ (t )) 
respectively, then (34) implies (19). This concludes the proof.
4 Fuzzy Observer with Measurable Premise Variables

Standard form of a fuzzy observer to the system model (13), (14) for measurable
premise variables is constructed as follows [10]
s
qe (t ) =  h i (θ (t )) ( ( A + Ai ) qe (t ) + Bu(t ) + J i ( y (t ) − ye (t ) ) ) , (35)
i =1
ye (t ) = Cqe (t ) (36)
where qe (t ) ∈  , and {J i ∈  n× m , i = 1, 2, , s} is the set of the gain matrices.

n
Theorem 2. The fuzzy observer (16), (17) is stable if there exist a positive definite
n×m
symmetric matrix P > 0, P ∈  n×m , and matrices Zi ∈  , i = 1, 2,, s such
that
P = PT > 0 , (37)
( A + Ai ) P + P ( A + Ai ) − ZiT C T − Zi C < 0 ∀i. (38)
Then Ji can be computed as
Ji = P −1 Zi . (39)
Proof. Introducing the estimation error between (13) and (16) as follows
e (t ) = q (t ) − qe (t ) (40)
and taking the time derivative of e(t) then it can be directly obtained
s
e (t ) = q (t ) − qe (t ) =  hi (θ (t )) ( ( A + Ai ) e (t ) − J i Ce (t ) ) , (41)
i =1
s
e (t ) =  hi (θ (t )) Aei e (t ) , (42)
i =1
respectively, where the observer system matrices are introduced as
Aei = A + Ai − J i C , i = 1, 2, , s . (43)

Defining the quadratic positive definite Lyapunov function of the form
v(e(t )) = eT (t ) Pe(t ) (44)
where P > 0, then after evaluation of the derivative with respect to t of (25) it is
obtained
v(e(t )) = eT (t ) Pe(t ) + eT (t ) Pe(t ) . (45)

Substituting of (23) in (26) gives
s s
v(e (t )) = eT (t ) P  hi (θ (t )) Aei e (t ) + eT (t )  hi (θ (t )) AeiT Pe (t ) , (46)
i =1 i =1
v(e (t )) = eT (t )  hi (θ (t )) ( PAei + AeiT P ) e (t ) ,

s
(47)
i =1
respectively. It is evident that (28) be negative if there exist a set of gain matrices
{J i ∈  n×m , i = 1, 2,, s} and a common symmetric positive definite matrix
P ∈  n× n such that
( A + Ai − J i C ) P + P ( A + Ai − J i C ) < 0 ∀ i .
T
(48)
Setting
PJ i = Z i (49)
then (29) implies (19). This concludes the proof.
5 Fuzzy Observer with Un-measurable Premise Variables
5.1 Basic Preliminaries

Proposition 1. (Quadratic performance) If a disturbance transfer matrix is
G zv ( s ) then for γ > 0, γ ∈  is
 (z (r ) z (r ) − γ v T (r )v (r ) ) dr > 0
∞
T
(50)
0
if γ > 0 is square of the infinity norm of the transfer matrix Gzv ( s ) .
Proof. It is evident that if
z ( s ) = Gzv ( s )v ( s ) (51)
is the transfer function matrix of dimension m× r and z ( s ), v ( s ) stands for the

Laplace transform of n dimensional objective vector and r dimensional distur-
bance vector, respectively, then (32) implies
z( s) ≤ Gzv ( s) v (s) (52)
and the infinity norm property states
1
Gvz ( s ) ∞
≤ Gvz ( s ) ≤ r Gvz ( s ) ∞
. (53)
m
Using notation Gvz (s) ∞ = γ , where γ is the value of the infinity norm of the
transfer matrix Gvz ( s ) , then the inequality (34) can be rewritten as
1 z ( s ) Gvz ( s )
0< ≤1< ≤ ≤ r . (54)
m γ v ( s ) γ
Thus, based on Parceval's theorem, (35) gives

1
∞ T 2
  z (r ) z (r ) dr 
z ( s) 0 
0< = , (55)
γ v ( s )
1
 ∞
 2
γ   vT (r )v (r ) dr 
0 
and subsequently
∞ ∞
0
z T (r ) z (r ) dr − γ  v T (r )u(r ) dr > 0 .
0
(56)
It is evident that (37) implies (31). This concludes the proof.
5.2 Fuzzy Observer Design

The proposed structure of the fuzzy observer is
s
qe (t ) = Aqe (t ) + Bu(t ) + J ( y (t ) − ye (t ) ) +  hi (θe (t )) Ai qe (t ) , (57)
i =1
ye (t ) = Cqe (t ) (58)
where qe (t ) ∈  is the estimation of the system state vector, and J ∈  n× m is the

n
observer gain matrix.
Theorem 3. The fuzzy observer (38), (39) is asymptotically stable if there exist a
positive definite symmetric matrix P ∈  n× n , a matrix Z ∈  n × m , and a positive
scalar γ ∈  such that
P = PT > 0, γ > 0 , (59)
 I n + PA + AT P − ZC − C T Z T PAv 
 <0 (60)
 * −γ I sn 
sn× sn
where I sn ∈  is the identity matrix, and
Av = [ A1 A2  As ] . (61)
Then the observer gain matrix J can be computed as
J = P −1 Z . (62)
Hereafter, * denotes the symmetric item in a symmetric matrix.
Proof. Using the observer estimation error (21) and taking the time derivative of
e(t) it can be directly obtained
s
e (t ) = ( A − JC ) e (t ) +  Ai ( hi (θ (t ))q (t ) − hi (θe (t ))qe (t ) ) , (63)
i =1
e(t ) = ( A − JC ) e(t ) + Av v (t ) = Ae e(t ) + Av v (t ) , (64)
respectively, where
Ae = A − JC , (65)
the disturbance input matrix is introduced in (42), the disturbance vector v(t) is
v T (t ) =  v1T (t ) v2T (t )  v sT (t )  , (66)
viT (t ) = hi (θ(t ))q(t ) − hi (θe (t ))qe (t ) (67)
and i ∈{1, 2,, s} .

Defining the quadratic Lyapunov function as follows
t
v(e (t )) = e T (t ) Pe (t ) +  ( eT (r )e (r ) − γ v T (r )v (r ) ) dr (68)
0
and considering (31), it is obvious that with infinity norm of the transfer matrix
Gev ( s) equals to γ , and with a positive definite symmetric matrix P ∈  n× n ,
then (49) be a positive function.
After evaluation the derivative of (49) with respect to t it can be obtained
v(e(t )) = eT (t ) Pe(t ) + eT (t ) Pe(t ) + eT (t )e(t ) − γ vT (t )v(t ) < 0 (69)
and substituting (45) in (50) gives

v(e (t )) = e T (t ) AeT Pe (t ) + eT (t ) PAe e (t ) +

(70)
+v (t ) A Pe (t ) + eT (t ) PAv v (t ) + eT (t )e (t ) − γ v T (t )v (t ) < 0 .
T T
v
Defining the composite vector
e T (t ) = e T (t ) v T (t )  (71)
then (51) can be rewritten as
v(e(t )) = eT (t ) P  e (t ) < 0 (72)

where
 I + PAe + AeT P PAv 

P =  n <0, (73)
 ∗ −γ I sn 
since it is evident that (53) be negative if P  < 0 .

Inserting (46) into (54) results in
 I n + P ( A − JC ) + ( A − JC )T P PAv 
 <0 (74)
 ∗ −γ I sn 
and setting
PJ = Z (75)
(55) implies (41). This concludes the proof.
6 Fault Residuals
6.1 Residual Generator

The residual signal can be considered naturally in the dependence of system out-
put and observer output as follows
r (t ) = Xqe (t ) + Yy (t ) . (76)
Thus, using (2), (21) it gives the result
r (t ) = X ( q(t ) − e(t ) ) + YCq(t ) = ( X + YC ) q(t ) − Xe(t ) . (77)

To obtain residual signals independent on the system state vector, it is neces-

sary to ensure that
X + YC = 0 . (78)
Equality (59) be satisfied e.g. by setting
X = −C , Y = Im . (79)
Therefore, signal of the residual generator (57) can now be written as
r (t ) = y (t ) − Cqe (t ) = y (t ) − ye (t ) , (80)
i.e. the residual signal is given as an observer innovation.
To prefer some residual properties r(t) can be defined as
r (t ) = Vr ( y(t ) − Cqe (t ) ) (81)

v× n
where matrix Vr ∈  , v ≤ rank C , is a weighing matrix.
In this sense, the residual parts associated with the linear sub-models can be de-
rived as well from the innovation signals, i.e. potentially defined as
ri (t ) = Vr e y (t ) = Vr ( y(t ) − Cqe (t ) ) = Vr Cei (t ) . (82)
Since (23) gives
s
e (t ) =  hi (θ (t )) Aei e (t ) (83)
i =1
then it yields
s s
r (t ) = Vr Ce (t ) =  hi (θ (t ))Vr CAei e (t ) =  hi (θ (t ))Vr Cei (t ) (84)
i =1 i =1
that means
s
r (t ) =  hi (θ (t ))ri (t ) . (85)
i =1
This principle can be generalized as follows
ri (t ) = Vri ( y(t ) − Cqe (t ) ) , (86)

s
r (t ) =  hi (θ (t ))ri (t ) (87)
i =1
v× n
where different weighting matrices Vri ∈  , v ≤ rank C , i = 1, 2,, s can be de-
fined to prefer e.g. directional properties of residual parts given by single linear
sub-models.
6.2 Structured Residual Generators of Sensor Faults

To design structured residual generators of sensor faults based on the state estima-
tors, all actuators are assumed to be fault-free and each estimator is driven by all
system inputs and all but one system outputs. In that sense it is possible according
with given nominal fault-free system model (13), (14) to define the set of struc-
tured estimators for k = 1, 2, , m as follows
s
q ke (t ) = Aqke (t ) + Bu(t ) + J k ( yk (t ) − yke (t ) ) +  hi (θe (t )) Ai qke (t ) , (88)
i =1
yk (t ) = C k qk (t ) , (89)
yke (t ) = C k qke (t ) , (90)
C k = Tk C , Tk = I m∅k (91)
where Tk is obtained by deleting the k-th row in identity matrix Im , qke (t ) ∈  is

n
n×( m−1)
the k-th estimation of the state, and J k ∈  is the gain matrix of the k-th
observer.
Theorem 4. All fuzzy observers from the set (69)-(72) are asymptotically stable if
n×n
for k = 1, 2, , m there exist positive definite symmetric matrices Pk ∈  , ma-
n×( m −1)
trices Zk ∈  , and positive scalars γ k ∈  such that
Pk = PkT > 0, γ > 0 , (92)
 I n + Pk A + AT Pk − Z k C k − C kT Z kT Pk Av 
 <0 (93)
 ∗ −γ I sn 
sn× sn
where I sn ∈  is the identity matrix, and Av is defined in (42).
Then the observer gain matrix J k can be computed as
J k = Pk−1 Zk ∀k . (94)
Proof. Defining the k-th observer estimation error as follows
ek (t ) = q (t ) − qke (t ) (95)
then
s
ek (t ) = ( A − J k C k ) ek (t ) +  Ai ( hi (θ (t ))q (t ) − hi (θek (t ))qke (t ) ) , (96)
i =1
ek (t ) = ( A − J k Ck ) ek (t ) + Av vk (t ) = Aek e(t ) + Av vk (t ) , (97)
respectively, where
Aek = A − J k C k , (98)
the disturbance input matrix is introduced in (42), the disturbance vector vk (t ) is
vkT (t ) = v1Tk (t ) v2Tk (t )  vskT (t )  , (99)
vikT (t ) = hi (θ(t ))q(t ) − hi (θek (t ))qek (t ) (100)
and i ∈ {1, 2,, s} , k ∈ {1, 2,, m} .

Defining the k-th quadratic Lyapunov function as follows
t
v(ek (t )) = ekT (t ) Pk ek (t ) +  ( ekT (r )ek (r ) − γ k v kT (r )vk (r ) ) dr (101)
0
and considering (31), it is obvious that with infinity norm of the transfer matrix
Gevk ( s ) equals to γ k , and with a positive definite symmetric matrix Pk ∈ 
n×n
,
then (81) is a positive function.
After evaluation the derivative of (49) with respect to t it can be obtained
v(ek (t )) = ekT (t ) Pk ek (t ) + ekT (t ) Pk ek (t ) + ekT (t )ek (t ) − γ k vkT (t )vk (t ) < 0 . (102)
Using analogy in (50) and (83), then exploiting (55) it can be obtained
 I n + Pk ( A − J k C k ) + ( A − J k C k )T Pk Pk Av 
 <0 (103)
 ∗ −γ k I sn 
and setting
Pk J k = Z k (104)
(84) implies (74). This concludes the proof.
Exploiting the model-based properties of the set of state estimators the set of
structured residual generators can be considered as
rk (t ) = Tk ( y(t ) − yek (t ) ) , ∀k . (105)
When all actuators are fault-free and a fault occurs in the l-th sensor the residuals
will satisfy the isolation logic
rk (t ) ≤ hk , k = l, rk (t ) > hk , k ≠ l . (106)
This residual set can only isolate a single sensor fault at the same time. The prin-
ciple can be generalized based on a regrouping of faults in such way that each re-
sidual will be designed to be sensitive to one group of sensor faults and insensitive
to others.
7 Illustrative Example
The nonlinear dynamics of the hydrostatic transmission was considered as in [6],
and this model was used for input/output signal generation.
The hydrostatic transmission dynamics is represented by a nonlinear fourth or-
der state-space model
q1 (t ) = − a11 q1 (t ) + b11u1 (t )
q 2 (t ) = − a22 q2 (t ) + b22 u2 (t )
q3 (t ) = a31q1 (t ) p (t ) − a33 q3 (t ) − a34 q2 (t )q4 (t )
q 4 (t ) = a43 q2 (t )q3 (t ) − a44 q4 (t )
where q1 (t ) is the normalized hydraulic pump angle, q2 (t ) is the normalized hy-

draulic motor angle, q3 (t ) is the pressure difference [bar], q4 (t ) is the hydraulic
motor speed [rad/s], p(t) is the speed of hydraulic pump [rad/s], u1 (t ) is the nor-
malized control signal of the hydraulic pump, and u2 (t ) is the normalized control
signal of the hydraulic motor. In given working point the model parameters are
a11 = 7.6923 a22 = 4.5455 a33 = 7.6054.10−4

a31 = 0.7877 a34 = 0.9235 b11 = 1.8590.103
a43 = 12.1967 a44 = 0.4143 b22 = 1.2879.103
Since the variables p(t ) ∈ 105, 300 and q2 (t ) ∈ 0, 1 are bounded on the pre-
scribed sectors then vector of the premise variables can be chosen as follows
θ (t ) = [θ1 (t ) θ2 (t )] = [ q2 (t ) p(t )] .
Thus, the set of nonlinear sector functions
b1 − q2 (t )
w11 (q2 (t )) = , b1 = 0, b2 = 1
b1 − b2
q2 (t ) − b2
w12 (q2 (t )) = = 1 − w11 (q2 (t ))
b1 − b2
c1 − p (t )
w21 ( p (t )) = , c1 = 105, c2 = 300
c1 − c2
p (t ) − c2
w22 ( p (t )) = = 1 − w21 ( p (t ))
c1 − c2
implies the next set of aggregated normalized membership functions
h1 (q2 (t ), p (t )) = w11 (q2 (t )) w21 ( p(t ))
h2 (q2 (t ), p(t )) = w12 (q2 (t )) w21 ( p (t ))
h3 (q2 (t ), p (t )) = w12 (q2 (t )) w12 ( p(t ))
h4 ( q2 (t ), p(t )) = w12 (q2 (t )) w22 ( p(t ))
The used sector functions of the premise variables take the forms as presented in
the Fig. 1.
The transformation of nonlinear differential equation systems into a TS fuzzy
system in standard form gives
 − a11 0 0 0  b11 0 
 0 − a22 0 0   0 b 
A + Ai =  , B= 22 
,
 a31ck 0 − a33 − a34 bl  0 0
   
 0 0 a43bl − a44  0 0
1 0 0 0 
 0 1 0 0 
C =  0 0 0.01 0  , C z =  
 0 0 0 −0.02 0 0 1 0 
with the associations
i = 1 ← ( l = 1, k = 1) i = 2 ← ( l = 2, k = 1)
i = 3 ← ( l = 2, k = 2 ) i = 4 ← ( l = 2, k = 2 )
w (q (t)) w (q (t))
11 2 12 2 w21(p(t)) w (p(t))
22
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 100 120 140 160 180 200 220 240 260 280 300 320
Fig. 1. Normalized sector function of θ1 (θ 2 )
The matrices P and Zi , i = 1, 2,3, 4 can be found by using convex optimization

techniques. Thus, solving (40), (41) with respect to the LMI matrix variables P, Z
using Self–Dual–Minimization (SeDuMi) package for Matlab [14], the feedback
gain matrix design problem in the constrained control was solved as feasible with
the results
11.7596   −62.3091 0.0000 0.0000 

 10.7835   0.0000 
P=  , Z =  0.0000 0.0000
 0.0143   0.0000 160.7955 0.0000 
   
 0.5689   0.0000 0.0000 −139.0731
γ = 55.8129
Inserting P and Zi into (43) the observer gain matrix was computed as follows
 −0.0530 0.0000 0.0000 

 0.0000 0.0000 0.0000 
J = 10 2  
 0.0000 112.5754 0.0000 
 
 0.0000 0.0000 −2.4445
which rise up a stable observer with eigenvalue spectrum

eigAe = {−2.3937 −4.5455 −5.3033 −112.5762} .
The matrix used in simulations was V = [10 0 0], the system was stabilized by
forced control policy, in the sense of static decoupling principle with respect to B
and C z [10]. Fig. 2 shows the set of solutions (the external system signal, the
fuzzy observer output signals, and the residual signal) according to changes in
p(t), where all values of p(t) belong to the prescribed sector interval (the left col-
umn), as well as in the fault regime (the right column) where the values of p(t) on
time interval t ∈ 40, 50 [s] don't belong to the prescribed sector of p(t) (an ex-
ternal action fault).
300 450
p(t) p(t)
280
400
260
240 350
220
300
200
180 250
160
200
140
150
120
100 100
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
time [s] time [s]
0.35 0.35
ze1(t) ze1(t)
z (t) ze2(t)
0.3 e2 0.3
0.25 0.25
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
time [s] time [s]
−5 −3
x 10 x 10
2 14
residual residual
1 12
0 10
−1 8
−2 6
−3 4
−4 2
−5 0
−6 −2
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
time [s] time [s]
Fig. 2. Residual response for the system in a fault-free, and in a fault regime
When an actuator fault appears the corresponding residual stays strongly differ-
ent from zero as it is computed and represented in Fig. 3.
8
residual
7
−1
0 10 20 30 40 50 60 70
time [s]
Fig. 3. Residual response for the system with the first actuator fault
8 Concluding Remarks
Modified design method is presented with aim to ensure robustness of fault resi-
dual generators for a class of nonlinear systems represented by Takagi–Sugeno
models. This goal is achieved by application of the bounded real lemma principle
in design conditions with respect to TS multi–model state observer for the system
with un-measurable premise variables. All presented versions of quadratic stability
conditions are derived in terms of optimization over LMI constraints using stan-
dard LMI numerical optimization procedures to manipulate the global observer
stability. Although the stability conditions based on use of the quadratic Lyapunov
function in a multi-model structure are conservative, as a common symmetric pos-
itive definite matrix verifying all Lyapunov inequalities is required, this principle
naturally exploits and extends the affine properties of TS fuzzy observers. Moreo-
ver, design task is solved with respect to measurable, as well as un-measurable
premise variables, using the same structure of residual generators.
It is significant to extend the presented approach to the case for fuzzy systems
with parametric perturbations, as well as to real convex polytopic-type uncertain-
ties, and to rise residual generator robustness. It is also encouraged to extend the
technique in fault tolerant control structures for TS fuzzy systems.
Acknowledgement. The work presented in this paper was supported by VEGA, Grant
Agency of Ministry of Education and Academy of Science of Slovak Republic under Grant
No. 1/0256/11. This support is very gratefully acknowledged.
References
[1] Bergsten, P., Palm, R., Driankov, D.: Observers for Takagi-Sugeno Fuzzy Systems.
IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics 32,
114–121 (2002)
[2] Blanke, M., Kinnaert, M., Lunze, J., Staroswiecki, M.: Diagnosis and Fault-Tolerant
Control. Springer, Berlin (2003)
[3] Chen, J., Patton, R.J.: Robust Model-Based Fault Diagnosis for Dynamic Systems.
Kluwer Academic Publishers, Norwell (1999)
[4] Ding, S.: Model-Based Fault Diagnosis Techniques. Design Schemes, Algorithms,
and Tools. Springer, Berlin (2008)
[5] Gao, Z., Shi, X., Ding, S.X.: Fuzzy State/Disturbance Observer Design for T-S Fuzzy
Systems with Application to Sensor Fault Estimation. IEEE Transactions on Systems,
Man, and Cybernetics, Part B: Cybernetics 38, 875–880 (2008)
[6] Gerland, P., Gross, D., Schulte, N., Kroll, A.: Robust Adaptive Fault Detection Using
Global State Information and Application to Mobile Working Machines. In: Confe-
rence on Control and Fault Tolerant Systems, Nice, France, pp. 813–818 (2010)
[7] Ichalal, D., Marx, B., Ragot, J., Maquin, D.: Design of Observers for Takagi-Sugeno
Discrete-Time Systems with Unmeasurable Premise Variables. An L2 approach. In:
Proceedings of the 17th World Congress IFAC, Seoul, Korea, pp. 2768–2773 (2008)
[8] Koshkouei, A.J., Zinober, A.S.I.: Partial Lipschitz Nonlinear Sliding Mode Observ-
ers. In: Proceedings of the 7th Mediterranean Conference on Control and Automation
MED 1999, Haifa, Israel, pp. 2350–2359 (1999)
[9] Krokavec, D., Filasová, A.: Dynamic Systems Diagnosis. Elfa, Košice (2007) (in
Slovak)
[10] Krokavec, D., Filasová, A., Hladký, V.: Design of Stable Fuzzy-Observer-Based Re-
sidual Generators for a Class of Nonlinear Systems. In: Proceedings of the 15th IEEE
International Conference on Intelligent Engineering Systems INES 2011, Poprad,
Slovakia, pp. 329–334 (2011)
[11] Krokavec, D., Filasová, A., Hladký, V.: TS Fuzzy Observer-Based Residual Genera-
tors for a Class of Nonlinear Systems. In: Proceedings of the 12th IEEE International
Symposium on Computational Intelligence and Informatics CINTI 2011, Budapest,
Hungary, pp. 45–50 (2011)
[12] Michels, K., Klawonn, F., Kruse, R., Nürnberger, A.: Fuzzy Control. Fundamentals,
Stability and Design of Fuzzy Controllers. Springer, Berlin (2006)
[13] Passino, K.M., Yurkovich, S.: Fuzzy Control. Addison-Wesley Longman, Berkeley
(1998)
[14] Peaucelle, D., Henrion, D., Labit, Y., Taitz, K.: User’s Guide for SeDuMi Interface
1.04. LAAS-CNRS, Toulouse (2002)
[15] Takagi, T., Sugeno, M.: Fuzzy Identification of Systems and its Applications to Modeling
and Control. IEEE Transactions on Systems, Man, and Cybernetics 15, 116–132 (1985)
[16] Tanaka, K., Wang, H.O.: Fuzzy Control Systems Design and Analysis. A Linear Ma-
trix Inequality Approach. John Wiley & Sons, New York (2001)
[17] Thau, F.E.: Observing the State of Nonlinear Dynamical Systems. International Jour-
nal of Control 17, 471–479 (1973)
[18] Wang, H.O., Tanaka, K., Griffin, M.F.: An Approach to Fuzzy Control of Nonlinear
Systems: Stability and Design Issues. IEEE Transactions on Fuzzy Systems 4, 14–23
(1996)
Lead-Lag Controller-Based Iterative Learning
Control Algorithms for 3D Crane Systems
Radu-Emil Precup1, Florin-Cristian Enache1, Mircea-Bogdan Rădac1,

Emil M. Petriu2, Stefan Preitl1, and Claudia-Adina Dragoş1
1
“Politehnica” University of Timisoara, Department of Automation
and Applied Informatics, Bd. V. Parvan 2, RO-300223 Timisoara
Romania radu.precup@aut.upt.ro, enacheflorin88@yahoo.com,
{mircea.radac,stefan.preitl,claudia.dragos}@aut.upt.ro
2
University of Ottawa, School of Electrical Engineering and Computer Science,
800 King Edward, Ottawa, ON, K1N 6N5 Canada
petriu@site.uottawa.ca
Abstract. This chapter deals with the application of two Iterative Learning Con-
trol (ILC) structures to the position control of 3D crane systems. The control sys-
tem structures are based on Cascade Learning (CL) and Previous and Current
Cycle Learning (PCCL) which improve the control system performance with fre-
quency domain designed lead-lag controllers for the x-axis and for the y-axis. The
parameters of continuous-time real PD learning rules which are also implemented
in real-world applications as lead-lag controllers are set such that to fulfill the
convergence conditions of CL and PCCL. Elements of anti-swing control for the
PCCL structure are discussed. Experimental results are given to solve the crane
position control problem of a 3D crane system laboratory equipment.
1 Introduction
The control systems for gantry crane systems play an important role in industrial
applications because of the good performance they must offer. These problems do
not represent simple task because of the Multi Input-Multi Output (MIMO) nonli-
near processes. The analysis of the recent literature shows that the control approach-
es to 3D crane systems are of great interest. Neural and fuzzy logic compensators are
introduced in anti-swing control schemes as shown in [1–3]. Time-optimal control
and visual feedback are applied in [4] to solve position and anti-swing control prob-
lems. The PID controller-based virtual environment teleoperation of hydraulic fore-
stry cranes is addressed [5]. An inertia theorem-based nonlinear controller with an
accelerated method based on projections to the goal is proposed in [6]. A MIMO
nonlinear tracking control structure is discussed in [7]. Feed-forward and input-
shaping techniques are suggested in [8–10]. The sliding mode control applied to
load swing suppression is analyzed in [11]. An interpolated delay-dependent gain
scheduling law is proposed in [12]. A fuzzy and CMAC architecture is investigated

26 R.-E. Precup et al.
in [13]. Genetic algorithms are employed in [14] to obtain the parameters in simple
models of 3D crane systems.
Iterative Learning Control (ILC) carries out the improvement of performance
indices (overshoot, settling time, etc.) of control systems executing repetitively the
same tasks in terms of several experiments (cycles, iterations). The information
from previous experiments is used in this context. Many learning rules are embed-
ded in ILC structures designed around the control systems whose performance are
improved [15–17]. ILC can be viewed as an optimization paradigm as pointed out
in [18], and thorough analyses concerning ILC are carried out in [19].
The performance improvement from an experiment to another one in ILC struc-
tures is ensured by using information acquired from previous experiments in terms
of inserting memory blocks. The repetitive tasks require the repetition of the con-
trol systems’ inputs (reference inputs and load disturbance inputs) and of the ini-
tial conditions as well, so the area of applications of ILC is constrained. However
this is not anymore a serious problem because, according to the classification giv-
en in [20], ILC can be combined in two ways with control loops:
• direct ILC, where ILC is used to determine the control signals or to contribute
to their computation [21, 22],
• indirect ILC, where the controllers produce the control signals and ILC is used
to update some parameters and/or variables for the control loop such that refer-
ence input or controller tuning parameters [23, 24].
This chapter is built upon the new control solutions which involve Cascade Learn-
ing (CL) and Previous and Current Cycle Learning (PCCL) and belong to the
category of direct ILC. These structures are defined in accordance with [17], and
they are considered in combination with control systems that contain frequency
domain designed lead-lag controllers proposed in [25] (for CL) and in [26] (for
PCCL). Extended experimental results with respect to those given in [25, 26] for
the position control problems related to the x-axis and to the y-axis are offered. In
addition, some elements of anti-swing control are discussed. The parameters of the
continuous-time real PD learning rules as lead-lag controllers are set such that to
fulfill the convergence conditions of learning processes in CL-based and PCCL-
based control system structures. The convergence conditions are inequalities
which employ H∞ norms calculated using a frequency domain approach. All con-
troller designs and experimental results are focused on a 3D crane system labora-
tory equipment that models industrial gantry crane systems [27].
This paper is structured as follows. The process models are discussed briefly in the
next section. Section 3 focuses on the ILC-based controller design procedures. Expe-
rimental results are given in Section 4. The conclusions are pointed out in Section 5.
2 3D Crane System Models

The following nonlinear state-space equations of the process in 3D crane systems
are obtained if no disturbance are considered and zero initial conditions are consi-
dered for all state variables except x1 [25, 26]:
Lead-Lag Controller-Based Iterative Learning Control Algorithms 27
x1 = x2 ,
x 2 = −T1 x2 − Tsy sgn( x2 ) − μ1 cos( x5 )[−T3 x10 − Tsz sgn( x10 )] + k1u1
+ k3 μ1 cos( x5 )u3 ,
x3 = x4 ,
x 4 = −T2 x4 − Tsx sgn( x4 ) − μ 2 sin( x5 ) sin( x7 )[−T3 x10 − Tsz sgn( x10 )]
+ k 2u 2 + k3 μ 2 sin( x5 ) sin( x7 )u3 ,
x5 = x6 ,
x 6 = −[T1 x2 − Tsy sgn( x2 )] sin( x5 ) / x9 + sin( x3 ) cos( x5 ) x82 / x9
+ cos( x5 ) cos( x7 )[k1 sin( x5 )u1 − k 2 cos( x5 ) sin( x7 )u2 − k3 μ 2 sin( x5 )
⋅ cos( x5 ) sin 2 ( x7 )u3 + k3 μ1 sin( x5 ) cos( x5 )u3 ] / x92 + μ 2 sin( x5 ) cos( x5 )
⋅ sin 2 ( x7 )[−T3 x10 − Tsz sgn( x10 )] / x9 + cos( x5 ) sin( x7 )[T2 x4
+ Tsx sgn( x4 )] / x9 − μ1 sin( x5 ) cos( x5 )[−T3 x10 − Tsz sgn( x10 )] / x9
− 2 x6 x10 / x9 ,
x 7 = x8 ,
x8 = k 2 sin( x7 ) cos( x7 )u2 /[ x92 sin 2 ( x5 )] − k3 μ1μ 2 sin 2 ( x7 ) cos( x7 )
⋅ u3 /[ x92 sin( x5 )] + μ 2 sin( x7 ) cos( x7 )[−T3 x10 − Tsz sgn( x10 )] / x9
− 2 x8 x10 / x9 + cos( x7 )[T2 x4 + Tsx sgn( x4 )] /[ x9 sin( x5 )],
x9 = x10 , , (1)
x10 = cos( x5 )[T1 x2 + Tsy sgn( x2 )] + x x sin ( x5 ) − k1 sin( x5 ) cos( x5 )
2
8 9
2
⋅ cos( x7 )u1 − k 2 sin 2 ( x5 ) sin( x7 ) cos( x7 )u 2 + k3 sin( x5 ) cos( x7 )[−μ 2

⋅ sin 2 ( x5 ) sin 2 ( x7 ) − μ1 cos 2 ( x5 ) − 1]u3 + μ 2 sin 2 ( x5 ) sin 2 ( x7 )[−T3 x10
− Tsz sgn( x10 )] + sin( x5 ) sin( x7 )[T2 x4 + Tsx sgn( x4 )] + μ1[−T3 x10 − Tsz
⋅ sgn( x10 )] + μ1 sin 2 ( x5 )[−T3 x10 − Tsz sgn( x10 )] + x62 x9 − T3 x10 − Tsz sgn( x10 ),
where the state variables are x1 – the distance of the cart from the centre of the rail,
x10 – the initial condition for x1 , x2 – the speed of the cart on the direction of x1 , x3
– the distance of the rail with the cart from the centre of the construction frame, x4 –
the speed of the rail with the cart on the direction of x3 , x5 – the acute angle be-
tween the lift-line of the payload and the rail, x6 – the angular speed that corres-
ponds to x5 , x7 – the acute angle between the lift-line of the payload and the vertic-
al line, x8 – the angular speed that corresponds to x7 , x9 – the length of the lift-line,
and x10 – the speed of the lift-line. The control signals in (1) are u1 , u2 and u3 that
correspond to the PWM duty cycles applied to the DC motors that actuate the
system on the axes x1 , x3 and x9 , respectively. The three axes x1 , x3 and x9 are
referred to as follows the x-axis, the y-axis, and the z-axis, respectively. The
parameters in (1) have the following values [25, 26] obtained from the first-principle
model of the process [27, 28]:
μ1 = 0.4156, μ 2 = 0.1431, k1 = 49.8636 , k 2 = 16.0336 , k 3 = −129 .8258,

T1 = 11.5242 , T2 = 26.3263, T3 = 217.3535, Tsx = 1.4903, Tsy = 6.4935, (2)
Tsz = 20.8333 .
The controlled output y can be one or more of the state variables x1 , x3 , x5 , x7

and x9 . The choice of this state variable depends on the specific control problems.
The state variables x1 , x3 are involved as controlled outputs in crane position
control problems, and the state variables x5 , x7 and x9 are involved as outputs in
anti-swing control problems.
The model (1) can be simplified by accepting that only the forces on the three
axes x1 , x3 and x9 affect the movement of the system. Furthermore, for zero ini-
tial conditions result the following process transfer functions H x (s) , H y (s) and
H z (s) are defined [14, 25, 26]:
H x (s) = x1 ( s) / u1 (s) = k x /[s (1 + Tx s)], H y (s ) = x3 (s ) / u2 (s) = k y /[s (1 + Ty s)],

(3)
H z (s) = x9 (s) / u3 (s) = k z /[s(1 + Tz s)],
where k x , ky and k z are the process gains, and Tx , Ty and Tz are the process time
constants. The least-squares identification based on real-world input-output data
taken from the laboratory equipment leads to the following parameter values [25,
26, 29] but other parameter values can be used [14]:
k x = 0.2939, Tx = 0.0587 s, k y = 0.2747, Ty = 0.0379 s,

(4)
k z = 0.1019, Tz = 0.0408 s.
For the sake of simplicity the three transfer functions defined in (3) are Single Input-
Single Output (SISO) processes in the controller design. Three SISO control systems
are designed, and the controller design procedures use the model simplification and
the interactions between these separately designed SISO control systems; the effects
of these interactions are important in anti-swing control problems.
The process models given in (3) can be used as benchmarks of more complex
processes. Such processes include servo systems [30–36], but other linear or non-
linear models can be used with effects on controller designs [37–48].
3 Controller Design Procedure

The direct ILC-based control system structures with CL and the ILC-based control
system structure with PCCL are presented in Fig. 1 and in Fig. 2, respectively
Lead-Lag Controller-Based Iterative
I Learning Control Algorithms 229
[25, 26], where r is the reeference input, y is the controlled output, e = r − y is thhe
control error, u is the con
ntrol signal, M is the memory block, the subscript j inddi-
cates the cycle (experimeent) index, C and C1 are the controller transfer functionns,
and P is the process transsfer function. The current control system (presented herre
as a control loop) is charracterized by the controller with the transfer function C C,
the reference input rj andd the index j + 1.
Fig. 1. Direct ILC-based con

ntrol system structure with CL.
Fig. 2. Direct ILC-based con

ntrol system structure with PCCL.
The controller with the transfer function C1 is referred to also as the learninng
rule. The disturbance inpu uts are not included in Fig. 1 and in Fig. 2 because theeir
repetitiveness cannot be guaranteed.
g Moreover, as shown in Section 1, the repetti-
tiveness is treated in diffeerent manners for direct and indirect ILC.
If the control error at thhe first cycle is finite and e0 ≠ 0 , the convergence conddi-
tions for the learning proccesses in these two ILC-based control system structurees
are expressed in the follow wing unified manner:
|| ei+1 / ei || ∞ ≤ γ ∀i ∈ N , 0 < γ < 1 , (55)
where the parameter γ determines

d the convergence speed, and the well accepteed
notation and frequency do
omain calculation for the H∞ norm is employed:
G ∞
= sup | G ( jω ) | , (6)
ω∈Ω
where j 2 = −1 , and Ω = [ωa , ωb ], 0 ≤ ωa < ωb – the frequency range of interest

that contains the frequencies ω of the controllers.
The convergence of learning processes in both CL and PCCL is guaranteed be-
cause the condition (5) leads to
ei ∞
≤ γ i e0 ∞
→ 0 as i → ∞ . (7)
Using the control system structures presented in Figs. 1 and 2, the convergence
condition (5) can be transformed into controller design conditions. The design
condition for the ILC-based control system structure with CL is
|| 1 − P C C1 /(1 + P C ) || ∞ ≤ γ < 1 , (8)
and the condition for the ILC-based control system structure with PCCL is
|| (1 − P C1 ) /(1 + P C ) || ∞ ≤ γ < 1 . (9)
The following unified controller design procedure is suggested:

• Step I. Carry out a frequency domain design to tune the parameters of the con-
troller with the transfer function C in the current cycle. The values of phase
margin are imposed to ensure not only stable control systems in the current
cycle that guarantees a finite e0 and also acceptable performance indices of
these control loops.
• Step II. Set the value of the parameter γ and carry out another frequency do-
main design that to tune the parameters of the controller with the transfer func-
tion C1 in the previous cycle. The parameters of C1 are tuned such that to fulfill
the controller design condition (8) for the ILC-based control system structure
with CL, and the controller design condition (9) for the ILC-based control sys-
tem structure with PCCL.
4 Experimental Results
The controller design procedure is tested by experiments on a 3D crane system

laboratory equipment [27] built around the experimental setup implemented in the
Intelligent Control Systems Laboratory of the “Politehnica” University of Timi-
soara, Romania (Fig. 3). The experimental setup consists of a rail moving along
the frame, a cart moving on the rail, and a payload being shifted up and down.
PWM-based actuated DC motors are implemented for the three axes.
Fig. 3. Photography of experimental setup.
The two ILC-based control system structures control separately the x-axis,
the y-axis and the angle, i.e., they control x1 , x 3 and x5 , respectively, using the
process transfer functions P( s) = H x ( s) , P(s) = H y ( s) for the x-axis and for the
y-axis, respectively. Two transfer functions can be defined for the anti-swing con-
trol problem in terms of considering either the control signals u1 or u2 , and one of
these functions is defined as [29]
P( s) = x5 ( s) / u1 ( s) = K z /(s 2 + az ) , (10)
which points out the need for oscillation damping, K z = 2.5 and a z = 24.53 .
The steps I and II of the controller design procedure use the lead-lag controllers
and the real PD learning rules with the unified transfer functions C (s ) and C1 ( s ) ,
respectively [25, 26]:
C ( s) = k c (1 + Ta s) /(1 + Tb s ), C1 ( s) = k1 (1 + Tc s ) /(1 + Td s) , (11)
where k c and k 1 are gains, and Ta , Tb , Tc and Td are time constants, Ta > Tb ,
Tc > Td .
The two steps of the controller design procedure presented in the previous
Section 3 were applied to obtain the same parameter values in the control system
structures for the position control problems (for the x-axis and for the y-axis). The
frequency domain approaches applied in the two steps of the design approach
resulted in the following parameter values for the x-axis and for the y-axis:
k c = 0.1, k1 = 1.5, Ta = 1 s, Tb = Td = 0.01 s, Tc = 0.07 s , (12)

and the parameter values for the anti-swing control problem are:
kc = k1 = 0.01, Ta = Tc = 0.1 s, Tb = Td = 0.01 s . (13)
The experimental results are presented as follows considering that the reference
inputs are rj in Figs. 1 and 2. The results are expressed as the responses of x1 and
x 3 as controlled outputs (the variables y j +1 in Figs. 1 and 2) after one iteration
Fig. 4. System’s responses obtained by experiments after 1 and after 52 iterations for the
ILC-based control system structure with CL in the position control problem for the x-axis.
ILC-based control system structure with PCCL in the position control problem for the
x-axis.
and after several iterations. The system responses of ILC-based control system
structure with CL in the position control problem for the x-axis are presented in
Fig. 4, and the system responses of the ILC-based control system structure with
PCCL in the position control problem for the x-axis are presented in Fig. 5.
The results presented in Figs. 4 and 5 show the performance improvement en-
sured by ILC reflected by reduced overshoots and settling times. The results also
show the faster convergence of the ILC-based control system structure with PCCL
in comparison with the control system with the ILC-based control system structure
with CL. In addition, the ILC-based control system structure with PCCL exhibits
less oscillatory behavior of the control signal compared to the ILC-based control
system structure with CL. That is the reason why the ILC-based control system
structure with PCCL is applied to the anti-swing control problem.
The experimental results in the anti-swing control problem for one of the angles
(with zero reference input) and the process transfer function defined in (10) are
presented in Fig. 6. These system responses correspond to the ILC-based control
ILC-based control system structure with PCCL in the anti-swing control problem.
system structure with PCCL. Fig. 6 shows that the results require additional per-
formance improvements in terms of considering different control system struc-
tures and of different controller tuning methods. The Extended Symmetrical
Optimum method [49] applied to PI and PID controller tuning can represent a
viable alternative in this context.
5 Conclusion
This chapter has offered two ILC-based control system structures that involve CL
and PCCL. The advantage of these control system structures is the performance
improvement in the context of PD controllers and learning rules applied to the
position control problem of a 3D crane system laboratory equipment.
The limitations of our control system structures concern the need for distur-
bance rejection and for performance improvement in the anti-swing control prob-
lem. Future research will aim the performance improvement in the anti-swing
control problem using different controller structures in the framework of discrete-
time formulations of ILC.
Acknowledgments. This work was supported by a grant of the Romanian National Author-
ity for Scientific Research, CNCS – UEFISCDI, project number PN-II-ID-PCE-2011-3-
0109, and by the cooperation between the Óbuda University, Budapest, Hungary, and the
“Politehnica” University of Timisoara, Romania, in the framework of the Hungarian-
Romanian Intergovernmental Scientific & Technological Cooperation Program.
References
[1] Toxqui, R., Yu, W.: Anti-swing control for overhead crane with neural compensation.
In: Proceedings of 2006 International Joint Conference on Neural Networks (IJCNN
2006), Vancouver, BC, Canada, pp. 9447–9453 (2006)
[2] Yu, W., Li, X., Irwin, G.W.: Stable anti-swing control for an overhead crane with
velocity estimation and fuzzy compensation. In: Lowen, R., Verschoren, A. (eds.)
Foundations of Generic Optimization, Applications of Fuzzy Control, Genetic Algo-
rithms and Neural Networks, vol. 2, pp. 223–240. Springer, Heidelberg (2008)
[3] Yu, W., Li, X.: Anti-swing control for an overhead crane with intelligent compensa-
tion. In: Proceedings of 3rd International Symposium on Resilient Control Systems
(ISRCS 2010), Idaho Falls, ID, USA, pp. 85–90 (2010)
[4] Yoshida, Y., Tabata, H.: Visual feedback control of an overhead crane and its combi-
nation with time-optimal control. In: Proceedings of 2008 IEEE/ASME International
Conference on Advanced Intelligent Mechatronics (AIM 2008), Xi’an, China,
pp. 1114–1119 (2008)
[5] Westerberg, S., Manchester, I.R., Mettin, U., La Hera, P., Shiriaev, A.: Virtual envi-
ronment teleoperation of a hydraulic forestry crane. In: Proceedings of 2008 IEEE
International Conference on Robotics and Automation (ICRA 2008), Pasadena, CA,
USA, pp. 4049–4054 (2008)
[6] Chang, C.Y., Chiang, K.H.: The nonlinear 3-D crane control with an intelligent
operating method. In: Proceedings of 2008 SICE Annual Conference, Tokyo, Japan,
pp. 2917–2921 (2008)
[7] Chwa, D.: Nonlinear tracking control of 3-D overhead cranes against the initial swing
angle and the variation of payload weight. IEEE Trans Contr. Syst. Technol. 17,
876–883 (2009)
[8] Ahmad, M.A., Ismail, R.M.T.R., Ramli, M.S.: Input shaping techniques for anti-sway
control of a 3-D gantry crane system. In: Proceedings of 2009 International
Conference on Mechatronics and Automation (ICMA 2009), Changchun, China,
pp. 2876–2881 (2009)
[9] Ahmad, M.A., Ismail, R.M.T.R., Ramli, M.S., Abd Ghani, N.M., Hambali, N.:
Investigations of feed-forward techniques for anti-sway control of 3-D gantry crane
system. In: Proceedings of 2009 IEEE Symposium on Industrial Electronics & Appli-
cations (ISIEA 2009), Kuala Lumpur, Malaysia, vol. 1, pp. 265–270 (2009)
[10] Kaneshige, A., Miyoshi, T., Terashima, K.: The development of an autonomous mo-
bile overhead crane system for the liquid tank transfer. In: Proceedings of 2009
IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM
2009), Singapore, pp. 630–635 (2009)
[11] Pisano, A., Scodina, S., Usai, E.: Load swing suppression in the 3-dimensional over-
head crane via second-order sliding-modes. In: Proceedings of 11th International
Workshop on Variable Structure Systems (VSS 2010), Mexico City, Mexico, pp.
452–457 (2010)
[12] Cuenca, Á., Salt, J., Sala, A., Piza, R.: A delay-dependent dual-rate PID controller
over an Ethernet network. IEEE Trans. Ind. Informat. 7, 18–29 (2011)
[13] Yu, W., Moreno-Armendariz, M.A., Ortiz Rodriguez, F.: Stable adaptive compensa-
tion with fuzzy CMAC for an overhead crane. Inf. Sci. 181, 4895–4907 (2011)
[14] Jovanović, Z., Antić, D., Stajić, Z., Milošević, M., Nikolić, S., Perić, S.: Genetic al-
gorithms applied in parameters determination of the 3D crane model. Facta Universi-
tatis, Series: Automatic Control and Robotics 10, 19–27 (2011)
[15] Bristow, D.A., Tharayil, M., Alleyne, A.G.: A survey of iterative learning control.
IEEE Control Syst. Mag. 26, 96–114 (2006)
[16] Ahn, H.S., Moore, K.L., Chen, Y.: Iterative learning control. Robustness and mono-
tonic convergence for interval systems. Springer, Heidelberg (2007)
[17] Xu, J.X., Panda, S.K., Lee, T.H.: Real-time iterative learning control. Design and ap-
plications. Springer, Heidelberg (2009)
[18] Owens, D.H., Hätönen, J.: Iterative learning control - An optimization paradigm. An-
nu. Rev. Control 29, 57–70 (2005)
[19] Abidi, K., Xu, J.X.: Iterative learning control for sampled-data systems: From theory
to practice. IEEE Trans. Ind. Electron. 58, 3002–3015 (2011)
[20] Wang, Y., Gao, F., Doyle, F.: Survey on iterative learning control, repetitive control,
and run-to-run control. J. Process Contr. 19, 1589–1600 (2009)
[21] Ruan, X., Bien, Z., Park, K.H.: Decentralized iterative learning control to large-scale
industrial processes for nonrepetitive trajectory tracking. IEEE Trans. Syst. Man Cy-
bern. A Syst. Humans 38, 238–252 (2008)
[22] Liu, T., Gao, F.: Robust two-dimensional iterative learning control for batch
processes with state delay and time-varying uncertainties. Chem. Eng. Sci. 66,
6134–6144 (2010)
[23] Tan, K., Zhao, S., Xu, J.X.: Online automatic tuning of a proportional integral deriva-
tive controller based on an iterative learning control approach. IET Contr. Theory
Appl. 1, 90–96 (2007)
[24] Wu, J., Ding, H.: Reference adjustment for a high-acceleration and high-precision
platform via A-type of iterative learning control. Proc. IMechE I J. Syst. Control
Eng. 221, 781–789 (2007)
[25] Precup, R.E., Enache, F.C., Rădac, M.B., Petriu, E.M., Dragoş, C.A., Preitl, S.:
Iterative learning control application to a 3D crane system. In: Proceedings of 8th
International Conference on Informatics in Control, Automation and Robotics
(ICINCO 2011), Noordwijkerhout, The Netherlands, vol. 1, pp. 117–122 (2011)
[26] Rădac, M.B., Enache, F.C., Precup, R.E., Petriu, E.M., Preitl, S., Dragoş, C.A.: Pre-
vious and current cycle learning approach to a 3D crane system laboratory equipment.
In: Proceedings of 15th International Conference on Intelligent Engineering Systems
(INES 2011), Poprad, Slovakia, pp. 197–202 (2011)
[27] Inteco Ltd., 3D crane, user’s manual. Inteco Ltd., Krakow, Poland (2008)
[28] Chen, W., Wu, Q., Tafazzoli, E., Saif, M.: Actuator fault diagnosis using high-order
sliding mode differentiator (HOSMD) and its application to a laboratory 3D crane. In:
Proceedings of 17th World Congress of the International Federation of Automatic
Control, Seoul, Korea, pp. 4809–4814 (2008)
[29] Enache, F.C.: Iterative learning control-based control solutions. Applications to a 3D
crane laboratory equipment. B.Sc. Thesis, Department of Automation and Applied In-
formatics, “Politehnica” University of Timisoara, Timisoara, Romania (2010)
[30] Paláncz, B., Benyó, Z., Kovács, L.: Control system professional suite. IEEE Control
Syst. Mag. 25, 67–75 (2005)
[31] Harmati, I., Lantos, B., Payandeh, S.: Fitted stratified manipulation with decomposed
path planning on submanifolds. Int. J. Robot. Autom. 20, 135–144 (2005)
[32] Vaščák, J.: Navigation of mobile robots using potential fields and computational in-
telligence means. Acta Polytechnica Hungarica 4, 63–74 (2007)
[33] Blažič, S., Matko, D., Škrjanc, I.: Adaptive law with a new leakage term. IET Control
Theory Appl. 4, 1533–1542 (2010)
[34] Vaščák, J., Madarász, L.: Adaptation of fuzzy cognitive maps - a comparison study.
Acta Polytechnica Hungarica 7, 109–122 (2010)
[35] Garcia, A., Luviano-Juarez, A., Chairez, I., Poznyak, A., Poznyak, T.: Projectional
dynamic neural network identifier for chaotic systems: Application to Chua’s circuit.
Int. J. Artif. Intell. 6, 1–18 (2011)
[36] Linda, O., Manic, M.: Uncertainty-robust design of interval type-2 fuzzy logic con-
troller for delta parallel robot. IEEE Trans. Ind. Informat. 7, 661–670 (2011)
[37] Baranyi, P., Kóczy, L.T.: A general and specialised solid cutting method for fuzzy
rule interpolation. J. Busefal 66, 13–22
[38] Baranyi, P., Yam, Y., Várkonyi-Kóczy, A., Patton, R.J.: SVD based reduction to
MISO TS fuzzy models. IEEE Trans. Ind. Electron. 50, 232–242 (2003)
[39] Horváth, L., Rudas, I.J.: Modelling and solving methods for engineers. Elsevier, Aca-
demic Press, Burlington (2004)
[40] Škrjanc, I., Blažič, S., Agamennoni, O.E.: Identification of dynamical systems with a
robust interval fuzzy model. Automatica 41, 327–332 (2005)
[41] Johanyák, Z.C., Kovács, S.: Fuzzy rule interpolation based on polar cuts. In: Reusch,
B. (ed.) Computational Intelligence, Theory and Applications, pp. 499–511. Springer,
Heidelberg (2006)
[42] Wilamowski, B.M., Cotton, N.J., Kaynak, O., Dundar, G.: Computing gradient vector
and Jacobian matrix in arbitrarily connected neural networks. IEEE Trans. Ind.
Electron. 55, 3784–3790 (2008)
[43] Iglesias, J.A., Angelov, P., Ledezma, A., Sanchis, A.: Evolving classification of
agents’ behaviors: a general approach. Evolving Syst. 1, 161–171 (2010)
[44] Johanyák, Z.C.: Student evaluation based on fuzzy rule interpolation. Int. J. Artif. In-
tell. 5, 37–55 (2010)
[45] Rajabioun, R.: Cuckoo optimization algorithm. Appl. Soft. Comp. 11, 5508–5518
(2011)
[46] Alfi, A., Fateh, M.M.: Intelligent identification and control using improved fuzzy par-
ticle swarm optimization. Expert Syst. Appl. 38, 12312–12317 (2011)
[47] Kasabov, N., Hamed, H.N.A.: Quantum-inspired particle swarm optimisation for in-
tegrated feature and parameter optimisation of evolving spiking neural networks. Int.
J. Artif. Intell. 7, 114–124 (2011)
[48] Madarász, L., Živčák, J.: Application of medical thermography in the diagnostics of
carpal tunnel syndrome. In: Proceedings of IEEE 12th International Symposium on
Computational Intelligence and Informatics (CINTI 2011), Budapest, Hungary,
pp. 535–539 (2011)
[49] Preitl, S., Precup, R.E.: An extension of tuning relations after symmetrical optimum
method for PI and PID controllers. Automatica 35, 1731–1736 (1999)
Algorithms for Pitch Distance Determination
Gyula Hermann
Óbuda University, H-1034 Budapest Bécsi út 94B, Hungary
Abstract. The paper describes algorithms used for the determination of pitch dis-
tances in the calibration of optical scales on a calibration machine. Two groups of
algorithms are used: the non-contextual and the contextual ones. The first group is
used to calculate the position of the edges in the digital image captured by the
camera is either based on the gradient or the so called Laplacian method without
any prior knowledge of the shape of the object. The contextual group of algorithm
make use of the a priori knowledge of the edge or shape and its position. The in-
formation carried by gray level values of the pixelsenables us to reach subpixel
resolution. The algorithms in the second group calculate the centreline of the
graduation or measuring line either using Gabor transformation or from the centre
of gravity formed by the gray values of the individual pixels. The resistance of the
algorithms to various distortions on the scale was investigated. Finally the whole
calibration system was described, uncertainty components are enumerated and ex-
perimental results are given.
1 Introduction
The micro- and nanotechnologies increases the demand for optical calibration
standards with submicron structures. Scales with periodic structures in one or two
direction are cheap and easy to use in the calibration of microscopes and other
measurement devices. Therefore the pitch distance, the mean distance between the
lines on such scales, has to be known precisely.
Fig. 1. Various types of scales

40 G. Hermann
Many kind of interferometric calibration systems for line scales have been built
at many laboratories. Some laboratories modified length measuring machines for
this purpose. Currently static line detection systems, mostly used in calibration la-
boratories, are to be replaced by systems using dynamic calibration technique,
wher the pitch distance is determined on the fly.
One of the earliest paper [2] describes the NIST length scale interferometer for
measuring graduated length scales. It discusses in detail not only the machine and
its operation, but elaborates also on the uncertainty, it’s sources, the required envi-
ronmental conditions and how they are kept under control or compensated.
Meli’s paper [11][12] presents another approach, based on laser diffractrometry,
according to Littrow principle, resulting in picometer measuring uncertainty. The
described method is suitable for 2D planar scales as well. In contrast to calibra-
tions by either conventional optical technology or SPM featuring interferometric
position detection, no local information about the grating standards is obtained.
The found pitch value is an average over the measured sample spot which has
usually a diameter form one up to a few millimeters.
An other approach is the so called flux area measurement used especially in the
determination of line width in the submicron range [5]
The paper by Druzuvec and his co-authors [4] discusses the effect of contami-
nation in the calibration uncertainty model. Significant influence in calibration un-
certainty budget is represented by the uncertainty of the line centre detection. The
paper discusses different types of line scale contamination like dirt spot, scratches,
line edge incorrectness and line intensity variations. They were simulated in order
to test the error correction capability of the line centre detection algorithm.
A good overview is given in [13] of the various calibration methods and measur-
ing equipments used by the major national metrology laboratories.
A major problem in the calibration process is the determination of the individu-
al pitch distances. A number of edge detection techniques are known from digital
image processing. The earlier schemes such as Roberts and Sobel operators are
gradient based [7]. An other approach is based on surface fitting [8]. The most
popular approaches, are based on convolving the images with Gaussian-like ker-
nels Canny derived Gaussian-like filters that maximize the signal to noise ratio
and minimize the localization error [3]. In a recent paper Galvan and Munos-
Gomes applied Gabor filter for the determination of the centre line points and sub-
sequently fitted a line to these points [6]. However their method is vulnerable to
line deterioration and. scale contamination. In the last years number of papers has
been published on fuzzy edge detection [1][10]. Unfortunately most of these algo-
rithms are not directly applicable to the problem because they change the informa-
tion content of the picture and require considerable computational effort.
2 Requirements
Graduated scales are made of various materials including steel, Invar, glass, glass-
ceramics, silicon and fused silica. The cross sectional shape can be rectangular,
H or U-form, or modified X called Tresca. The machine should be able to accom-
modate and calibrate all of them up to a length of 400 mm, with a minimal pitch
Algorithms for Pitch Distance Determination 41
distance of 1µm. The resolution of the measuring system has to be better 0,02µm.
The image of the lines should be captured optically. The structure has to be
equipped with the appropriate number of high precision temperature sensors to
collect enough data for determining the thermal distribution and compensating the
effect of the resulting deformation.
The construction of the linear scale calibration system consists of three physi-
cally separated parts: the solid steel bed of the Zeiss measuring machine and the
carriage module, the optical microscope and image capturing module and the
measuring system. The motion and the image evaluation is executed and coordi-
nated by a personal computer. The carriage’s position is determined by an HP het-
erodyne laser interferometer.
CCD camera
Vertical & tilt

Illumination Linear stage adjustment
Cross slide
Solid steel Laser inter-

base ferometer
Vibration isola-
tion pad
Fig. 2. The overall construction of the calibration system
Linear scales are calibrated by moving the carriage and measuring it’s dis-
placement by the interferometers. The measuring value is calculated as the sum of
the laser interferometer reading and the indication of the digital microscope. From
here out follows that an important factor in the calibration process is the determi-
nation of the centreline.
3 The Various Definitions of Pitch Distance

The pitch distance can be defined on three ways given in the subsequent picture.
Fig. 3. Pitch distance definitions

42 G. Hermann
Whereas the first two versions given in the picture can be measured directly, the
third one requires more processing. The difficulty is generated by line thickness
variation and the straightness error of the edges.
When a bright line on a black background is viewed on sample-level, column
by column, theoretically the pattern obtained is an ideal step-shaped waveform. In
real-life, this waveform is not that ideal, but the line will be a bit more blurred. On
sample-level, the blurring will result into a Gaussian curve. This will make the de-
tection of the edges even more complex.
4 Edge Detection Methods
From the point of conceptual view, the edge detection methods are categorized in-
to contextual and non-contextual approaches. The non-contextual methods work
autonomously without any priori knowledge about the scene and the edges. They
are flexible in the sense that they are not limited to specific images. However, they
are based on local processing focused on the area of neighbouring pixels. The con-
textual methods are guided by a priori knowledge about the edges or the scene.
They perform accurately only in a precise context. It is clear that autonomous de-
tectors are appropriate for general-purpose applications. However, contextual de-
tectors are adapted to specific applications that always include images with same
scenes or objects.
Sub-pixel rendering is an image processing technique to increase the apparent
resolution of an image. It takes advantage of the fact that each pixel is actually com-
posed of individual sub-pixels with greater detail. Sub-pixel approach is used to im-
prove the accuracy of localization in edge detection. As it was mentioned, edges can
be localized using either the local maxima or zero-crossing methods, however, by
applying interpolation on the pixels we can localize them more accurate.
The majority of the different edge detection methods can be grouped into two
categories:
First Order Derivative Based Edge Detection(Gradient method): It detects the
edges by looking for the maximum and minimum in the first derivative of the im-
age. Sharpening an image results in the detection of fine details as well as enhanc-
ing blurred ones. The magnitude of the gradient is the most powerful technique
that forms the basis for various approaches to sharpening. The gradient vector
points in the direction of maximum rate of change. For a function fix, y), the mag-
nitude of the gradient of f at coordinates (x, y) is defined as:
∇f (x, y) = ( ∂ x f ( x , y )) 2 + ( ∂ y f ( x , y )) 2
while the gradient orientation is given by:
 ∂ y f ( x, y) 
∇ f ( x , y ) = arctan  
 ∂ x f ( x , y) 
Second Order Derivative Based Edge Detection (Laplacian based Edge Detec-
tion): The Laplacian method searches for zero crossings in the second derivative
of the image to find edges. An edge has the one-dimensional shape of a ramp and
calculating the derivative of the image can highlight its location. Suppose we have
the following signal, with an edge shown by the jump in intensity below:
Fig. 4. Intensity variation in the neighbourhood of an edge
If we take the gradient of this signal (which, in one dimension, is just the first
derivative with respect to t) we get the following:
Fig. 5. First derivative of the intensity function
Clearly, the derivative shows a maximum located at the centre of the edge in the
original signal. This method of locating an edge, is characteristic of the “gradient
filter” family of edge detection filters and includes the Sobel method.
A pixel location is declared an edge location if the value of the gradient ex-
ceeds some threshold. As mentioned before, edges will have higher pixel intensity
values than those surrounding it. So once a threshold is set, one can compare the
gradient value to the threshold value and detect an edge whenever the threshold is
exceeded. Furthermore, when the first derivative is at a maximum, the second de-
rivative is zero. As a result, another alternative to finding the location of an edge is
to locate the zeros in the second derivative. This method is known as the Lapla-
cian and the second derivative of the signal is shown below:
44 G. Hermann
Fig. 6. Second derivative of the intensity function
In the edge based approaches each horizontal line profile within the region of
interest in the image is analysed. The centre of the left and the right edge is used
and the edge locations are determined with a moment based edge operator. A line
is fitted through all these centres using only points within 2σ. The intersection of
this fitted line with the reference line is used as the scale line position. The line fit-
ting can done either using least square fit or a min-max fit based on a convex hull
algorithm [9].
The second approach is based on the use of complex Gabor filter to obtain sub-
pixel resolution:
h( x) = g ( x)[cos(ϖ 0 x ) + j sin(ϖ 0 x) ]
where g(x) is a Gaussian function.
Fig. 7. Detecting the centre of a line using Gabor filter
From the above figure it can be seen that the zero crossing of phase signal is the
appropriate feature to find the line centre. Based on the detected lines centres, it is
necessary to fit a straight line. In order to eliminate the effected of the outlier data
points an M-estimator is used.
An other way is, to calculate the centre of “gravity” of the gray values of the
individual pixels:
 g r
i, j i, j
rcentre =
i j
 gi j
i, j
where r is the distance from the origin of the coordinate system and gi,j-s are the
gray scales values of the different pixels.
5 The Influence of Contamination and Its Elimination

The centreline detection algorithm was tested with different deteriorated scales,
some of them are gives in the figure below.
Fig. 8. Lines with distorsion and contaminations
First the centreline of the original image was calculated and after that the line
was “repaired” either by filling the white spots or cutting of the bulges and repeat-
ing the calculation again. It was found that the displacement of the centre line due
to these distortions was in our cases less, then one pixel resulting in centreline
shift less then 20 nm.
6 The Physical Realization of the Optical System
Linear scales are calibrated by moving the carriage and measuring it’s displace-
ment by the interferometers. The result of the measurement corresponds exactly to
the scale pitch. The determining factor to this correspondence is the line centring
process. The scale structures are captured optically. The optical system is assem-
bled together from various commercially available modules. The microscope con-
sists of an objective, a control element for focusing, based on a piezo actuator a
tube lens, and a camera connection module.
The microscope with the CMOS camera is mounted on a beam fixed to the steel
base. Digital measuring microscope enables precise estimation of the line edge
quality and precise location of lines. To process the image data captured to differ-
ent approach had been followed in the subsequent paragraphs.
46 G. Hermann
Fig. 9. The optical system on a test bench
7 The Illumination
As it was already mentioned from the optical point of view linear scales can be
transparent (made of glass, silica or Zerodur) and opaque (made of various met-
als). Therefore the machine has to be equipped with both through the lens and
backlight illumination.
For opaque (metal) scales diffuse axial illumination is envisaged. The lens
looks through a beamsplitter that is reflecting light on the scale. Illumination is
coaxial to the imaging system.
For transparent scales backlight illumination is applied. Backlighting is an ex-
cellent way to improve applications that need to measure edges, because silhouet-
ting the object provides high contrast for improved edge detection while reducing
unneeded surface detail. Normal backlights, however, can reduce the sharpness of
the edge, because diffuse reflections occur from the broad area output and thus re-
duce the accuracy of gauging and inspection systems. Telecentric backlight
illumination, is based on the same principles as telecentric measuring lenses, uses
parallel rays, it avoids these problems. It also increases illumination compared to a
standard backlight, allowing faster measurements.
Fig. 10. The illumination system during assembly

8 Experimental Results
The algorithms presented in section 4 have been implemented on personal com-
puter built in the camera. It is responsible for the edge and centre line determina-
tion and communicates its results to another personal computer collecting data
from the laser interferometer and controlling the motion of the stage with the aid
of microcontrollers.
Fig. 11. The calibration machine in operation
Fig. 12. The user interface

48 G. Hermann
The subsequent picture illustrates the operator interface on the personal com-
puter controlling the global system. It enables user to define the measuring points
either manually (aperiodic measurement) or automatically with a given period.
The measuring mode (left or right edge, centre) can be selected by a radio button.
After successful execution of the calibration process the calibration documenta-
tion, including the uncertainty calculation, is automatically generated and printed.
The sources of measurement uncertainty can be divided into five groups:
Table 1. Uncertainty components
Translation stage Imaging system

Yaw of the stage x motion Magnification
Pitch of the stage x motion Camera orientation
Abbe offset in the x and z directions Image distorsion
Interferometer Half line width
Laser wavelength Resolution of edge detection
Air temperature Scale properties
Air pressure Scale misaligment
Air humidity Temperature deviation
Air CO2 concentration Thermal expansion coefficient
Edlen equation Scale support
Interferometer dead path Line quality and evaluation
Interferometer alignment Illumination inhomogenity
Interferometer resolution Measurement
Drift influence Repeatibility
We have calibrated the complete system and the optical system separately. It
was found that the uncertainty of the complete system is about 60 nm while the
edge and the centreline detection was less then 20 nm.
9 Conclusion
The paper discusses the various edge determination algorithms, presenting their
advantages and drawbacks. It describes the high resolution optical microscope re-
sponsible for capturing line data. The line data is processed by filtering technique
in order to determine the line centers. This enables dynamic calibration of grat-
ings, hereby reducing calibration time and costs.
Acknowledgments. The author thanks Dr. Gy. Ákos, dr. Miklós Báthor and Zsolt Bánki
for their valuable contribution and István Nagy for implementing the illumination system.
References
[1] Barkhoda, W., Tab, F.A., Shahryari, O.-K.: Fuzzy Edge Detection Based on Pixel’s
Gradient and Standard Deviation Values. In: Proc. of the International Multiconfe-
rence on Computer Science and Information Technology, pp. 7–10
[2] Beers, J.S., Penzes, W.B.: Y 14.5-1973 The NIST Length Scale Interferometer. Jour-
nal of Research of the National Institute of Standards and Technology 104, 225–252
(1999)
[3] Canny, J.: A Computational Approach to Edge Detection. IEEE Transactions on Pat-
tern Analysis and Machine Intelligence 8(6), 679–698 (1986)
[4] Druzovec, B., Acko, B., Godina, A., Welzer, T.: Simulation of line scale contamina-
tion in calibration uncertainty model. International Journal of Simulation Model 73,
113–123 (2008)
[5] Fiekowsky, P.: The end of threshold: Subwavelength Optical Linewidth Measurement
Using the Flux-Area Technique. Photomask Japan poster 4066-67 (2000)
[6] Galván, C., Munos-Gomez, J.A.: Robust line detection for line scale calibration
[7] Gonzales, R.C., Wintz, P.: Digital Image Processing. Addison-Wesley, Menlo Park
(1987)
[8] Haralick, R.M.: Digital edge steps from zero crossing of second directional deriva-
tives. IEEE Transactions on Pattern Analysis and Machine Intelligence 6, 58–68
(1984)
[9] Hermann, G.: Robust convex-hull based algorithm for straightness and flatness de-
termination. Acta Polytechnica Hungarica 4(4), 111–120 (2007)
[10] Looney, C.G.: Fuzzy and rule-based image convolution. Mathematics and Computers
in Simulation 51, 209–219 (2000)
[11] Korotkin, V.I., Pulkin, S.A., Vitushkin, A.L., Vitushkin, L.F.: Laser interferometric
diffractometry for the measurements of diffraction grating spacing. Applied Op-
tics 35(24), 4782–4786 (1996)
[12] Meli, F., Thalman, R., Blattner, P.: High precision pitch calibration of gratings using
diffraction interferometry. In: Proc. of the 1st. Conf. on Precision Engineering and
Nanotechnology, Bremen, vol. 2, pp. 252–255 (1999)
[13] WGDM-7 Preliminary comparison on nano-metrology according to the rules of CCL
key comparisons. Nano 3 Line scale standards. Final report. PTB Braun-schweig
(2003)
Objects for Visualization of Process Data
in Supervisory Control
Iveta Zolotová, Roman Mihaľ, and Rastislav Hošák
Technical University of Košice, Department of Cybernetics and Artificial Intelligence,

Košice, Slovakia
{iveta.zolotova,roman.mihal,rastislav.hosak}@tuke.sk
Abstract. Supervisory control system represents a central unifying part of

information and control system of industrial plant processes. Particularly at
present, supervisory control applications even allocates a big share of factory
software investments. The fundamental goal of projecting supervisory application
is to establish stable and fast communication with technological process level to
obtain data from processes and to design visualization screens in a suitable form to
ensure readily visible information and control functions of processes in a plant.
That is realized by visualization screens developed using graphical objects from
different fields of industries which gives system builders much more effective
tools for designing of visualization application and better visibility of process
elements on screen. It is very closely related to trend objects used in visualization
as monitoring components. This article presents concept of trend objects used in
visualization screens of supervisory control application using software products
from Wonderware Corp.
Keywords: Trend object, supervisory control, visualization, graphical object.
1 Introduction
In industrial sphere the issue of supervisory control is also known by the acronym
SCADA/HMI (Supervisory Control And Data Acquisition / Human – Machine
Interface) and presents central unifying position in the factory control system. The
heart of supervisory control application is visualization of processes in
manufactory with all the elements represented by graphical objects in visualization
screens like trends, alarms, buttons, text arrays and many others.
In the past, the most emphasis in the development of visualization applications was
focused on sufficient interpretation of information by means of text objects or basic
static graphics due to the lack of libraries with graphical symbols and objects in
environments of software tools intended to create applications of supervisory control.
Certainly it was also caused by low performance and possibilities of computational
resources in the former period. With the growing computer capabilities and
development of information technologies the conveniences of modern graphical
symbols in software for development of supervisory control applications has appeared.

52 I. Zolotová, R. Mihaľ, and R. Hošák
At present a great emphasis in the issue of software tools for visualization of

processes is focused on reducing application development time. From this and
other reasons modern supervisory control software contains directly the entire
libraries of graphic symbols reserved for particular industries and it’s technology
parts to gain more specific, graphically appropriate and technologically more
functional graphical objects implemented in process visualization screens.
One of the most important component of general realization of visualization
application in conditions of real-time process control is trend object. It is usually
represented by separated graphical symbol with options that involve possibilities to
show progress of various process variables obtained from technological process
level.
These mentioned facilities also contains the connection of supervisory control
software InTouch and Application Server from Wonderware Corp., which by means
of ArchestrA technology adds the possibility to develop visualization application
with ArchestrA graphical symbols library. These graphical objects have a modern
graphical appearance and offer different edit possibilities in development
environment of Application Server, called ArchestrA IDE (Integrated Development
Environment).
2 Supervisory Control
Supervisory control [1], [2], [3] represents a system, that provides monitoring and
control of geographically large technological processes. Within this system there
are included all the equipment and functionality for acquisition, processing,
transmission and visualization of necessary information about controlled
processes.
Supervisory control system in an industrial plant operates these primary
functionalities:
• data acquisition from technological level of controlled system,
• visualization – alarms, trends, data and process flow presentation,
• technological process control,
• communication with machinery level and other subsystems at a higher
control level,
• data archiving – cooperation with database system.
The most pronounced component of SCADA/HMI application is visualization.
Visualization of processes [3], [4] means application of theoretical, technical,
software and communication issues within industrial plant to visualize defined
objects concerning technological processes and their automatic control system
with the aim to support real-time decision making and control.
Current visualization software tools offer full graphical user interface for
development and deployment of applications. In many cases this interface has the
object oriented representation possibility of system elements or it contains
preformatted libraries with graphical objects for different parts and sectors of
industrial equipment [5].
Objects for Visualization of Process Data in Supervisory Control 53
3 Graphical Objects in Visualization

In this section of article graphical objects used in process visualization are
described in ArchestrA IDE (Integrated Development Environment) as an
environment of industrial software products from Wonderware Corp.
The best known product is InTouch that provides development and running of
complete visualization applications with predefined graphic tools. A superstructure
above InTouch makes product Wonderware Application Server, which introduces
technology ArchestrA presented as unifying software architecture with products and
services for effective design and deployment of automation and information
application in industrial sphere [6].
ArchestrA technology interconnect functionality of InTouch and Application
Server into development environment ArchestrA IDE where is used object-based
representation for conveying process variables and complex of elements with
predefined options.
InTouch offers graphics using custom graphics designs or using graphics libraries
with predefined graphical symbols libraries from a tool Wizards Selection. After
InTouch integration with Application Server there is an opportunity to use libraries of
graphical ArchestrA symbols.
The integration concept of visualization software tool InTouch and Application
Server enriches the overall supervisory application of modern graphical objects from
the ArchestrA symbols library. In WindowMaker, the visualization development
environment of InTouch, it is possible to use InTouch graphic symbols and also
ArchestrA graphics after installation and functional combining between both products,
InTouch and Application Server.
ArchestrA IDE has an graphical user interface with object representation and also
contains other various objects with predefined functionality and miscellaneous
possibilities to create and deploy complex applications.
These modern graphical ArchestrA objects have an option to advanced editation,
modifiability to the needs of application and particularly preformatted properties.
ArchestrA graphical objects are accessible for developer of application from
the graphical library situated in integrated development environment called
ArchestrA IDE in the section Graphic Toolbox. Here are graphical objects sorted
in specific toolsets according to their usability domain [7].
Graphical objects in ArchestrA IDE can be adjusted in a solution where using
this graphical objects are very closely related to application objects and are a part
of other functional object in ArchestrA. This graphical object is usually added to
application object representing a process variable so the graphical object is linked
to the automation object.
Some graphical objects, specifically those from library Displays from graphic
toolbox, provide functionality only in a case of linkage with particular application
objects. Graphical objects from ArchestrA library toolset Displays implemented in
visualization application afford features of communication status or system
information therefore this graphics must be joined to the special application objects to
make these information accessible for visualization [8].
Several graphical elements and objects listed in graphic libraries with
predefined functionality intended for different industrial parts, can be considered
as fully complete and locked from graphical and functional point of view.
Although there is possiblity to do modifications on these objects also in ArchestrA
infrastructure through the Symbol Editor tool while common modifications of
graphical objects are available in configuration and development environment
WindowMaker of visualization product InTouch. The main development screen
consists of a grid panel and various components designed to creating graphic
elements with specific functionality.
Environment of the Symbol Editor is closely connected with .NET platform to
develop objects using graphic elements and scripting from .NET. Graphical
objects from ArchestrA library have modern design properties and consist of
graphical layers from elementary overlay to the shades and reflections. An option
to import InTouch SmartSymbol graphical objects generate opportunity to
combine ArchestrA and InTouch graphic elements. There is just the possibility to
create, configure and modify the Trend object in Symbol Editor.
So process of modification and complete configuration of ArchestrA graphical
objects is provided by a tool in ArchestrA IDE called Symbol Editor (Figure 1.)
triggered with opening a graphical object.
Fig. 1. Symbol Editor for creating, modification and configuration of graphical objects used
in visualization
4 Trends
Trend object is very important element in visualization application on supervisory
control level where monitoring of process variables in real-time conditions is
ongoing [11]. Trends demonstrate imaging of obtained data sequence of certain

process variable in a particular timeframe.
So human operator in dispatching center has outlined a controlled system situation
in real-time conditions thanks to existence of trend object functionality in visualization
application [9]. Therefore, we can say that presence of trend object in visualization
strengthens the informational value of data obtained from technological level.
From the time sequence of observed variables it can be resulting trend of
situation progress in controlled system - increasing the value of certain variables,
growth arrest or subsequent decline.
Readily visible graphical interpretation allows human operator to predict process
situation in next time period. We can argue that trend objects in visualization
application support visibility and prediction of process variables values [10].
An important part of trends is the option to view historical data progress using
setting process data archivation in industrial database server. This makes human
operators possible to see what happened in controlled process at certain moments
or a period of time.
In a case of process variable portraying in a short time interval there is not need to
use solution with industrial database server because trend objects in visualization
tools usually have a possibility to store process data for a short time in a buffer.
Nowadays the trend object in visualization application consists of modern
graphical layout with more functionalities and configurable options [11]. The
largest part of trend object is field intended to a process variables rendering
formed by the graph area with an option to adjust the grid.
The area of plotting graph consists of monitored part usually with grid chart
within the background performing readily visible displaying of pens of data
gained from process variables. The background should stay colorless to increase
and force the visibility of portrayed pens. The graph area for displaying is
composed of value axis Y and time axis X therefore pens are normally showing
the dependency of a process variable from the time.
So each process variable pen is represented as a line created by a series of
monitored and obtained values in time sequence from technological process level.
There is usually a legend integrated like a component of trend object or situated
near the object representing information in the usual form of block of information
about pens like color of the pen, process variable name or tag, actual value of
variable, commentary. One of frequently used functions of the trend object is to
display the cursors in the graph, which means creating a cross section
perpendicular to the timeline of displayed pens, and also to show the value change
of variables over time period in a numerical way.
The trend object within the supervisory application is usually implemented in
visualization screen or directly in external program modules specialized for advanced
trending and with some added functionalities and process data statistics [12].
4.1 Trend Objects

Trend object as a standard part of graphic symbols library in InTouch is also
included in newer versions of Application Server. This graphical object Trend has
a several differences from trend object in InTouch, it is situated in environment of

ArchestrA IDE, but not accessible from the section Graphic Toolbox like other
graphics. It is implemented as a component of a Symbol Editor where is placed in
the graphical elements toolbar.
During editing of the trend object in the main window of Symbol Editor, on the
left side there are individual elements and layers from which the trend object
composed. One of these elements is also Trend Client which provides trend object
functionality, it has no significance for configuration, but only ensures running of
trend object services in the background.
The configuration phase of trend object in Symbol Editor begins with options
in the Edit animations window and with the first card Pens which presents creation
and setting individual portrayed pens (Figure 2.). Each of pens can be named
according to the process variable and is linked to the source of plotted information
– observed process variable. The process variable can be represented by
automation object from ArchestrA IDE or by a tag from InTouch.
Fig. 2. Trend object configuration with historical options in own object’s menu, opened in
Symbol Editor
Process variable selection on automation object in ArchestrA IDE is realized in

Galaxy Browser – a tool intended to browse properties and data variables of
various objects in a project of ArchestrA IDE called Galaxy. The process variable
on automation object obtained from technological level of controlled system can
be entered in this form:
actual_temperature1.PV
InTouch tag-based process variable can be also used, but must be entered with
the provider of data in this form:
intouch:tagname
The area of Pen Details let to refer to a variable or text description or
combination of both text and data from variable as in this example:
"Actual value: " + intouch:tagname
In the next there are accessible options to set the range of process variable data,
pen style options, appearance issues of plot area, chart options and retrieval mode
with setting trend duration. ArchestrA trend object has additional scripting
possibilities to customize its settings and functionality.
In case of historical trending it is necessary to configure some application
objects to historize process data and of course an option of communication with
process database Historian Server in Historical Sources section of trend object and
complete with linkage to the variables in database.
5 Application Example
Controlled system is represented by a laboratory model intended for educational

and research purposes. Mentioned laboratory model Intelligent House is designed
with industrial control network LonWorks used in the field of building
automation. The application was developed in a package of Wonderware software
called Wonderware System Platform using products InTouch and Application
Server with ArchestrA technology.
5.1 Model Intelligent House
The Intelligent House model (Figure 3.) consists of two floors, which are rooms in
the house [13]:
• In the room upstairs is realized temperature control with temperature
sensor and heating device simulated by bulb. The room is equipped also
with two fans to possibility of air circulating and thereby the change of
temperature.
• In the room downstairs is realized control of lighting. There is placed
lighting sensor, the source of light is bulb and light coming from
surroundings through jalousie controlled by a stepper motor.
Essential controlled variables in the model Intelligent House there are lighting and
temperature. These variables are technologically represented in controlled system
by two bulbs in separated rooms. One of them perform lighting function and the
second one is used like a heating device.
Fig. 3. Model Intelligent House
System control is provided by network LonWorks composed of sPoint modules

and has next features:
• sensing lighting in the room,
• sensing temperature in the room,
• warm air exhausting,
• sensing lighting intensity,
• sensing heating intensity of heater.
Communication with higher level of control is implemented with Factory Suite
Gateway.
5.2 Application
Visualization application of model Intelligent House (Figure 4.) was created using
InTouch and ArchestrA software technologies and consists of main screen and
secondary screens.
On the main visualization screen there is implemented a real-time trend to show
progress of sensed process variables from technological level of control. The
plotting area consists of four pens with different color representing four process
variables of laboratory model.
Under the plotting area there is situated a component part of trend object Pen
Selector with checkboxes to select plotting of all four or only individual process
variables. Each pen can be selected by the checkbox to be showed on the chart.
The section Pen Selector also contains further information about pens and
associated process variables.
Fig. 4. The main visualization screen of controlled model Intelligent House with trend
object
In front of the screen in the part named Historical is a button to invoke a

separate window with historical trends. There is used trend object the same as for
real time data progress showing but configured to show process variables from
database with the same section of plotting area and Pen Selector.
Comparison of trend objects within the main visualization screen can be seen
on next figures. The first one (Figure 5.) represents InTouch trend object used for
visualize process data from technological level of laboratory model Intelligent
house same as the second one on Figure 6. from ArchestrA graphical objects.
Fig. 5. InTouch trend object integrated to the main visualization screen

Fig. 6. ArchestrA trend object integrated to the main visualization screen
6 Conclusion
The most pronounced advantage of containment of modern graphical objects with
some specific predefined functionality in visualization software for developing
visualization screens of supervisory control application is significantly faster
design and development of this application with the great investment to the future
application requirements including effective modification, maintenance and
deployment.
Trend object in visualization represents essential function of monitoring element
in visualization application within supervisory control of plant processes. Trend
provides monitoring of process variables data progress in certain time period and
also allows to review manufacturing processes data from a historical perspective.
Implemented and presented technologies participate on educational activities
and research in the Department of Cybernetics and Artificial Intelligence which is
also engaged in other products with various technologies of graphical trend of
values displaying with the clients of databases, ActiveX Controls panels, trends
for statistical process control.
Acknowledgments. This publication is the result of the project implementation

Development of Centre of Information and Communication Technologies for Knowledge
Systems (project number: 26220120030) supported by the Research & Development
Operational Programme funded by the ERDF (50%) and VEGA -1/0286/11 (50%). This
work was concluded in our laboratory with the software and licenses provided by
Wonderware Educational Consignment Support from Pantek (CS) s.r.o.
References
[1] Landryová, L., Babiuch, M.: Modelování objektů v průmyslových aplikacích
(Modeling Objects of Industrial Applications). In: Handbook of Research on Social
Dimensions of Semantic Technologies and Web Services, ch. XXXVI, pp. 743–759.
Information Science Reference (2009) ISBN 978-1-60566-651-8
[2] Popescu, D., Meng, Q., Ionete, A.A., Ionete, C., Aguridan, R.: Remote vs. Simulated,
Virtual or Real-Time Automation Laboratory. In: International Conference on
Automation and Logistics, Shenyang, China, pp. 1410–1415 (2009) ISBN 978-1-
4244-4794-7
[3] Flochová, J., Mrafko, L., Mudrončík, D., Auxt, F.: On Industrial Techniques and
Information Technology Control Classrooms. In: Proceeding of IEEE Conference on
Decision and Control 2007, New Orleans, pp. 1874–1879 (2007) ISBN 1-4244-1498-9
[4] Liguš, J., Zolotová, I., Karch, P., Ligušová, J.: Information and Control System of
Traverse and its Integration into Cybernetic Centre. Electronics and Electrical
Engineering (6), 147–152 (2010) ISSN 1392-1215
[5] Kolota, J.: A remote laboratory for learning with automatic control systems and
process visualization. International Journal of Engineering Education 27(5), 1130–
1138 (2011) ISSN: 0949149X
[6] Wonderware Information, http://www.wonderware.com
[7] Creating and Managing Archestra Graphics User’s Guide, Wonderware Archestra
user manual, Invensys Systems, Inc.
[8] Zolotová, I., Laciňák, L., Laciňák, S.: Machine and site supervisory control of
flexible production system. In: AEI 2011: International Conference on Applied
Electrical Engineering and Informatics 2011, pp. 27–31. TU, Italy (2011) ISBN 978-
80-553-0740-4
[9] Ponsa, P., Vilanova, R., Diaz, M., Goma, A.: Interface Design Improvement in
Supervisory Control. e-Minds: International Journal on Human-Computer Interaction 1(3)
(2007) ISSN 1697-9613
[10] Frederick, P., Gattass, M., Mediano, M.R.: Efficient visualization of graphical
objects. In: Proceedings of XII Brazilian Symposium on Computer Graphics and
Image Processing, Campinas, Brazil, pp. 73–81 (1999) ISBN 0-7695-0481-7
[11] Božek, P., Barborák, O.: Key benefits of virtual technology. Intellektuaľnyje Sistemy
v Proizvodstve 15(1), 115–119 (2010) ISSN 1813-7911
[12] Bober, P., Zgodavová, K.: Educational web tool for statistical process control. In: ICL
2011: 14th International Conference on Interactive Collaborative Learning:
Proceedings, pp. 460–464. IEEE, Piešťany (2011) ISBN 978-1-4577-1746-8
[13] Mihaľ, R.: Alarmy, trendy a bezpečnosť v supervíznom riadení. Diploma thesis.
TUKE, FEI, Košice (2010)
Mesurement of Energy Flows and CO2
Emissions Balance of the Low-Potential Heat
Source in Combination with a Cogeneration
Unit
Natália Jasminská
Technical university of Košice, Mechanical Engineering Faculty, Department of Power

Engineering, Vysokoškolská 4, 042 00 Košice, Slovakia
natalia.jasminska@tuke.sk
Abstract. Traditional ways of heating with natural gas or electric energy can be
effectively replaced or supplemented by alternative heating which can utilise heat
from the ambient environment. Improving the energy efficiency and developing
the renewable energy which closely relate to the general economy and social poli-
cy in most countries. They have a real potential to contribute to the sustainable
development and economic growth and can influence all areas of economic activi-
ty and are one of the sophisticated method out of the economic crisis. The article
is devoted to the use of non-traditional energy sources, which are the method of
heating and hot water preparation using heat pumps in combination with a cogene-
ration unit. The experimental part balances the system of heat pump and cogenera-
tion unit, the effect of the different parameters of the system operation in running
of the whole system.
1 Introduction
Growth in fossil fuels prices and the environmental consequences of their
extensive exploitation move the energy efficiency in recent years in the centre of
economic and political attention, whether at global, European as well as national
level.
The dependence on imported energy sources and the associated risks of price
fluctuations on the transformation of global energy markets in our economy
requires adequate response. Projected solutions are in the greatest possible but
cost-effective utilisation of domestic energy sources as well as in systematic
reduction of energy demand and the overall energy consumption.
Traditional ways of heating with natural gas or electric energy can be
effectively replaced or supplemented by alternative heating which can utilise heat
from the ambient environment around us.
The current energy situation leads to the general effort to exploit
unconventional sources of energy, among other things, so called low-potential
heat sources [1, 2]. Especially the systems of heat supply with low energy

64 N. Jasminská
potential using solar energy, geothermal water energy and the ambient
environment introduce a promising alternative in comparison with the limited
fossil energy carriers [6, 8, 10]. The suitability of their use depends on the
geographic area and climate zone in which we live [13, 14].
Despite the fact that in Slovakia there are real possibilities for using low
temperature energy sources, their exploitation in comparison with the other
European Union countries is negligible. The advantages of technologies that use
energy sources of low-potential in comparison with traditional energy carriers are
not only in the energy saving of fossil fuels within the heat supply systems of
buildings and constructions, but are related also to the ecological purity of the
environment [7].
The main objective of the contribution is to demonstrate the need and relevance
to focus on the alternative methods of exploitation and the related rationalization
of energy consumption in reducing the ecological load of the environment,
reducing the production of CO2 in the atmosphere.
2 Experiment
Measurement of low-temperature source of energy flows, which was formed by
tandem of a heat pump and a cogeneration unit had run in the object „Retirement
house and the house of social services (hereinafter referred as to RHHSS), which
also served as an experimental facility.
2.1 Characteristics of Low-Potential Energy

Source - Engagement of the System
The system was installed on the basis of using the tandem connection of the low-
potential energy source – heat pump and cogeneration unit.
The installation of the system heat pump and cogeneration unit is not only in-
teresting by the fact of using the combined production of the electric and thermal
energy, but the system is unique also because the cogeneration units produces the
electric energy, which powers the heat pump, and at the same time the cogenera-
tion unit and the heat pump generate heat, which supplies the heating water for the
object of RHHSS.
2.1.1 Principle of the Function of the Heat Pump in the System
Using the heat exchanger on the primary side (evaporator) is from the energy
source (well water) collected heat. As a result of added electric energy (by com-
pressor) is the heating water in the heat exchanger on the water side (condenser)
heated under control settings to + 15 °C up to + 60 °C. By doing so must be taken
into account the cooling power of the heat pump. [3, 14, 16, 19].
Heat pump takes away the low-potential heat from the water that is pumped
from the well. The well is the part of the object with the yielding of approximately
9 l.s-1 and after cooling down the water is channelled off to the absorbing well
Mesurement of Energy Flows and CO2 Emissions Balance 65
(Fig. 1). The parameters of the wells pumping and absorbing are verified by the
pumping test. The temperature of the pumped water is around 10 °C.
From the view of heat exchanger protection within the heat pump is the pre-
liminary heat exchanger designed as demountable. The thermal parameters of the
heat exchanger are 12 °C/8 °C – 11 °C/7° C power output of 90 kW.
The electric energy for the driver of the heat pump is produced in cogeneration
unit. At the same time the waste heat increases the temperature of the heated heat-
ing medium that increases the efficiency of utilisation of the primary fuel – gas.
Not covered power peaks are supplied from the existing interchange heating sta-
tion. For the possibility of the optimal switching heat pump and cogeneration units
is for the heating used the existing accumulation tank with the volume of 0,4 m3. It
also serves as the torus to eliminate the mutual influence of the circulating pumps.
Fig. 1. Scheme of the location of the low-potential heat source for the heat pump
2.1.2 Principle of the Cogeneration Unit Function within the System
It is a combustion 4-cylinder gas engine. The electric energy source is the asyn-
chronous generator with the output 30 kW. The operational temperature gradient
of the heating water is from 50 °C to 70 °C. The fuel is natural gas.
The generated electric energy is used to power drive the heat pump and to
power drive the other pumps. Any excess electric energy is sold to the distribution
network.
66 N. Jasminská
The cogeneration unit is placed in the system in series together with the heat
pump and at the same time increases the temperature of the heating water. If nec-
essary, it supplies the heat for heating or hot water.
2.2 Characteristics of Low-Potential Energy

Source - Engagement of the System
Temperature
In order to conduct the evaluation of the energy operation of the heat pump and
cogeneration unit system, they were chosen on the system 11 measuring points, on
which were installed 11 temperature recorders PCMICRO TE (Fig. 3).
The attached temperature sensor had a low thermal capacity with a low thermal
resistance between the pipe and the sensor with the capacity of 10 400 records.
Fig. 2. Software PCMICROLOG
The temperature recorders were placed in a total number of 11 units on the sys-
tem pipeline. Recording of the measured values was carried out automatically at
regular intervals 1 minute. Temperature recorders were then in consequence con-
nected to the computer by means of the software PCMICROLOG and the data
obtained were saved every day into a separate file (Fig. 2).
Scheme description (Fig. 3)
In the absorbing well (S1) with the depth of 8 m is placed the pressure pump
which is continuously connected with the activity of the heat pump (HP) and
cogeneration unit (CGU). The heat exchanger (HE) separates the water from the
well from the water on the primary side of HP to prevent clogging of the heat
exchanger of the heat pump by incrustation from the water. The heat exchanger
is a plate, demountable. The TR1 temperature meter measures the temperature
from the source well and TR2 temperature of the return water into the absorbing
well.
Fig. 3. Scheme of the connection heat pump and cogeneration unit with the energy and
temperature measurers
On the branch is installed the water-gauge/flow meter, which serves to measure

the volume flow of the water taken from the well and repeated release of water to
the well.
On the primary side of the HP are installed the circulatory pumps P1 – P4,
which are switched on together with HP and CGU.
The temperature meter TR3 measures the temperature of the water running out
from the heat pump and TR4 measures the temperature of the return to HP. For the
measurement of the energy flows are placed on the system the measurers M1 –
M7. Measurer M1 measures the energy taken from the low-potential heat source –
from the well and serves for the calculation of COP heat pump.
On the secondary side is installed the heat measurer M2, which measures the
amount of heat produced in heat pump and also serves for the calculation COP
heat pump. Measurer M5 – electrometer measures the electricity consumed to
power drive the HP.
The output water from the HP is the input for CGU and its temperature is in-
creased by the temperature of the heat produced in CGU. CGU burns the natural
gas which is measured by the gas meter G1. Thermal energy produced in CGU is
measured with the measurer M3. The measurer M6 measures the electric energy
produced in CGU. CGU has its own mixing valve MV1, which forms the circula-
tion circuit for CGU and provides the water temperature for CGU to prevent the
condensation of combustion gases. The temperature recorder TR10 records the
temperature of water flowing out of the HP after mixing after the additional heat-
ing from the CGU.
68 N. Jasminská
Three-way valve MV2 is the switching valve for heating the water, which
supplies the heat to the central heating (CH) and hot utility water HW. MV2 is set
to the mode MW only. Hot water is heated up in three tanks – accumulation tanks
with the capacity of 3 x 800 l through the plate exchanger.
The thermal energy for CH is recharged into the accumulation tank AT, where
is installed the heat meter M4. The pump P5 enables to take-off the heat from AT
to the system of CH. The pump P5 works continuously in the time of heating.
Bivalent heat source for additional heating of CH is from the interchange heat-
ing station (IHS) of the company TEHO Košice. Bivalent source is connected
directly to the system CH, where the water from the engine room and the IHS is
mixed.
For the overall measurement and evaluation of energy balances in the system
serves the measurer M7, which records produced – excess electric energy in CGU,
or lack of the electricity supplied to the engine room from the distribution net-
work.
Flow
To record the individual energy flows in the heat pump were measured the flows
by means of a flow meter on the circuit of the anti-freeze fluid, heated circuit and
on the heated side of the hot water preparation. The volume flow rate of the con-
sumed natural gas in the cogeneration unit was measured in the cogeneration unit
by means of the gas meter.
Thermal energy
The amount of thermal energy was recorded by means of the measurers M1-M4
positioned on the heat pump and cogeneration unit system pipeline (Fig. 3).
Electricity
Electricity was measured with the measurers M5 – M7. Positioning of the electric-
ity measurers is shown on the Fig. 3.
3 Results of Measurements
The measurement on the system heat pump and cogeneration unit “Experimental
source of heating“ has run for a long time.
To evaluate the measured data was selected the time period for the mode 1,
preparation of hot water and heating and mode 2, only hot water preparation. The
goal of evaluation is to present the energy flows as well as determine the value
COP and SPF for hot water preparation and heating, to evaluate the operation of
the heat pump and cogeneration unit and consequently also for the overall system.
The following results will continue to present the course of these values in de-
pendence on different parameters.
The system heat pump and cogeneration unit is throughout the year operated in
two modes:
 Mode 1 – it represents the mode of hot water preparation and heating,
 Mode 2 – in this mode the heating of central heating is not in operation,
the energy is supplied only for the hot water preparation.
3.1 Energy Balance of the Mode 1

During the selected measured time period (9 445 minutes) was the system HP and
CGU in operation for 5 969 minutes, which represents 63.2 % from the total
measured time period with the number of 133 switching of the heat pump and
cogeneration unit.
Average temperature in the well was 9.78 °C. The temperature on the output
from the heat pump reached the value of 44.40 °C and on the output from the co-
generation unit reached the value of 68.32 °C. The temperature development in the
system HP + CGU is shown on the Fig. 4.
Development of the external temperatures for the measured time period was as
follows:
 Highest temperature: 16 °C,
 Lowest temperature: 5 °C,
 Average temperature: 11 °C.
For energetic evaluation of the heat pump is used the term “coefficient of per-
formance” - COP (coefficient of performance) [5], which is defined according to
STS ES 14511 (STS – Slovak Technical Standard; ES – European Standard) as
the ratio of heat power output to the electric input power of the heat pump system
[17,18,19].
4.65 (1)
where:
Qk – the supplied thermal output (kWh),
Pk – the heat pump power consumption (kWh),
EM2 – energy produced by the heat pump (kWh),
EM5 – energy supplied to the heat pump (kWh),
m1 – mode 1.
Considerably greater importance for system efficiency evaluation however has the
seasonal power factor SPF more than coefficient of performance. The calculation
of seasonal performance factor SPF is not defined in the standard STS ES 14511.
If the factor SPF shall represent the significant parameters it is necessary to spec-
ify the system boundaries to determine heat and energy requirements, which were
included in the calculation, and take into account also the input powers of the
additional equipment and energy losses.
So, the SPF may be defined as a ratio of the heat delivered by the heat pump
during the overall period of operation in a calendar year and total amount of spent
driving energy of a compressor and other ancillary (auxiliary) equipment of circu-
lation (pumps, fans, defroster of vaporizers and other equipment consuming the
driving primary energy).
70 N. Jasminská
Typical day - mode 1 TR1

80 TR2
70 TR3
60 TR4
Temerature (ºC)
50 TR5
40 TR6
30 TR7
TR8
20
TR9
10
TR10
0
TR11
0:00 3:20 6:40 10:00 13:20 16:40 20:00 23:20
Tam
Time (hours)
TR 1 – well water temperature, TR 2 –temperature of water returning back to the well,

TR 3 – heat pump water temperature, TR 4 – temperature of return water of the heat pump,
TR 5 – temperature on the output from the cogeneration unit, TR 6 – temperature of return water
of the cogeneration unit, TR 7 –temperature of mixing, TR 8 – temperature of water inputting the
heating, TR 9 temperature of return water of heating, TR 10 – temperature of water inputting
the hot water heating, TR 11 temperature of return water of the hot water heating, Tam – outer
measured temperature
Fig. 4. Course of system measurement temperatures of HP + CGU in selected time segment

for typical day in the mode 1
Power input of ancillary (auxiliary) equipment of top quality units is approxi-

mately 10 % of the total required power input (for less quality units, 15 % may be
considered) [3], in our we considered the value of 11 %. So the SPF may be de-
termined according to the following equation:
∑
4.15 (2)
∑
where:
∑Qc – Amount of delivered heat rates (kWh),
∑Qt – total heat amount (kWh),
∑Pt – total power inputs (kWh),
EM2 – Energy produced by heat pump (kWh),
EM5 – Energy delivered to heat pump (kWh).
E(P1+ P2+ P3+ P4+ P5) – energy necessary to drive the pumps P1-5 (kWh).
m1 – mode 1.
3.1.1 Utilization of Natural Gas Use in the Mode 1
Low-energetic system of heat pump and cogeneration unit connection is proposed

for the purposes of minimizing of the primary energy consumption with the direct
consequence of lower operational costs when comparing with the classical heat
sources and IHS prospectively.
The system is proposed so that the cogeneration unit is driven by natural gas,
produced electric energy serves for drive of heat pump and, in the case of excess
of non-consumed electric energy, it is delivered to the public electric network.
Heat energy produced in cogeneration unit is delivered to the system for hot water
heating and central heating prospectively.
Based on measurements reached for the selected time segment it is possible to
determine the measured energy amounts obtained and consumed from individual
energetic sources:
QW m1 = 25.82 GJ (7 172.22 kWh),
QHP m1 = 32.90 GJ (9 138.89 kWh),
QCGUm1 = 22.40 GJ (6 222.22 kWh),
QHW m1 = 30.60 GJ (8 500.00 kWh),
QCH m1 = 24.70 GJ (6 861.11 kWh).
where:
QW m1 – amount of heat obtained from the low-potential source – well,
QHP m1 – amount of heat produced by the heat pump,
QCGU m1 – amount of heat obtained in cogeneration unit,
QHW m1 – amount of heat delivered to the hot technical water,
QCH m1 – amount of heat delivered to the heating,
m1 – mode 1.
Amount of electric energy produced in cogeneration unit for the selected meas-
ured time segment is ECGU m1 = 8.94 GJ (2 483.33 kWh).
Amount of electric energy consumed in heat pump for the selected measured
time segment is EHP m1 = 6.44 GJ (1 788.89 kWh).
Based on the measured data it is possible to determine the amount of excessive
electric energy Enm1 (delivered to the network in kWh) from the following
equation:
0.28 i. e. 28 % (3)
The cogeneration unit utilized 905.0 m3 of natural gas during the selected period.
The manufacturer of the cogeneration unit indicates the maximum electric power
output of 25 kW and maximum heat rate of 47 kW using the fuel for 90.5 %. The
total efficiency (real usage of fuel utilization) was determined from the following
equation:
72 N. Jasminská
0.80 i. e. 80 % (4)
·
where:
ηr – real fuel utilization (%),
PE – CGU electric power output (kWh),
PH – CGU heat rate (kWh),
qNG – Heating capacity of the natural gas (MJ.m-3),
– Volumetric flow of the natural gas (m3.s-1).
With 80 % fuel utilization, the CGU thermal losses are 20 %.
Based on the above mentioned calculations, it is possible to appraise the con-
nection system operation of the thermal pump and cogeneration unit by means of
simple energy budget representing a sum of some parameters influencing the en-
ergy consumption and demand for the system operation and which are important
for energy source appraisal.
3.2 Energy Budget of the Mode 2

In the mode 2, i. e. in the mode of hot water preparation only, measured date al-
ways represents a time segment. During the selected measured time segment (10
047 minutes) HP and CGU systems were in operation for the duration of 3 347
minutes representing 33.30 % of the overall measured section in amount of 133
switching-on of the heat pump and cogeneration unit. Average temperature in the
well was 10.61 ºC. Temperature on the heat pump output reached the value of
49.40 ºC and on the cogeneration unit output reached the value of 67.62 ºC.
Course of external temperatures within the measured time segment was:
 The highest temperature: 15 ºC,
 The lowest temperature: 9 ºC,
 Average temperature: 12 ºC.
The connection system operation of the thermal pump and cogeneration unit in
mode 2 is illustrated in the figure 5 which was selected as typical measurement day.
In the mode 2, the COP reached the value calculated according to the equation:
4.27 (5)
Where:
Qlps – Heat from the low-potential source (well) (kWh),
Php – Heat pump power input (kWh),
EM2 – Energy produced by heat pump (kWh),
EM5 – Energy delivered to the heat pump (kWh).
m2 – mode 2.
Typical day mode 2 TR1
80 TR2
70 TR3
Temperature (°C)
60 TR4
50 TR5
40 TR6
30 TR7
20 TR8
10
TR9
0
TR10
0:00 3:20 6:40 10:00 13:20 16:40 20:00 23:20
TR11
Time (hours) Tam
TR 1 – well water temperature, TR 2 –temperature of water returning back to the well,

TR 3 – heat pump water temperature, TR 4 – temperature of return water of the heat pump
TR 5 – temperature on the output from the cogeneration unit, TR 6 – temperature of return water
of the cogeneration unit, TR 7 –temperature of mixing, TR 8 – temperature of water inputting the
heating, TR 9 temperature of return water of heating, TR 10 – temperature of water inputting
the hot water heating, TR 11 temperature of return water of the hot water heating, Tam – outer
measured temperature
Fig. 5. Course of system measurement temperatures of HP + CGU in selected time segment

for typical day in the mode 2
Coefficient of performance of the heat pump in the mode 2 shows the similar
course as that in the mode 1. In consequence of higher condensation temperatures
during the hot water heating the values of the course are lower when comparing
them with mode 1.
Seasonal factor SPF was determined as in the mode 1 according to the follow-
ing equation:
∑
3.77 (-) (6)
∑
Where:
∑Qt – total heat amount (kWh),
∑Pt – total power inputs (kWh),
EM2 – energy produced by the heat pump (kWh),
EM5 – energy delivered to the heat pump (kWh),
E(P1+ P2+ P3+ P4+ P5) – energy required for drive of P1-5 system pumps (kWh).
74 N. Jasminská
3.2.1 Utilization of Natural Gas Usage in the Mode 2
In the mode 2, the system of heat pump and the cogeneration unit delivers the heat
to prepare the hot water only. Based on the measurements obtained for the se-
lected time segment it is possible to determine the amounts of energies gained and
consumed from the individual energetic sources.
QW m2 = 13.25 GJ (3 680.56 kWh),
QHP m2 = 17.30 GJ (4 805.56 kWh),
QCGUm2 = 11.60 GJ (3 222.22 kWh),
QHW m2 = 28.90 GJ (8 027.78 kWh),
QCH m2 = 00.00 GJ (0.00 kWh).
where:
QW m2 – amount of heat obtained from the low-potential source – well,
QHP m2 – amount of heat produced by the heat pump,
QCGU m2 – amount of heat obtained in the cogeneration unit,
QHW m2 – amount of heat delivered to the hot technical water,
QCH m2 – amount of heat delivered to the heating,
m2 – mode 2.
Amount of electric energy produced in the cogeneration unit for the selected
measured time segment is ECGU m2 = 5.58 GJ (1 550.00 kWh).
Amount of electric energy consumed in the heat pump for the selected meas-
ured time segment is EHP m2 = 4.51 GJ (1 252.78 kWh).
Based on the measured data it is possible to determine the amount of excessive
electric energy Enm2 (delivered to the network in kWh) from the equation:
0,19 i.e. 19 % (7)
In the mode 2, the cogeneration unit utilized for the selected time segment
575.50 m3 of natural gas. Even in the mode 2, real fuel utilization (total efficiency)
is 80 %, with the CGU losses being 20 %.
4 Discussion
The connection system operation of the thermal pump and cogeneration unit in
mode 2 is illustrated in the figure 4 which was selected as typical measurement of
mode 1 and figure 5 representing the typical day of the mode 2.
Fundamental of the low-temperature system of „Experimental heat source“ is
the division of the heat produced into the heat delivered to the heating system and
warm water preparation in the mode 1 and in the case of the mode 2 to hot water
preparation only and electric energy consumption in the heat pump. A decisive
factor for the energy consumption in the low-temperature system is the operation
method of the cogeneration unit during a day.
Heating of objects through the system of heat pump operates in the batch mode
with switching on and off occurring quite frequently. With reducing temperature
of heated water of 56/51°C with heating bodies, number of switching on and off is
increased. On the other hand, the cogeneration unit is constructed for the continu-
ous run of operation.
Important increase of the number of switching on and off during the opera-
tion in batch mode of heat deliveries stands for significant increase of failure
condition occurrence probability simultaneously. Presently, nor even required
level of the repair and maintenance system of the cogeneration unit is solved
where the repair durations cause long cogeneration unit standstill.
As seen in the figures 4 and 5, the heat pump switched together with the cogenera-
tion unit in the switching amount of 20 with the average pause periods of 20 minutes.
Higher frequency of switching influences the efficiency as well as life time of
the system very negatively. Proportion of switching on the efficiency losses is ap-
proximately 10 %.
The utilization of the fuel in both modes is mainly influenced by:
• Proportion of the production of heat and electric energy in CGU (this
factor is considered to be a constant because the proportion of the produc-
tion of electric energy and heat is given by the producer).
• COP influences the gas utilization in the CGU very significantly
(Figure 6). The higher is COP, the higher is the gas utilization.
• Electric power output excess, as COP, influences the gas utilization but
contrarily, the lower is the electric power output, the higher is the gas
utilization in the CGU.
Natural gas utilization is approximately the same in the mode 1. In the mode 2
lower COP, but also lower electric power output excess are reached.
Optimum solution for maximum gas utilization in the system of HP and CGU is
to reach the maximum COP with minimum electric power output excess.
2.200 En = 0 %
2.000 En = 5 %
Gas utilization (-)
1.800
En = 10
%
1.600
En = 15
1.400 %
En = 20
1.200 %
En = 25
1.000 %
3.0 3.5 4.0 4.5 5.0 5.5 En = 30
%
COP of the heat pump (-)
En – electric power output excess (%) – electric energy delivered into the electric system
Fig. 6. Gas utilization in the system of HP + CGU depending on COP and electric power
output excess
76 N. Jasminská
In the figure 6, the course of fuel utilization depending on COP and electric
power output excess for various values is illustrated.
As seen in the Figure 6, the CGU operation for electric power output seems to
be the most suitable in the given HP + CGU tandem which corresponds with the
HP requirement in the given time. I. e. the electric energy excess has to be equal
zero and minimum prospectively.
5 Ecological Evaluation of the Heat Pump and Cogeneration

Unit System
Development of ecological situation related to atmosphere pollution by energetic
plants was focused on elimination of the CO2 production from primary source of
the company TEKO Košice municipal heating station operated in combined cycle
Table 1. CO2 emission production during heat production by various technologies
CO2 EMISSION PRODUCTION WHEN PRODUCING HEAT BY VARIOUS

TECHNOLOGIES
Coefficient Efficiency
COP
of CO2 of energy CO2 emis- CO2 emis-
Heat production when pro-
emissions delivery sion pro- sion pro-
method ducing
when pro- depending duction duction
the heat
ducing on source
(kg.kWh-1) (%) (-) (kg.kWh-1) (kg.GJ-1)
Water/water
1 heat pump
0,2050 87 3,10 0,0760 21,1146
heating bodies
Water/water
heat pump
2
Floor central 0,2050 87 4,20 0,0561 15,5846
heating
Water/water
heat pump
3
Capillary 0,2050 87 5,40 0,0436 12,1213
central heating
Water/water
heat pump
4
Floor central 0,2050 87 3,20 0,0736 20,4548
heating
Direct central
5 heating
0,2050 87 1,00 0,2356 65,4553
Electric
Biomass 1,00
6 0,0859 87 0,0987 27,4170
Heat
Gas
7 0,2500 87 1,00 0,2874 79,8190
Heat
Heat from
8 steam-gas 0,2791 79 1,00 0,3533 98,1429
cycle
CHSS/TEKO
9 0,3333 79 1,00 0,4219 117,2025
Heat
of production of heat and electric energy on the fuel basis of black coal and natu-
ral gas which assures the supporting services of CHSS (Central Heat Supply Sys-
tem) – top level supplies for heating and hot water preparation through the com-
pany TEHO into the object of RHHSS.
For parameterisation of the CO2 greenhouse gas exposition, no model solution
was constructed, but coefficients of “CO2 emission production during heat produc-
tion from various fuel kinds” (Table 1) was used.
Determination of basic comparing position according to the coefficient of the
CO2 emission production (Table 1) in TEKO transformed to the place of heat
consumption for the heat pump and cogeneration connection system was deter-
mined according to the following equation:
· 982 5380 · 0,4219 414 532,8 kg. year

414,5 t. year
where:
Qi - Amount of heat necessary for CH and HW heating inclusive of heat
from CHSS in the consumption place (kWh.year-1),
- CO2 emission production calculated to the place of heat
consumption (kg.year-1, t.year-1),
- Specific CO2 TEKO emission production calculated to the place of
heat consumption (kg.kWh-1).
From methodical reasons, two options were defined for calculations:

Option No. 1 – heat pump and cogeneration unit tandem as well as supporting
services from CHSS assuming that the cogeneration unit will cover the electric
energy demand for heat pump and excessive electric energy will be sold to the
public electric network.
Option No. 2 - operation of single heat pump and supporting services from CHSS
assuming that the electric energy necessary for heat pump drive will be supplied
from the public electric network completely.
Determination of emission production in the Option No. 1 – heat supply of build-
ing from local source of the heat pump and cogeneration unit tandem and deliver
of heat from the system of CHSS to assure the supporting services will be accord-
ing the equations:
MNG · 457 256 · 0,230 105 168,88 kg. year

105,17 t. year
· 276 667 · 0,4219 116 725,8 kg. year

116,72 t. year
78 N. Jasminská
CO2 emission production of the heat pump and cogeneration unit tandem and
supporting services from CHSS was determined from the equation:
105 168,88 116 725,8 221 894,68 kg . year

kde: 221,89 t . year
where:
- CO2 emission production of the heat pump and cogeneration unit
tandem (kg.year-1, t.year-1),
- CO2 emission production of the supporting services from CHSS
(kg.year-1, t.year1),
- CO2 emission production in the Option No.1 (kg.year-1, t.year-1),
- Amount of heat from CHSS (kWh.year-1),

- Coefficient of CO2 emission from the natural gas fuel (kg.kWh-1),
- Specific CO2 TEKO production emission calculated to the
consumption place (kg.kWh-1).
- Natural gas consumption for heat and electric energy production for
the heat pump drive (kWh.year-1).
Determination of the emission production in the Option No. 2 – heat supply of
a building from local heat source, heat pump, heat deliveries for supporting ser-
vices from CHSS and electric energy deliveries for drive of heat pump from pub-
lic electric network:
CO2 emission production by heat pump
· 572 268 · 0,0760 43 492,36 kg. year

43,49 t. year
where:
- CO2 emission production by heat pump (kg.year-1, t.year-1),
- Amount of heat produced in heat pump (kWh.year-1),
- Calculated specific CO2 emission production by heat pump
(water/water) supplied from the public electric network by
electro-energetic mix of electric energy (kg.kWh-1),
CO2 emission production by delivery of supporting services from SCZT
· 410 270 · 0,4219 173 092,90 kg. year

173,11 t. year
Where:
- CO2 emission production of supporting services from CHSS
(kg.year-1, t.year-1),
- Amount of heat from CHSS (kWh.year-1).
- Specific CO2 TEKO emission production calculated to the place of
consumption (kg. kWh-1).
CO2 emission production in the Option No. 2:
43 492,36 173 092,9 216 585,26 kg . year

216,58 t . year
where:
- CO2 emission production by heat pump (kg.year-1, t.year-1),
- CO2 emission production of supporting services from CHSS
(kg.year-1, t.year-1),
- CO2 emission production in the Option No. 2 (kg.year-1, t.year-1).
Saved CO2 emission production in the heat pump and cogeneration unit con-
nection system
Yearly volumes of saved CO2 emissions were calculated by difference of CO2

emissions in individual options against the comparing base as if the overall heat
delivery (Q = 3 537 GJ.year-1) was saturated from CHSS.
Options under consideration are in the conditions of energetic heat supply from
renewable resources a very important local contribution to obligations of Slovakia
to reduce the CO2 production. Both options are comparably valuable by their con-
tribution to reduction of the CO2 greenhouse gas by approximately 50 %.
Table 2. Saved CO2 emissions in individual options
Index Option No. 1 Option No. 2

-1
Saved CO2 emissions (t.year ) 192,6 197,0
-1
Comparing base of CO2 emissions (t.year ) 414,5 414,5
Saved emissions (%) 46,4 47,4
Conclusion of the Ecological Appraisal
Transformation of the energetic market toward to renewable resources of energy is

planned for next 40 years and as such, it will require spending of some
80 N. Jasminská
transformation cost. It is suitable to solve the measuring of the transformation cost

through the measurement of CO2 emission production with the energy delivery to
the market in parallel [4, 9, 11, 12]. As a suitable instrument in the form of eco-
nomical stimulus, green bonus as a price of inverse value of the CO2 emission unit
measured in €.t-1 of non-produced CO2 emissions.
Applying the above mentioned principles, it is possible to reduce the energy
consumption and emissions of CO2 greenhouse type using the individual tech-
nologies. Technology changing the conditions among the parameters investigated
is the technology using the low-potential energetic source through the heat pump.
6 Conclusion
As a conclusion it is to be stated that the balance of low-temperature energetic
system of connection showed existing reserves in the efficiency of combined
sources of heat and electric energy production in method of management of opti-
mizing of required performance.
Heat pump and cogeneration unit connection system was implemented for the
purposes of heat production from renewable resource and replacement of electric
energy from public electric network for the heat pump drive by electric energy
produced in cogeneration unit burning the natural gas. Energy (gas) is led to the
place of consumption directly whereby energetic losses are minimized.
Characteristics of the given solution:
• Gas burnt in cogeneration unit is changed into heat and produced electric
energy is utilized in the heat pump during the heat production (this value
depends on hear pump COP) multiply.
• Heat originated upon the electric energy production in a combustion en-
gine is used for temperature increase outputting from the heat pump. So
the heat pump is allowed to work in the mode with lower temperature
guaranteeing higher COP value.
• The equipment tandem is operated according to the heat demand in paral-
lel. Eventual excessive electric energy is consumed in the process of heat
energy distribution for circulating pumps or into the electric network.
• Resulting electric effectiveness is expressed by „gas utilization“ (gas in-
putting the cycle in amount of 100 % produces the heat and, thank to
electric energy, the heat pump produces heat from low-potential source –
well). Total energy produced exceeds the gas energy inputting the cycle.
Determining parameter for durably effective operation of the heat pump and the
cogeneration unit tandem is the same concurrent reliability (no-failure operation)
of individual elements of the tandem directing to the categorical imperative.
Because it results from the energetic statistics, that 40 % of consumption falls
to buildings, than the analysed system of low-potential, renewable energy source
use in energetic economy of ten thousands is usable conceptual energetic inten-
tion, the effectiveness of which will be manifested by estimation from 2015
plainly. Generally, this horizon is considered as a beginning of more assertive

burdening of fossil fuel use by progressive consumption tax.
Use of primary energy from low-potential source – ground water of the Košice
town is a strategic trend advantageous for further conceptual solution in develop-
ment of energetic supply of the Košice town as a whole, but mainly its individual
heat consumers (residential and municipal sphere) with CO2 reduction potential of
43 %, however legislation environment shall be created as well as financial gov-
ernment support shall be provided to implement and operate such systems effec-
tively, if Slovakia wants to meet the EU requirements, i. e. to cover 20 % of en-
ergy consumption by renewable resources until 2020.
Acknowledgments. “This article originated thanks to support within the Research and
Development Operational Programme for the project: Centre for efficient integration of the
renewable energy sources with the code ITMS: 26220220064, co-financed by the resources
of the European Regional Development Fund”.
References
[1] Ajah, A.N., et al.: On the robustness, effectiveness and reliability of chemical and
mechanical heat pumps for low-temperature heat source district heating: A compara-
tive simulation-based analysis and evaluation. Energy 33(6), 908–929 (2008)
[2] Bullard, C.W.: Heat pumps and combined heat and power a technology assessment.
Energy Policy 16(6), 579–593 (1988) ISSN: 0301-4215
[3] Čarnogurská, M., Příhoda, M., Brestovič, T.: Teplotná a napäťová analýza
membránovej steny výparníka kotla (Thermal and stress analysis of the membrane
wall of the boiler evaporator). In: Tepelná Technika v Teorii a Praxi: Vědecká Konfe-
rence k 50. výročí Založení Katedry Tepelné Techniky: Sborník Přednášek, August
31-September 2, pp. 26–31. VŠB-TU, Ostravice. - Ostrava (2011) ISBN 978-80-248-
2463-5
[4] Čarnogurská, M., Brestovič, T., Příhoda, M.: Modelling of Nitrogen Oxides Forma-
tion Applying Dimensional Analysis. Chemical and Process Engineering 32(3),
175–184 (2011) ISSN 0208-6425
[5] Girault, M., et al.: Estimation of time-varying heat sources through inversion of a low
order model built with the Modal Identification Method from insitu temperature mea-
surements. International Journal of Heat and Mass Transfer 53(1-3), 206–219 (2010)
[6] Kalsson, F., Fáhlen, P.: Capacity-controlled ground source heat pumps in hydronic
heating systems. International Journal of Refrigeration 30(2), 221–229 (2007)
[7] Kulcar, B., Goricanec, D., Krope, J.: Economy of exploiting heat from low-
temperature geothermal sources using a heat pump. Energy and Buildings 40(3),
323–329 (2008)
[8] Lichner, M., Nagy, R.: Energetická a ekonomická efektívnosť obnovy budov na
bývanie v Košiciach (Energetic and economic efficiency of the residential buildings
renovation in Košice). In: Vykurovanie 2008, pp. 443–449. SSTP, Bratislava (2008)
ISBN: 9788089216192
[9] Liping, J., Bin, C.: An Input-output. Model to Analyze Sector Linkages and CO2
Emissions Procedia Environmental Sciences 2, 1841–1845 (2010)
82 N. Jasminská
[10] Pinka, J., Wittenbergerer, G., Sidorová, M.: Možnosti využitia geotermálnej energie
na Slovensku (Possibilities of geothermal energy utilization in Slovakia). Zborník
Vedeckých Prác VŠB-TU Ostrava 51(1), 225–230 (2005) ISSN 0474-8476
[11] Rao, G., Wang, Y., Li, K., Wang, N.: Analysis on Chongqing Industry CO2
Emissions Efficiency Difference and Its Emissions Reduction Potentials. Energy
Procedia 5, 2230–2235 (2011)
[12] Quadrelli, R., Sierra, P.: The energy–climate challenge: Recent trends in CO2 emis-
sions from fuel combustion. Energy Policy 35(11), 5938–5952 (2007)
[13] Sakellari, D., Forsén, M., Lundqvist, P.: Investigating control strategies for a domes-
tic low-temperature heat pump heating system. International Journal of Refrigera-
tion 29(4), 547–555 (2006)
[14] Haiwen, S., et al.: Quasi-dynamic energy-saving judgment of electric-driven seawater
source heat pump district heating system over boiler house district heating system.
Energy and Buildings 42(12), 2424–2430 (2010)
[15] Tauš, P., Taušová, M.: Ekonomické posúdenie návratnosti vybraných druhov OZE
(Economic return evaluation of selected types of renewable energy sources). In:
Možnosti Financovania Ekoenergetických Projektov v Podmienkach SR a EÚ,
pp. S134–S140. Dom techniky ZSVTS, Košice (2006) ISBN: 8023202626
[16] Urchueguía, J.F., et al.: Comparison between the energy performance of a ground
coupled water to water heat pump system and an air to water heat pump system for
heating and cooling in typical conditions of the European Mediterranean coast. Ener-
gy Conversion and Management 49(10), 2917–2923 (2008)
[17] Vranay, F., Vranayová, Z., Lukašik, D.: Application of renewable energy sources in
central heat supply systems. Czasopismo Techniczne: Budownictwo 107(4), 221–228
(2010) ISSN: 0011-4561
[18] Vranay, F., Vranayová, Z.: Heating for intelligent building - case study. Visnik
nacionaľnovo Universitetu Ľvivska Politechnika (627), 226–231 (2008) ISSN: 0321-
0499
[19] Bi, Y., et al.: Comprehensive exergy analysis of a ground-source heat pump system
for both building heating and cooling modes. Applied Energy 86(12), 2560–2565
(2009)
Part II
New Approaches in Mechatronics,
Modeling and Human Knowledge
Intelligent Supervisory System for Small
Turbojet Engines
Rudolf Andoga1, Ladislav Madarász2, Tomáš Karoľ2, Ladislav Főző3,

and Vladimír Gašpar2
1
Technical University of Košice, Faculty of Aeronautics, Department of Avionics
2
3
Technical University of Košice, Faculty of Aeronautics, Department of Aviation
Engineering
{rudolf.andoga,ladislav.madarasz,tomas.karol,
ladislav.fozo,vladimir.gaspar}@tuke.sk
Abstract. The paper deals with design of an intelligent supervisory system usable
in the area of engineering applications of small turbojet engines. Such supervisory
system consists of diagnostic and control system applied in different phases of the
engine’s operation. In order to design it we need to know the object and its models
that will be realized by means of situational modeling concept and neural
networks. Diagnostic system will be based on these types of models applied into
the majority voting method of online diagnostic. Intelligent supervisory system
will also include off-line pre-start diagnostics for such engine with perspective
health monitor database.
Keywords: turbojet engines, diagnostics, artificial intelligence, neural networks.
1 Introduction
Turbojet engines in general can be treated as complex dynamic systems that are a
vital part of the whole aircraft and their operation is crucial for safety of aircraft.
Nowadays application of small turbojet engines is beginning to shift from
auxiliary power units into area of unmanned flight systems propulsion. Therefore
it is becoming also necessary to achieve high level of operability of such engines.
Application and design of such intelligent methods of supervisory control will
allow expanding its application in standard sized engines used for propulsion of
transport and also military manned aircraft [7].
The aim of the article is to present a system that detects failures and eliminates
them in a complex system. Failure detection is divided into two basic parts – pre-
start auxiliary systems diagnostics and online diagnostics. In the area of pre-start
diagnostics and system supervision a real-time system of diagnostics is proposed
and is applied in the phase of engine start-up. The system is aimed to check all
auxiliary systems, evaluate their state and proceed with engine start-up with

86 R. Andoga et al.
feedback from its sensors. Most of the present start-up systems do not use real-
time feedback and their start-up sequence is usually time based. Expert systems
and fuzzy inference systems in the area of the engine’s start are proposed that can
intelligently drive the engine from cold into its hot state [3,7].
The first chosen diagnosed parameter during operational state of the engine is
one of the most crucial parameters of any engine and it is the speed of the engine.
The present control systems usually control the speed of the engine and failure in
the area of speed measurement can lead to a catastrophic failure of the whole
engine. We propose a novel diagnostic and backup system that will be tested on a
small experimental turbojet engine MPM-20/21 and act as a diagnostic and
backup system for optical sensor that measures the speed of the engine. We
propose two different dynamic models of the speed of the engine with the input
from the fuel flow supply that will act as a backup to the optical measurement.
The first model is in the form of a trained neural network on the measured data of
the MPM-20/21 engine and the second model used as a backup is done with the
method of successive integrations again from the measured data of the engine.
However the diagnostic and backup system is modular so any other forms can be
used as backup measurements, like speed measurement using voltage of an
alternator or measurement of speed using magnetic field. The presented approach
will be applied into a modular architecture that can be further expanded and
generalized for other parameters of small turbojet engines with its perspective
application in normal sized turbine engines.
2 Small Turbojet Engines

The main application of small turbine engines has always been in the area of using
such engines as primary starters of main aircraft engines and auxiliary power units
in aircraft. Industrial applications include use of these engines mainly as power
generator units [3,7]. Novel applications use such engines in the area of unmanned
aerial vehicles and other small aircraft in the form of turbojet engines. This vast
area of applications implies research and advancements in control systems of these
engines. Small turbine engines also represent a good class of experimental
laboratory engines that is suitable for real-world experiments and as
thermodynamic processes in such engines can be compared to standard engines,
obtained results and advancements can be expanded also for this class of engines.
One of the steps to improve efficiency, safety and lifespan of such complex
system is to design an intelligent supervisory system of such small engine that will
operate above its traditional or advanced control system. To achieve this aim we
use a small turbojet engine MPM-20/21 in the Laboratory of intelligent control
systems of aircraft engines.
The experimental engine MPM-20/21 has been derived from the TS – 20/21
engine, which is a turbo-starter turbo-shaft engine previously used for starting
engines AL-7F and AL-21F. The engine has been rebuilt to a state, where it
represents a single stream engine with radial compressor and a single stage non-
cooled turbine and outlet jet. The basic scheme of the engine is shown in the
figure 1.
Intelligent Supervisory System for Small Turbojet Engines 87
Fig. 1. The measured data of fuel flow and engine speed
All used sensors, except fuel flow and rotations sensor, are analogue and utilize
voltage output. The data acquisition has been done in LabView environment. The
digital measurement of parameters of MPM-20 engine in real time is important to
create a model and control systems including the intelligent supervisory control
system complying with FADEC definition („Full Authority Digital Electronic
Engine Control“). The graph in the figure 2 shows dynamic changes of parameters
of the engine to changes of fuel supply input.
Fig. 2. The measured data of fuel flow and engine speed
The following basic parameters are measured:

• air temperature at the outlet from the diffuser of the radial compressor -
T2C [°C];
• gas temperatures in front of the gas turbine - T3C [°C];
• gas temperature beyond the gas turbine - T4C [°C];
• static pressure of air beyond the compressor P2 [Ata];
• static pressure of gases in front of the gas turbine P3 [Ata];
• static pressure of gases beyond the gas turbine P4 [Ata];
• fuel flow supply Qpal [l/min];
• thrust Th [kg];
• rotations of the turbine/compressor, n1 [rpm].
88 R. Andoga et al.
All parameters of the engine can be used to create fully autonomous control
system that runs on the control computer and is used either to control the chosen
parameters of the engine and also used to control different auxiliary systems in
order to perform actions like starting up the engine, health monitoring, backing up
important sensors and shutting the engine down.
3 Modeling the Small Turbojet Engine

In order to design and build any control system including a supervisory one, it is
needed to create mathematical models of the chosen object of control. Different
aims require design of different mathematical models. For the small turbojet
engine MPM we build three different types of mathematical models:
• analytic models for design and precise analysis of thermodynamic
processes,
• dynamic control models for design of real-time controllers of the
engine,
• diagnostic models to simulate errorless outputs of different sensors
and for backup.
3.1 Analytic Modeling

Static and dynamic properties of turbojet engines (MPM-20) can also be described
by a mathematical model of operation single stream engine under equilibrium or
non-equilibrium conditions. This will allow modeling the thrust, fuel
consumption, pressures and temperatures of the engine by different altitudes and
velocities in the chosen cuts of the engine [8].
The steady operation of the engine is such a regime, where in every element of
the engine same thermodynamic processes are realized. Operation of an engine in
its steady operation can be described by:
• algebraic equations of balance of mass flow of working materials
through nodes of the engine, equations of output balance, equations of
regulation rules and equations describing particular oddities of an
engine. A system of equations expresses that for given outer
conditions of operation of an engine, characteristics of all nodes of an
engines and preset values of control parameters (fuel supply, cross
section of the output nozzle, angle of compressor blades), operation of
the engine will settle itself on one and only one regime.
• graphically by utilization of knowledge of characteristics of all parts
(output, compressor, turbine, etc) of the engine and their preset curves
of joint operations (e.g. lines of stable rations of T3c/T1c in
compressor). Designation of all curves of the engine is done in a way
that we will try to fulfill continuity conditions for all parts of the
engine and characteristics of all these parts are given. These
characteristics can be found by direct measurement, computation, etc.
3.2 Situational Modeling of the MPM-20 Engine

The situational model of the engine is suitable for control systems design and can be
also employed in its diagnostic system. The main advantage of such model is the
ability to describe individual dynamical dependencies that more complex and cannot
be depicted by static analytic models. In experimental data driven modeling we will
consider T2c, T4c, P2c, n as output parameters dependant on the parameter of fuel
supply Qpal that can be seen in the figure 2. These dependencies are not stationary
and are changing during course of turbojet’s engine operation. In such case it isn’t
possible to easily set operating points in multi-dimensional non linear parametric
space. The operating point will lie on a functional of the following parameters:
Op = f (Qp,N1,T4,T2, P2) (1)
To decompose the model into subspaces, we will use the methodology of
situational modeling and decompose the model into sub-models representing
certain operating points of the engine.
Opi = fi (N1,T4,T2, P2), i=1…n (2)
where n is the regimes count. This decomposition can be done by expert knowledge
or with use of clustering algorithms. We propose the decomposition of the model
into a set of three operating points (n=3), or in terms of situational modeling into
three distinct situations [8]. That is the startup of the engine, stable operation of the
engine and its shutdown. Every situational model is further decomposed into a set of
non equivalent sub-models, which are interconnected according to the basic physical
dependencies in the engine and are treated as black box systems. Every one of these
sub-models is then represented by a neural network or fuzzy inference system,
Fig. 3. Modular architecture of the dynamic engine model

90 R. Andoga et al.
which models the individual parameter dependencies and further decomposes the
operating points into local operating points which are then represented as local neural
or fuzzy models. Furthermore all models have to be put in an adaptive structure that
will be able to decide, which model to use for certain situational frame. The modular
architecture of such system is shown in the figure 3. A classifier in the form of neural
network represents the gate which gates outputs of individual models to give a correct
prediction. Use of this model allows us to simulate whole operation of the engine with
also highly non-linear atypical situations such as startup and shut down of the engine.
The model has only a single input parameter in the form of fuel flow input Qpal.
The inputs for the classifier neural network are state variables resulting from
the model, the only input to the model is the fuel flow parameter. The output of
the network will be defined as a vector:
Ou = [x1, … , xn] (3)
where n is the number of situational model frames and xi = {0;1},in our case n=3.
Further and more precise description of the model can be found in [8].
3.3 Diagnostic Models – Neural Network Model

In the process of training of the neural network that will act as the one of the
backup model the following input data has been used – fuel flow supply Qpal
shown in the figure 4 and as output the speed of the compressor n that was
obtained directly by measurement of the engine MPM-20/21. The model can also
be applied within the situational model of the engine in appropriate frame.
Fig. 4. The measured data of fuel flow

Fig. 5. The measured data of engine speed
Based on testing of different topologies, the best neural network topology for
modeling of the chosen dependency is a neural network with three hidden layers
with 30 neurons in the first, 20 neurons in the second and 25 neurons in the third
hidden layer. The neural network has an input window of delayed inputs of 4
neurons and a single output neuron. The mean absolute error (MAE) of the
network is MAE = 150,34 rpm and the maximum absolute error MAAE = 1467,19
rpm, that represents 0.4% and 3.1% of the operational parameters of the MPM-
20/21 engine. The figures 6 and 7 illustrate approximation of the measured
reference signal and the error of this approximation.
4 Intelligent Supervisory Control System

The intelligent supervisory system for small turbojet engine can be decomposed
into four basic parts.
• Pre-start diagnostic system
• Online diagnostic system
• Backup system
• Start-up control system
92 R. Andoga et al.
Fig. 6. The modeled engine speed and the measured engine speed
Fig. 7. Error of the neural network with topology 30-25

Fig. 8. The intelligent supervisory engine control system – a general scheme
All the systems operate with mathematical models of the engine as described in
the previous chapters. The following figure illustrates how is the intelligent
supervisory system designed on the highest level of situational control for the
small turbojet engine.
The individual parts of the intelligent supervisory system are described in the
following chapters.
5 Back-Up System
Its main role ([2], [3]) is fault tolerance, and this is achieved by using redundant
resources, which we call backup. In implementing the backup, there are two basic
relationships between the backed up and the backup element.
Static backup operates continuously with a system. The input and output of the
system are permanently connected. Its main advantage is the ability to mask errors
and impermeability of errors on the system’s output. On the other hand, its
disadvantage is the increased energy consumption and low mean time between
failures (because all elements are simultaneously worn out).
Dynamic backup starts operations in a system when the backup object gets to
the error state. It is activated by using a switch. With dynamic reserve a longer
mean time between failures and lower power consumption than with static reserve
is achieved. However, the disadvantage is that during switching occurs a
temporary loss of output signal. In this backup method three basic modes of
94 R. Andoga et al.
operation are distinguished (it dependents in proportion to the size of performance

in relation to backup and backed up device):
• unloaded - backup element - when it is not used - is completely
disconnected from power (zero power),
• lightweight - backup element is turned on only at reduced power in
the time, when the backed up element is working,
• loaded - represents the opposite extreme to the unloaded backup (the
backup works on 100% performance together with backed up
element).
Hybrid backup tries to connect the positives of the static and dynamic backup and
eliminate their weaknesses.
6 Diagnostic System
6.1 Diagnostic Systems in General

Diagnostics represents analysis of immediate state of the object ([2], [3], [4], [6]).
It consists of two phases- fault detection and fault localization.
Diagnostic system (DS)
It is a system designed and focused to evaluate the state of the object. It consists of
diagnostic tools (sensors, testers, diagnostic algorithms, modeling techniques, ...),
diagnosed system and operator. Diagnostic tools can be implemented as a part of the
object (internal) or separately (external resources). Division of diagnostic systems:
• ON-LINE DS - evaluates the continuous status of an object at work
(Fig. 9),
• OFF-LINE DS - diagnostics in this type of DS is most frequently
realized by testing when the object of diagnosing is off (Fig. 10).
Fig. 9. Online diagnostic system

Diagnostics has great importance for aircraft’s engines and aircraft, because if
error occurred during the flight, this fault must be detected enough early to make
the necessary actions for safe continue of the flight or landing. One of the primary
methods used in practice (for diagnostics as well as for backup) is the majority
method, which is based on the static backup (Fig. 11) and has the ability to capture
failures of input elements. Number of inputs fed into the evaluation element must
be odd to avoid the case where a number of false and correct signals are the same.
A more detailed description of the designed system with the majority method is
given in one of the following chapters.
Fig. 10. Offline diagnostic system
Fig. 11. The general scheme of the NMR structure (n- modular redundant)
6.2 Pre-start Diagnostics

Pre-start diagnostics is executed before the actual engine start-up procedure is
launched. That means that all auxiliary subsystems of the engine like fuel pumps,
96 R. Andoga et al.
oil pumps, ignition, electromagnetic valves are powered only after the pre-start
diagnostics has passed. In this case the default hydro-mechanical unit is replaced
by computer controlled servovalve LUN 6743. This represents the action element
and its control algorithms run in Matlab and LabView environment. The flowchart
is depicted in Fig. 12.
An advantage of this control approach is the fact that it is possible to change the
start signal and the whole control signal straight from the computer, which enables
us large possibilities when using different methods and processes of control. It is
also possible to change the start of each executive element, whether continuously
after a specific activity or it is possible to delay the time of particular element's
start. On the other hand questions about the sensors and action elements reliability
come to focus because their fail will result in feedback failure and the entire
system control mechanism. Here it is necessary to carry out error diagnostics and
backup of chosen measured parameters.
START
Start the palications

Turn on the
MATLAB and
main switch
LabVIEW
NO The engine
connection switch is
turned on?
YES
Turn on the spark Turn on the oil

Engine startup
plugs pump
NO The speed NO The speed

of compresor is over of compresor is over
8 000 RPM? 20 000 RPM?
YES
YES
Execute- the „start

Turn on
signal“/
the fuel
Execute- fuzzy Disconnect the Disconect the
pump
startup regulator electrical starter spark plugs
NO
Turn off the engine?
YES
Turn off the

main switch
STOP
Fig. 12. Electrical control for engine startup
To ensure an error free start and operation of the engine, there is a need to
check all systems and sensors functionality, before the engine start has begun.
This is why the prestart diagnostics is a crucial element in the engine diagnostics
process. For carrying the prestart diagnostics out a diagnostic expert system was
chosen as the best approach. An expert system is a computer program that keeps
the knowledge of an expert about the subject area in form of a knowledge base
and uses it for problem solving on an expert level.
The proposed scheme of a diagnostic expert system for the prestart diagnostics
of the MPM-20 is depicted in Fig. 13.
Fig. 13. The expert system used for pre-start diagnostics
The main goal of the proposed expert system is to execute a check of every
monitored motor parameters. Next, based on the previous check, it has to decide,
whether the monitored part of the measuring chain works correctly and so that
whether the engine can be started.
The operating principle can be summed up into two main points:
• Measured data from the MPM-20 are compared with the created
knowledge base using the control mechanism.
• The output of this process with conclusions (recommendations) for the
user, are depicted using the graphical interface.
The proposed knowledge base represented with table is depicted in Tab. 1.
Table 1. Rule base for pre-start diagnostics
Num IF CONDITION THEN CONCLUSION

power supply failure
1. UQpal ≠ <23, 25> V of the flow meter
Qpal
2. UP2C ≠ <20, 28> V of the pressure
sensor P2C
3. UP3C ≠ <20, 28> V of the pressure
sensor P3C
4 valid conditions 2. or 3. of the pressure
sensors
5. UFt ≠ <10, 12> V of the engine thrust
sensor Ft
6. Un ≠ <22, 24> V
of the speed meter n
7. UEZOP≠ <32, 27> V
( EZOP)
it is not possible
any of the conditions conditions
8. start the engine
<1 … 7> valid
MPM 20
it is possible start
9. all conditions <1 … 7> invalid
the engine MPM 20
98 R. Andoga et al.
Individual rules from the knowledge base are implemented in the LabVIEW
environment using mathematical blocks.
6.3 The Majority Method for Small Turbojet Engines

The designed system is an online diagnostic system based [2] on static method of
backup and used for checking the multiple backed up devices - to detect the failure
of one of them. In this case we confine ourselves only to a system with three
backups, but generally it can be n (odd number). In such a system we assume that
the probability of failure is bigger for one of them, usually two work correctly
(Fig. 14). Measuring element, whose task is to detect the faulted element is called
a quorum element.
On the input of the quorum element are input variables (I1, I2, I3), which can
be measured identically (the same measuring method of the value on all inputs), or
may represent signals obtained from the measurement of the variables by various
methods. We propose here the signals to be the results of the models (neural and
linear) described in the first chapter and optical sensor measurement that will be
the last signal.
Fig. 14. Adapted scheme of the quorum element
Subsequently, the inputs go through the block of pair comparison, where based
on defined permitted error residue εi, pov, their comparison and evaluation. The
output signal that exceeds the permissible error residues does not enter the next
block, where the average value – I (7) is calculated.
I i
I = i (4)
n
Where n – is a number of inputs, which do not exceed the permissible error
residues.
This is in the form of negative feedback supplied to the input of block qi. In
case, where more inputs fail in the pair comparison, the block signalizing total
breakdown of the system is activated, this is indicated by the corresponding output
V4 and the system can enter a failsafe state.
Block qi has a characteristics (5) in the form:
1, for : ε i ≤ ε , pov 
  (5)
qi = 
i

0, for : ε i > ε i , pov 
It serves as a member of the validation (Fig. 12), where its output is a logical
binary output (0,1). This output (Vi – from V1 to V3) directly indicates if the input
of quorum element represents measurement of error or not.
Fig. 15. Standard deviation of error of the dynamic model
Inputs to the system (quorum element) represented three different methods of

measurement of the operating speed MPM 20, allowed error residues (RPM) were
defined, based on empirical experience. For qi:
Optical sensor
1, for : ε i ≤ 2000 

q1 =   (6)
0, for : ε i > 2000 
100 R. Andoga et al.
Approximation of the operating speed based on the neural net
1, for : ε i ≤ 550 

q2 =   (7)
0, for : ε i > 550 
Calculation based on a linear model of turbojet engine
1, for : ε i ≤ 1000 

q3 =   (8)
0, for : ε i > 1000
For these values the table of the block of pair comparison was constructed, see the
following (Table 2).
Table 2. The table of the block comparison
Pair comparison of Allowed value of error

i and j inputs: (εi,pov+ εj,pov)[RPM]:
I 1 a I2 2550
I 1 a I3 3000
I 2 a I3 1550
Adapted scheme of quorum element for the proposed inputs is on the next
figure (Fig. 15).
Fig. 16. Standard deviation of error of the dynamic model

6.4 Simulation of the Voting Majority Method

The following figures illustrate operation of the implemented voting majority
method for the small turbojet engine with 3 different types of speed assessment.
The first one is done by means of optical sensor (OS), the second one is
represented by a neural network model (NM) and the last one is represented by
approximation with a linear dynamic model (LM). The figure 17 illustrates the
simulated errors of different types of speed assessments. Error of the neural model
occurs at the time of 45 seconds, error of the linear model occurs at the time of 59
seconds and error of the optical sensor occurs at 69 seconds. Figure 18 shows the
resulting simulation with running diagnostic/backup system. We can see that the
quorum element eliminates all errors that have considerable amplitude of around
5000 rpm. The system will undergo further evaluation and will also be evaluated
in real-world conditions on the small turbojet engine.
Fig. 17. Measured and calculated engine speed together with errors of inputs of the quorum
element
Fig. 18. Measured and calculated engine speed together with errors of inputs of the quorum
element
7 Startup Control
In design of control algorithms as elements of an integrated control circuit fuzzy

inference systems can be successfully used. Such system has been used to design a
startup controller for MPM 20 - the experimental small turbojet engine. This
controller is acting only by startup of the engine and its aim is to decrease the
temperature overshoot by startup that decreases life cycle of the engine and in
certain cases can lead to turbine engine damage. The present startup techniques
are mainly time based, what means that the fuel flow input is increased in a time
based function rather than parameter based. The control algorithm can be seen as
one of the controllers in the figure 8 and is bound with digitally controlled servo
vent for fuel supply control.
The basic idea is to decompose the startup process of the engine into model
micro-situational frames, where one rule of the inference system would
correspond to one micro-situational frame in the start-up macro-situational frame.
Each rule in the form of <if> … <then> … postulate has a corresponding output
value of fuel supply assigned. In this way we can handle not only the classical
situations at startup, but also emergency situations like stall of the engine or fire in
the engine at startup. Three inputs and one output were chosen for the rules so the
rule looks like this:
IF T4c IS L{T4c} AND dT4c IS L{dT4c} and n IS L{n} THEN Qpal IS L{Qpal} (9)
where
• T4c – temperature of gases behind the turbine
• dT4c – derivation of temperature of gases behind the turbine
• n – speed of the engine
• Qpal – fuel flow supply to the engine
• L – function that assigns fuzzy membership function in corresponding
universe of the afore mentioned parameters
In this way the fuzzy controller handles 60 micro-situations with a small excerpt
shown in the following table.
Where rules in yellow color represent critical temperature states, blue color
represent stall of the engine, red color represents rule handling fire of the engine
and green color represents the final rule after execution of which control is
handled to another controller and engine transits from startup macro-situation into
another macro situational frame. The last figure shows the control surface of the
startup macro-situation for MPM-20 engine in case of dT4c is zero.
Table 3. Knowledge base of the start-up controller
Fig. 19. The control surface of the fuzzy inference system startup controller
The pilot simulations have shown that the proposed intelligent supervisory control
system can be efficiently applied to the laboratory test engine. The system is
modular by its design and as a framework can incorporate different algorithms and
approaches that can be evaluated. As a backup/diagnostic system, it is usable in
eliminating faulty measurement or computation of the engine’s speed and is
usable in the control algorithm. Implementation of the neural model and other
methods of computing of different parameters will increase the safety of operation

of the engine. The system will also undergo real-world testing in laboratory
conditions with the MPM-20/21 engine. We also aim to apply further adaptive
techniques and filtering algorithms [10] into the modular intelligent supervisory
control system and test them in real-world conditions.
No. 1/0298/12 – “Digital control of complex systems with two degrees of freedom” and by
KEGA Grant Agency of Ministry of Education and Academy of Science of Slovak
Republic under Grant No.018TUKE-4/2012 “Progressive methods of education in the area
of control and modeling of complex systems object oriented on aircraft turbo-compressor
engines”. This support is very gratefully acknowledged.
References
[1] Harris, C., Hong, X., Gan, Q.: Adaptive Modeling, Estimation and Fusion from Data,
p. 323. Springer (2006) ISBN 3-540-42686-8
[2] Hlavička, J., Racek, S., Blažek, T.: Digital Fault tolerant systems, p. 330. ČVUT,
Praha (1992) (in Slovak language) ISBN 80-01-00852-5
[3] Jonathan, S.L., Turso, J.A., Shah, N., Sowers, T.S., Owen, T.S.: A Demonstration of
a Retrofit Architecture for Intelligent Control and Diagnostics of a Turbofan Engine.
NASA/TM -2005- 214019 (2005)
[4] Kelemen, M., Lazar, T., Klecun, R.: Ergatic systems and safety in aviation (Ergatické
systémy a bezpečnosť v letectve: edukácia a inteligencia zručností v leteckej prevádzke:
vysokoškolská učebnica), vol. 1, p. 316. Akadémia ozbrojených síl generála M.R.
Štefánika, vyd - Liptovský Mikuláš (2009) (CD-ROM) ISBN 978-80-8040-383-6
[5] Kreidl, M.: Diagnostics systems, p. 314. ČVUT, Praha (1997) (in Czech language)
ISBN 80-01-01670-6
[6] Lazar, T., et al.: Development tendencies and modeling of avionics systems, MoSR,
Bratislava, p. 160 (2000) (in Slovak language) ISBN 80-88842-26-3
[7] Linke-Diesenger, A.: Systems of commercial turbofan engines: an introduction to
systems functions. Springer (2008) ISBN 978-3-540-73618-9
[8] Madarász, L., Andoga, R., Főző, L.: Intelligent Technologies in Modeling and
Control of Turbojet Engines. In: Er, M.J. (ed.) New Trends in Technologies: Control,
Management, Computational Intelligence and Network Systems, Sciyo, pp. 17–38
(2010) ISBN: 978-953-307-213-5,
http://www.intechopen.com/articles/show/title/intelligent
-technologies-in-modeling-and-control-of-turbojet-engines
[9] Sanjay, G.: Fundamentals of Aircraft Engine Control Design Course, Lecture. NASA
Glenn Research Center (September 15, 2007)
[10] Várkonyi, T.A., Tar, J.K., Bitó, J.F., Rudas, I.J.: Simple Noise Reduction in the
Adaptive Synchronization of Coupled Neurons by Robust Fixed Point
Transformation. In: Proc. of the 15th IEEE International Conference on Intelligent
Engineering Systems (INES 2011), Poprad, Slovakia, June 23-25, pp. 297–302
(2011) ISBN: 978-1-4244-8954-1
Product Definition Using Knowledge in Human
Influence Representation
László Horváth and Imre J. Rudas
Óbuda University, John von Neumann Faculty of Informatics, Institute of Intelligent

Engineering Systems, Budapest, Hungary
Abstract. Joining to a recent trend for including human defined active knowledge
in product model to replace direct definition of product objects the authors of this
chapter defined a new method for the representation of human participation in
product model construction. This is one of the results authors achieved in the topic
of human request driven product model definition. The proposed model represen-
tation applies human request based active knowledge and assists handling infor-
mation for the lifecycle of product as a contribution to Product Lifecycle
Management (PLM) technology. Current PLM technology represents well-proven
engineering practice utilized at so many leading companies. Recognizing this situ-
ation, the authors proposed implementation of the proposed modeling in profes-
sional PLM environments using application programming interface tools. In this
chapter, product model space and contextual engineering process are introduced.
Outside affects on product definition, contextual chain for outside affects and con-
nections with actual product definition are explained. Because current product
modeling had not capabilities for description of human thinking process in the
product model, a new solution was conceptualized that use knowledge based con-
textual chain for product definition. The authors introduced extended model space
where extension spaces constitute a contextual chain and map content to product
features in model space.
Introduction
In recent years, great advances are stimulated in knowledge intensive Product Li-
fecycle Management (PLM) by industrial competition. High level and reliable
new products are required with short innovation cycle. By now, robust computer
modeling and virtual prototyping capabilities have been considered as essential re-
sources for successful engineering. Partial solutions such as CAD/CAM/CAE sys-
tems at the end of the past century are not suitable any more. Full PLM service is
necessary by using of a fully integrated model for the product. Demand for model-
ing system with capability of simple modification and development of product
model assumes advanced handling of relationships within the product model. At
the same time, unorganized nature of relationships in product model requires re-
presentation of knowledge for product feature definition within the model. This
knowledge is utilized at automatic definition and redefinition of product objects
for frequent changes of product and its variants.

106 L. Horváth and I.J. Rudas
The authors of this chapter analyze product modeling technology in industrial

PLM systems and develop new ideas, approaches, and methods to enhance know-
ledge based definition of engineering objects in the course of product definition
for ten years. They consider the above characterized robust and flexible engineer-
ing technology as starting point for their product model related research. This re-
search is supported by the tendency of feature and active knowledge based con-
struction of product model producing excellent surface for an urged new
generation of knowledge definition and application.
In the above context, product model is considered by the authors as it was de-
fined in the STEP (Standard for the Exchange of Product Model Data, ISO 10303)
product model standard. Despite basic results in STEP were elaborated during
eighties and nineties, they still apply for product modeling. Phrase engineering ob-
ject refers to full integration of objects demanded by different engineering areas in
product model during lifecycle of product [14].
Results of the work presented in this paper will be tested in the experimental
PLM system at the Laboratory of Intelligent Engineering Systems (LIES) of the
Institute of Intelligent Engineering Systems, John von Neumann Faculty of Infor-
matics, Óbuda University where leading industrial product lifecycle management
technology is available to verify the results.
This chapter is a rewritten and extended variant of the paper entitled Control of
Product Object Definition by Background of Human Influence [17] that was pre-
sented in the 15th IEEE International Conference on Intelligent Engineering Sys-
tems 2011, Poprad, Slovakia. It is organized as follows. It starts with a discussion
of preliminaries and basic considerations of the proposed modeling. Model space,
its main connections, and contextual engineering process are introduced. The next
part explains outside affects on product definition, contextual chain for outside af-
fects, and connection of outside world to product definition. Following this, it is
emphasized that current product modeling has no capabilities for description of
human thinking process in the product model and a new modeling method is pro-
posed as a contribution to solution for this problem. The chapter is continued by
an introduction to extended model space concept where extension spaces are ap-
plied to represent contextual chain of knowledge and for mapping knowledge con-
tent to product features in the model space. Finally, main connections of the
proposed modeling for its implementation in a PLM system are outlined.
1 Preliminary and Basic Considerations

Product modeling applies model space in which product representation is con-
structed. Current model space is a consequence of shape centered nature of engi-
neering. Model space in its current form is outlined in Fig. 1. Product is
constructed by features, including form, part, and even higher level features. Fea-
ture is an object that requires capability for the definition of its parameters. Rela-
tionships of feature parameters are defined by rule, reaction, equation, and other
features. Relationship features are increasingly knowledge representations. Results
are conserved by constraint features.
Product Definition Using Knowledge in Human Influence Representation 107
For the handling of model space, group of authorized humans applies functio-
nality for human control of feature definition. Handling procedures generate and
modify features according to direct human instruction or using human defined re-
lationship features. Connections with physical world are increasingly important
and exchange information with scans, sensors, and equipment controls.
Related recent publications concentrate on knowledge definition, feature prin-
ciple application, human and modeling procedure communication, feature parame-
ter definition, and PLM integration problems. In the following, first important ref-
erences are cited from published works of researchers then previous related
researches by the authors are briefed.
The STEP product model standard established standardized basics of feature
based product models. It applies generic and application oriented resources to de-
fine object oriented product model. Main areas of its application can implement it
by using of application protocols those are still under development for typical en-
gineering areas. Leading companies apply STEP principles and methods at devel-
opment and application of PLM systems [1].
During the last decade, leading PLM technology reached full feature applica-
tion in order to establish application oriented product definition. In [2], feature is
defined as a coalescence of parametric and semantics in order to support function-
oriented product modeling by handling geometry data on the basis of manipulated
expertise and assure the possibility of communication between users as well as in-
side the engineering modeling environment. Very complex products should be de-
scribed in product model including elements for embedded electric, electronic, and
computer functions. Mathematical model of a turbojet engine in [3] is an example
to include description of complex relationships in engineering utilizing methods
from computational intelligence. A typical extended feature principle in [4] is ap-
plied for reconfiguration features at runtime in case of software products. Authors
of [5] give an extended definition of life cycle features for product model by inte-
gration of the downstream activities in the design process. They define relation-
ships between product features for applications such as assembly, disassembly, re-
cycling, maintainability, and environment.
Product lifecycle management (PLM) is considered in the book [6] as a new pa-
radigm to manage products all the way across their lifecycles, in the most effective
way. That book outlines a wide area of engineering activities including simulation
that demands highly organized information system in order to support integrated
definition of product entities for lifecycle application. Author of [7] highlights
product lifecycle management (PLM) as perhaps the most important enterprise in-
formation technology application for supporting product and service innovation
and discusses critical issues and challenges around product engineering. He em-
phasizes problems caused by lack of standard engineering processes as a founda-
tion for PLM. In [8], functional behavior and structure model is introduced as the
engineering design conceptual space. In the function behavior and structure mod-
el, a design process schema is suggested as a way to understand the typical life
cycle of a decision-making process in design. All knowledge sources are involved
in this conceptual model.
In [9], knowledge acquisition processes are defined to capture structured

knowledge systematically, and knowledge representation technology is estab-
lished to store the knowledge, preserving important relationships. The objective
is to establish solution that is better than conventional engineering data bases.
Author of [10] discusses nature of expert performance in engineering and em-
phasizes that expertise is not simply a matter of possessing talent, but it is the
result of a dedicated application to a chosen field. In [11], issues at capture, re-
presentation and retrieval of design intent are discussed, a definition for design
intent is presented, and a context-based inference system is proposed to capture
design intent from product data. Design space is defined on corporation level
and design level.
Virtual space theory and methodology apply results from virtual reality sys-
tems. Paper [12] discusses impact of virtual reality on quality of collaboration
design processes. It is focused on collaboration scopes and requirements, partici-
pants’ behavior, and exploited interaction modalities. A method is applied to syn-
chronous and remote collaboration. Authors emphasize that this represents the
most critical communication in industry.
The proposed method is grounded during former works by the authors of this
chapter in modeling of engineering intent [13], definition of product behavior in
product model [14], knowledge representation in product model [15], and infor-
mation content based product modeling [16]. The authors of this chapter also ana-
lyzed change propagation to engineering objects. Definitions for change affect
zone and adaptive action was described in [18]. Modeling of human intent was
motivated by definite demand for modeling that makes decision considering actual
intent of several human possible. For that reason, intent representation should be
included in product model. Product behavior was considered as part of human in-
tent and correlation was revealed between product behavior and the related prod-
uct features. Knowledge representation is analyzed in the context of active defini-
tion of product features and content of product information is defined as
background knowledge for feature definition. Finally, direct preliminary of the re-
ported research by the authors is application of adaptive action to advanced active
knowledge based control of product features. According the concept by the au-
thors, analysis of change propagation in product model prepares definition of
adaptive actions. Executable status of adaptive action is the prerequisite to realize
the control of product features in the model space.
The contextual knowledge based and human request initiated product feature
definition in this chapter is based on the contextual engineering process in Fig. 2.
Two connected contextual chains are defined for the product model and the out-
side affects of product modeling. Although this chapter discusses only contextual
chains are defined for the product definition, the authors prepare their product
modeling method for future extension to contextual connections from outside
world.
Product Definition Using Kn
nowledge in Human Influence Representation 1009
Fig. 1. Model space and its main

m connections
Contextual chain for product model serves product definition. In this case, ouut-
side affects can be consid dered by using of parameters of product features and acc-
tive relations such as rulees and reactions. Features describing influencing humanns
are in CM1 contextual con nnection with features describing human intent. Produuct
behaviors are defined in thhe context of human intent (CM2).
Fig. 2. Contextual engineerin

ng process
Human intent includess modification request by the human and it is defined bby
requested engineering objjectives, product features, and methods for the definitioon
of product feature param meters. Decisions on engineering objects are done in thhe
context of product behav vior (CM3). Finally, generation of engineering objects inn-
cluding product and otherr features is done in the context of decisions (CM4).
Contextual chain for outside affects starts from environment of company.
Company activity featurees are defined in the context of this environment (CE11).
Project features are defineed in the context of company activity features (CE2). Thhe
contextual chain continues with context between project and engineering problem
solving (CE3) then engineering problem solving and product definition features
(CE4). Finally, the product model and the outside affects of product modeling con-
textual chains are in contextual connection. The authors of this paper proposed de-
finition of influencing human features in the context of product definition features
(CDC). This means that environment affects must be defined as human influences
together with definition of human intent. Because product definition can decide on
product objects directly, contextual connection CDE is also necessary.
2 Outside Affects on Product Definition
Outside affect features were placed within the twin contextual chain as it was
shown in Fig. 2. Fig 3 details groups of these features. Company environment acts
on company through CE1. It includes features for standards, law, and controls from
authorities. Outside professionals can define knowledge for proposed or mandato-
ry application. Here come marketing and customer demands related information
units in the modeling system. Company experts must be in the knowledge of
finance and economy related actual information.
It is obvious that a project for development and application of products works
in the context of company (CE2). Company level standard features act from here.
Also company level affects by strategy and decisions, budget, and measures.
Instructions modify product definition.
Problem solving at engineering modify product definition by direct definition
of task (CE3). Features at this level are defined for structure and activities, opera-
tions management, specifications and instructions, and problem to be solved. This
problem solving aspected definition of product development task supposes
application of engineering objectives at the definition of human request on product
definition.
Features in problem solving at engineering act on product definition features
through the contextual connection CE4 (Fig. 2). At the same time, contextual chain
for product model can be bypassed by CDE when product definition experts decide
direct product feature definition. The authors of this paper do not suggest utiliza-
tion of this possibility because it breaks the proposed decision system. Direct con-
text represents current practice of product definition.
The proposed contextual chain for product model is connected through CDC and
M4
C (Fig. 4). Above all, generation of engineering objects handles product fea-
tures, parameters, relationships, and constraints in the model space. As a conse-
quence of application the above method, direct interactions modify product or
knowledge when the proposed indirect interactions for modification product and
knowledge are not applied.
Fig. 3. Contextual chain for outside affects
3 Current and Prop

posed Methods
Problem solving at engiineering covers a special functionality that influencees
product definition. Autho orized humans who are working on product definitioon
think about object definittion both for product features and knowledge for produuct
feature definition.
Fig. 4. Actual product definiition
One of the actual challlenges in product modeling is that current systems havve
no capabilities for systemmatic description of this thinking process in the produuct
model. At the same time, it is important to know the way to a change request be-
cause coordination of req quests or revision of earlier decisions often needs it. Inn-
evitably, currently applied
d modeling methods allow for definition of knowledge iin
which some elements of human thinking process can be represented in produuct
model. However, engineeering objective, way to decision, purpose of modification,
allowed changes, and con ntextual chains can not be represented. Frequent modiffi-
cation of product model during
d its development needs systematic representation at
least main corner points ofo human thinking process in order to reconstruction huu-
man request for product definition
d during entire lifecycle of a product. High flexx-
ibility of virtual technology stimulates product definition through high number oof
changes.
As a consequence of th he above situation, a problem is generated by direct intee-
ractions for modification of product or knowledge features because definition oof
knowledge in current mod deling systems is restricted by slight representation capaa-
bilities. Some elements off human thinking process can be represented by using oof
features activated by chan nged knowledge or related parameters in product modeel.
However this is also considered as direct way of product model modification.
Fig. 5. Current product defin

nition in PLM
The method proposed byb the authors of this chapter is sketched in Fig. 6. Problem
solving at engineering is in contextual connection (CE4) with product definition.
Human thinking process is i communicated with the product modeling by using oof
human request recording functionality.
f This functionality is connected with produuct
ual connection CDC. Human thinking process on object de-
definition through contextu
finition is communicated d and sequence of elements, engineering objective, annd
intended influence are reccorded. These records answer questions what, why, annd
how engineer requests mo odification of the product model, respectively. Records arare
handled as requested but still inactive features in the product model.
Knowledge enhanced by b contextual chain for product model defines and conn-
neering objects through contextual connection CM4. Acc-
trols generation of engin
cording to main purposee of the proposed method, this control modifies activve
knowledge entities in thee model space. Direct control of product objects is stiill
possible according to meaasures in the workgroup for product modeling.
Fig. 6. The proposed producct definition in PLM
Contextual chain for product

p model is outlined in Fig. 7. Influencing human is
in connection with recordding human thinking process on object definition. The ree-
sulted record is applied at
a definition of knowledge enhanced by contextual chaiin
for product model. At thee meantime, this result is placed in the product model iin
g human (CM1).
the context of influencing
The next step in the coontextual chain for product model is collecting of objec-
tives and influences for engineering objects from actual requests (influence aat-
tempts) by different influeencing humans. Contextual connection CM2 is applied foor
this purpose. Definition of
o behaviors and situations from engineering objectivees
for engineering objects asssures engineering objective driven definition of produuct
objects. Contextual conneection CM3 serves communication between behavior deffi-
nition and decision on requested influences. Decision is supported by analysis oof
change propagation to eng gineering objects in change affect zone. In the meantim
me,
Fig. 7. Contextual chain for product model

adaptive actions are defin

ned for coordinated change requests. Generation of enggi-
hrough contextual connection CM4. The generated moddel
neering objects is done th
entities are considered ass temporary results and are often placed in the produuct
model as variants.
Fig. 8. Extended model spacce
4 Extending Modell Space Concept
The classical and proven solution for the representation of product information is
the model space. As it waas explained in the above sections of this chapter, the obb-
jective of the research intrroduced by the authors is extension of the knowledge re-
presentation capabilities in model space. For this purpose, extension spaces arre
applied to represent contextual chain for product model by mapping knowledge

content to product features in model space (Fig. 8).
Influence space represents influencing humans by mapping influence features
to product features. Influence features identify human with characteristics affect-
ing influencing and are in contextual connection CM1 with request features in the
intent space. Request features represent requests by humans in the form of engi-
neering objective and intended influence for elements of human thinking process
and are in contextual connection CM2 with behavior features in the behavior space.
Behavior features in behavior space represent product behaviors by situations
and sets of circumstances to define situations by relevant engineering object pa-
rameters. Behavior features are in contextual connection CM3 with decision fea-
tures in decision space. Decision features represent coordinated decisions and are
in contextual connection CM4 with engineering objects in model space.
Processing human requests is essential part of the proposed procedure (Fig 9).
Human requests generation of product or other features or gives objective for
definition of these features. The aim is increasingly objective based request. How-
ever, availability of procedures for objective based product feature definition is re-
stricted and this problem area requires additional research.
Both objective and feature request is defined for new or modified features be-
cause objective is also defined as model modification feature. The first step in
processing human request is to define consequences for the requested new or mod-
ified features. These consequences must be analyzed both for existing and decided
features. Requests are represented in the intent space for a product, a product
family or other group of engineering task configuration. Because typically
multiple humans submit requests for the same objective or product feature,
coordination of the requests is necessary. The method outlined in Fig. 9 assumes
coordinated requests. Coordination is discussed in [19].
Decision process applies two basic concepts. They are contextual chain and
change affect zone (CAZ) [16]. Contextual chain denotes a direction for propaga-
tion while CAZ integrates contextual chains for a coordinated request. By using of
these supports, decision space represents coordinated decisions in the form of
adaptive actions. Adaptive action is defined by process of coordination and ap-
proval of requests. Elaboration of coordination process constitutes one of the most
important future researches. Generation of engineering objects in model space is
done through contextual connection CM4 as it was explained above.
Adaptive action based product feature generation through contextual connec-
tion CM4 is outlined in Fig. 10. Model space includes product feature, parameter,
relationship, constraint and other feature definitions as it was introduced in Fig. 1.
Engineering object generation through contextual connection CM4 changes product
and relationship features for the modification of product elements and knowledge,
respectively. Product, and active rule and reaction features are defined in accor-
dance with status of change in adaptive action definition.
Fig. 9. Processing human req

quests
Fig. 10. Adaptive action baseed feature generation
5 Implementation in
i PLM Systems
As it was explained abov ve, the new knowledge based method for product definni-
tion is devoted to be extension to product model in currently applied PLM syys-
tems. The method considers awaited development and extension of these system ms
dapted to a more sophisticated model space and also it is
in the future. It can be ad
suitable to flexible develo
opment of product modeling especially to accommodaate
new methodological elem ments. It preserves absolute interactive nature of recennt
product modeling in ordeer to free definition of products on the basis of decisionns
made by experienced perrsonnel at industrial companies. However, recent trend is

including knowledge for product feature definition instead of direct definition oof
feature by its parameters.. The authors of this chapter joined to this trend as theey
explained and discussed above
a for key issues of their new method.
The main connectionss of the proposed extension at its implementation in a
PLM system are outlined d in Fig. 11. It is supposed that PLM functionality inn-
cludes software developm ment libraries in application programming interface (APII).
API serves extension by a gateway to communicate with PLM functions includinng
influencing human interfaace, group work management, product data managemennt,
product data representatioon, feature handling procedures, and model space mann-
agement. The proposed method
m needs only extensions to handling of new featurees.
For this reason, new featu
ure handling procedures, new feature definitions, and neew
elements for user interfacee are required.
Fig. 11. Implementation in PLM

P systems
Current product modelling is based on definition of engineering objects in modd-

el space as features for well-defined
w modification. In an object oriented modelinng
environment, feature handling procedures are mapped to feature objects. Interacc-
tive product definition su
urface is available for human at product definition by uus-
ing viewport to see into the model space and other interactive communication
tools. From outside programs, access is provided to the interactive product defini-
tion surface. Outside feature handling procedures are mapped to objects for new
entities in product model. In this way, the modeling method by the authors well
fits to the philosophy and structure of current industrial PLM systems.
6 Conclusion
A new method was introduced and characterized as a contribution to efforts in in-
cluding knowledge for the definition of product features in advanced product
models available in comprehensive PLM systems. The authors of the proposed
new method recognized that knowledge based solution must be defined for analy-
sis of coordinated requests for the modification of product model. Fortunately, the
feature principle has been extended to the full product model in current PLM
technology. The proposed extension is also full feature driven so that features in it
modify the same product model in a changed way.
In the proposed method, model space of current PLM systems is controlled by
new contextual engineering process. Contextual chain for outside affects on prod-
uct definition is also important to consider. The authors prepared this contextual
chain for the connection with the proposed extensions by future research. Current
product modeling systems have no capabilities for systematic description of hu-
man thinking process. This capability is intended to produce by the proposed con-
textual chain serving a new way of product definition. Extended model space of-
fers extension spaces to represent contextual chain for product model by mapping
knowledge content to product features in model space. Future research includes is-
sues for new methods, processes, and feature types. These issues were mentioned
at above relevant discussions in this chapter.
Acknowledgments. The authors gratefully acknowledge the grant provided by the KTIA-
OTKA Fund for Research of the Hungarian Government. Project number is K 68029. Fi-
nancial support from the Óbuda University is also acknowledged. The authors also ac-
knowledge the grant for Research Groups from the Óbuda University.
References
[1] Jardim-Goncalves, R., Figay, N., Steiger-Garcao, A.: Enabling interoperability of
STEP Application Protocols at meta-data and knowledge level. International Journal
of Technology Management 36(4), 402–421 (2006)
[2] Bronsvoort, W.F., Bidarra, R., Nyirenda, P.J.: Developments in Feature Modelling.
Computer-Aided Design and Applications 5(3), 655–664 (2006)
[3] Főző, L., Andoga, R., Madarász, L.: Mathematical Model of a Small Turbojet Engine
MPM-20. In: Rudas, I.J., Fodor, J., Kacprzyk, J. (eds.) Computational Intelligence in
Engineering. SCI, vol. 313, pp. 313–322. Springer, Heidelberg (2010)
[4] Lee, J.J., Muthig, D.: Feature-Oriented Analysis and Specification of Dynamic Prod-
uct Reconfiguration. In: Mei, H. (ed.) ICSR 2008. LNCS, vol. 5030, pp. 154–165.
Springer, Heidelberg (2008)
[5] Sya, M., Masclea, C.: Product design analysis based on life cycle features. Journal of
Engineering Design 22(6), 387–406 (2011)
[6] Stark, J.: Product Lifecycle Management: 21st Century Paradigm for Product Realisa-
tion, p. 441. Birkhäuser (2004)
[7] Hewett: A Product Lifecycle Management (PLM): Critical Issues and Challenges in
Implementation. Information Technology and Product Development, Annals of In-
formation Systems 5(Part 1), 81–105 (2009)
[8] Colomboa, G., Moscaa, A., Sartori, F.: Towards the design of intelligent CAD sys-
tems: An ontological approach. Advanced Engineering Informatics 21(2), 153–168
(2007)
[9] Preece, A., Flett, A., Sleeman, D., Curry, D., Meany, N., Perry, P.: Better knowledge
management through knowledge engineering. IEEE, Intelligent Systems 16(1), 36–43
(2005)
[10] Cross, N.: Expertise in design: an overview. Design Studies 25(5), 427–441 (2004)
[11] Iyer, G.R., Mills, J.J., Barber, S., Devarajan, V., Maitra, S.: Using a Context-based
Inference Approach to Capture Design Intent from Legacy CAD. Computer-Aided
Design & Applications 3(1-4), 269–278 (2006)
[12] Germani, M., Mengoni, M., Peruzzini, M.: Method for Evaluating VR-based Tools
for Collaborative Design. In: New World Situation: New Directions in Concurrent
Engineering, pp. 451–464 (2010)
[13] Horváth, L., Rudas, I.J.: Intelligent Human-Computer Communication of Engineers
at Extended Companies. Journal of Advanced Computational Intelligence and Intelli-
gent Informatics 10(4), 510–516 (2006)
[14] Horváth, L., Rudas, I.J.: Adaptive Modeling for Robot Systems Based on Behaviors
and Feature Driven Shape Descriptions. WSEAS Transactions on Information
Science and Applications, 1761–1771 (2005)
[15] Horváth, L., Rudas, I.J.: Knowledge Technology for Product Modeling. In: Know-
ledge in Context - Few Faces of the Knowledge Society, ch. 5, pp. 113–137. Walters
Kluwer (2010)
[16] Horváth, L., Rudas, I.J.: Engineering Objective-Driven Product Lifecycle Manage-
ment with Enhanced Human Control. In: Rudas, I.J., Fodor, J., Kacprzyk, J. (eds.)
Towards Intelligent Engineering and Information Technology. SCI, vol. 243,
pp. 543–556. Springer, Heidelberg (2009)
[17] Horváth, L., Rudas, I.J.: Control of Product Object Definition by Background of Hu-
man Influence. In: Proc. of the INES 2011. 15th IEEE International Conference on
Intelligent Engineering Systems, Poprad, Slovakia, pp. 45–50 (2011)
[18] Horváth, L.: New Design Objective and Human Intent Based Management of
Changes for Product Modeling. Acta Polytechnica Hungarica 4(1), 17–30 (2007)
[19] Horváth, L., Rudas, I.J.: Knowledge Engineering for Modeling and Simulation in Vir-
tual Product Development. In: Proc. of the 5th International Symposium on Computa-
tional Intelligence and Intelligent Informatics, Floriana, Malta, pp. 111–116 (2011)
Usage Dependent Rehabilitation and
Maintenance Model for Special Engineering
Structures
Kornélia Ambrus-Somogyi1 and András Bakó2

1
Óbuda University, Rejtő Sándor Faculty, Institute of Media Technology
and Light Industry, Budapest, Hungary
a_somogyi.kornelia@rkk.uni-obuda.hu
2
Óbuda University, Neumann János Faculty, Budapest, Hungary
bako@uni-obuda.hu
Abstract. An engineering structures rehabilitation and maintenance model is pre-

sented. The problem is to give the maintenance and rehabilitation action for each
year of the planning period. There are given several conditions to fulfill. Two
objectives should be presented: the system optimum is to minimize the total action
cost, the user optimum is to maximize the user benefit. The problem is to deter-
mine such an optimum which fulfill the given conditions and either user or system
optimum.
There are several models to solve this problem: linear or integer programming,
heuristic, stochastic etc. In the paper two models are presented. The first is a sto-
chastic programming model which is formulated for maintenance and rehabilita-
tion of a road network. The other model is a linear programming which was
formulated for other engineering structures.
Keywords: M+R management, linear programming model, Markov deterioration

matrix, stochastic model.
1 Introduction
With the recognition that the engineering structures maintenance and repair needs
far exceed the resources available to address these needs, many firms have to turn
to the development of ESMS (Engineering Structures Management System). The
ESMS is a method to improve the allocation of these limited resources and the
condition of their engineering structures. The ESMS is based on performance
modeling because of without the actual condition of the structures and the future
deterioration process any model could not be build up.
In its basic term the ESMS refers to the careful allocation of funds available for
these purposes: maintenance, repair and rehabilitation to ensure that the funds are
used in the most effective way. Specifically an ESMS is a rational and systematic
approach to organizing and carrying out the activity related to planning, designing,
constructing and replacing structures.

126 K. Ambrus-Somogyi and A. Bakó
The one part of the total ESMS is a computer program. The software associated
with an ESMS should provide the following functions:
- A database which contains the necessary information needed for this
purposes including inventory and inspection data, and information re-
lated to maintenance, repair, and rehabilitation actions and effective-
ness. This base contains historical data (deterioration past and future
maintenance and rehabilitation actions, cost, etc.).
- A mayor maintenance, rehabilitation and replacement component,
which contains the actions and its prices, etc.
- Heuristic or optimization procedure which gives the cheapest main-
tenance and rehabilitation actions.
- Deterioration model which determine the future condition state of an
engineering structure depending on the actual condition state and the
time period.
Several models should be built up to solve the problem. The most simple is the
heuristic where instead of the optimal solution feasible solution is determined.
This is not the cheapest and the best solution, but this “dirty” algorithm gives a
solution in short time. These models are usually a ranking one, where the possible
solutions are ranked using some idea.
Other solution is based on an integer (or 0, 1) programming. In this case the
maintenance and rehabilitation actions are multiplied by an unknown vector ele-
ment, which is a decision variable. (in the case of 0, 1 programming). It is difficult
to use this model in the case of large problems because the size of the model.
Other possible model is based on probability. Such a simple and usable model
is suggested by Prekopa [6]. This model solves the Pavement Management Prob-
lem. (see Bako[5]). This method is summarized in the next chapter. It is easy to
see, that this model is usable in the case of other engineering structures.
2 Stochastic Model
A stochastic programming model is presented in this chapter. Let us denoted the
homogenous set of road sections by i, i=1, 2,…, I. The cardinality of the set i is ni
and the sections are denoted by j, j=1, 2, , ni Each section could be in Ni different
conditions. The road condition index is k, k=1, 2,…, Ni. The area of the i-th group
and j-th road section is Tij. In the beginning of the t-th year the road condition of
the sections are determined. Let us denoted this condition by X ijk( t ) which has only
two value
(t ) = 1, when the set j - th section of set i in the t period is in k condition, (2.1)
X ijk 
 0, otherwise
In the set i the total area of the sections being condition states k is
ni
 X ijk(t )Tij (2.2)

j =1
Usage Dependent Rehabilitation and Maintenance Model 127
Let be a decision variable (t )

x ijhk which belongs to set i condition state k time pe-
riod t section j. x ijhk
(t )
= 1, when in the section being condition state k such a main-
tenance action is applied, that his condition will be h, and 0 otherwise.
These decision variables are applied only h=k, k+1,… Ni because each main-
tenance action will result a better condition than the earlier one.
(t )
The action cost is denoted by cijhk , h=k, k+1,…, Ni
The change of the condition during the actual year is modeled by the probabili-
ty variable Yijh(t + 1 ) .
1, if the conditiondo not changeafter theplanningperiodof the section

(t +1) =  j whichis in conditionh after theactualmaintenance, (2.3)
Yijh 
0, otherwise

Other possible condition is to order that the area of the road sections after a period
has to stay in the same condition with a big probability, and should not be re-
duced.
 ni ni 
P
  X ijk(t )Tij ≤  Yijh(t +1)Tij xijhk
(t )  ≥ p
 k (2.4)
 j =1 j =1 h ≥ k 
If such a solution is accepted which fulfill the given conditions with minimal cost,
the following problem has to be solved.
I N i ni
(t ) x (t )
min  cijhk ijhk (2.5)
i =1 k =1 j =1 h ≥ k
Suppose that
 ni ( ) ni 
(t +1)T x (t )  ≥ p , for all i and k
P  X ijk
t
Tij ≤ Yijh
 j =1 ij ijhk
 k
 j =1 h≥k  (2.6)
(t ) = 1, for all i and j
 xijhk
h≥k
3 Optimization Model
Earlier version of that algorithm was given by Ambrus-Somogyi at all [1,2]. The
stochastic model is formulated in the case of a transportation firm, where there are
several types of vehicles and several numbers of cars are in each type(see [3]).
In our model the following denotation will be used
- The type of the engineering system(cars): tip (tip1, …, tipm, …, tipM) –
for example Volvo autobus, etc., m is the type index, M is the number
of types.
- The number of the car types m is denoted by dbm

- The car type m is built up by N different peaces, m=1, 2, M.
- The number of different parts of the certain cars signs N. We can sup-
pose that the number of parts at the car types is equal. Here we see the
parts no elementary parts (screw,…), but bigger, common repairable
unit (brakes, steering-gear, etc.). Signe rmn the number of n-th parts of
m-th types car. These means, we have in all db m ⋅ rmn piece such part.
- The parts can be in different condition states. The condition of a part
is described by different types of deterioration parameters (for exam-
ple visual characterization or result of instrumental monitoring, etc,).
Let S be the number of condition states parameters. Mark 1 denotes
the best, and mark 4 denotes the worst condition: perfect – mark 1,
fault, does not disturb the normal use – mark 2, fault, disturbs the
normal use – mark 3, useless – mark 4.
- The altering of the condition, the deterioration depends on the other
factors – for example efficiency, transported amount – too. Denotes
the number of different values of theses F, the adequate index let be f.
- The possible maintenance types are denoted with vector p
( p1 ,, pk ,, p K ) – k is the index, K is the number of possible main-
tenance operations. The intervention can be simple repairing, the
change of a smaller unit, the change of a bigger unit or the change the
car part.)
- In case of the multiperiod algorithm the index of year is t, the number
of years is T.
The size of the vectors and matrices depend on the number of parameters and the
different values of parameters. If the number of parameters is S and the number of
different notes is 4, then the number of possible stage is S 4 , for example if S = 4
than 44=256. It means, the size of vectors is 256, the number of elements of matrix
is 256⋅ 256 . Denote the size of matrix L.
This number determines the size of the unknown variable vector Z, the element
of this vector denote proportion. The number of Z vectors is M ⋅ N ⋅ F ⋅ K ⋅ T .
The l-th element of vector Zmnfkt shows the car parts belonging to m, n, f, k, t
indexes and is in the l-th condition, that how many percent of this part the pk main-
l
tenance must be realized. This element is zmnfkt, or ( Z mnfkt ) l . We use an upper
index, if we refer an element of vectors or matrices.
In the model a Markov transition probability matrix is used to determine the de-
terioration. The element i, j of that matrix means the probability, that after a plan-
ning period this condition of this element does not change, the element i, j+1
means that the condition state deteriorates to some lower level.
To create the matrix first we divide the possible conditions into discrete condi-
tion states. The condition state can be divided for example 10 stages 10%, 20%,
30%, …100%. At any point in time probabilities are given for the likelihood of the
structure being in each condition states and there are defined in a “transition
matrix”. This matrix is used to predict the condition state after a time period (year,
2 years or more years).
Let us denote the Markov transition probability matrix by Qmnfk which belongs
to n-th part of the m-th type car, the f-th running efficiency and the k-th mainten-
ance action. The number of different matrices is M ⋅ N ⋅ F ⋅ K , the number of rows
and columns of matrices is the number of quality condition stages, L. The l-th
il
element of i-th row of Qmnfk matrix is qmnfk , or ( Q mnfk ) il gives the probability that
if the car parts which at the starting time is in i-th stage, at the end of the planning
period get into the l-th condition stage.
Let us denote the unknown vector Vmnft, that gives the fraction of those engi-
neering systems(car parts) which belongs to the n-th parts of the m-th type, to the
f-th running efficiency at the end the t-th time period.
Let us denote the vector bmnf, which gives the fraction of the different condition
stages of the n-th parts of the m-th type, belong to the f-th running efficiency at the
starting time of planning.
There are several conditions to fulfill. The first condition (3.1) is related to the
fraction of the engineering systems (car parts) at the initial year:
K
 UZ mnfk1 = b mnf , m = 1, 2,  , M n = 1, 2,  , N f = 1,  , F (3.1)
k =1
where U L⋅ L size unit-matrix. We must choose such a Z vector in first year,

which gives the starting bmnf vector in case of all car types, all parts and all run-
ning efficiency.
The second condition defines the vector Vmnf1 (3.2), the fraction of structures,
of the car parts at the end of the first planning:
K
 Q mnfk Z mnfk1 = Vmnf1 , m = 1,  , M n = 1, , N f = 1,  , F (3.2)
k =1
The next condition (3.3) applies to the mediate years. This means, that the Vmnft,
the fractions at the end of t-th time period gives the starting distribution for the
(t+1)-th period. For each year the following conditions must be fulfilled:
M N
 UZ mnfk (t +1) − Vmnft = 0, f = 1, , F k = 1, , K t = 1, , T − 1 (3.3)
m =1 n =1
This condition defines the unknown Vmnft vector.

One of the maintenance policies has to be applied (3.4) for every structure (car
parts) in each year:
M N F K
    Z mnfkt = 1, t = 1,  , T . (3.4)
m =1 n =1 f =1 k =1
The car parts are divided into 3 groups: acceptable (good), unacceptable (bad) and
the rest. Let us denote the three set by G the good, R the bad and E the set of other
structures and by H the whole set of structures. The following relations are realiz-
ing to the sets:
G∩R =∅ G∩E =∅
R∩E =∅ (3.5)
G∪R∪E = H
The following conditions (3.6) are related to these sets in the initial year
M N F K M N F
  (Q
m=1 n=1 f =1 k =1
mnfk Zmnfk1 )l ≥ α1  (bmnf )l , l ∈ G
m=1 n=1 f =1
M N F K M N F
 (Q
m=1 n=1 f =1 k =1
mnfk Zmnfk1 )l ≤ α 2  (bmnf )l , l ∈ R ,
m=1 n=1 f =1
(3.6)
M N F K
(b E )l ≤  (Qmnfk Zmnfk1 )l ≤ (b E )l , l ∈ E
m=1 n=1 f =1 k =1
where
- G, R, E are given above,
M N F
- (b
m=1 n=1 f =1
mnf )l , l ∈G the fraction of structures in the good set be-
fore the planning period,

M N F K
-     (Q mnfk Z mnfk1 ) l , l ∈ G the fraction of structures in the
m =1 n =1 f =1 k =1
good set after the first year,

M N F
-   (b mnf ) l , l ∈ R the fraction of structures in the bad set before
m =1 n =1 f =1
the planning period,,

M N F K
- (Q
m=1 n=1 f =1 k =1
mnfk mnfk1 Z )l , l ∈ R the fraction of structures in the
bad set after the first year

M N F K
-     (Q mnfk Z mnfk1 ) l , l ∈ E the fraction of structures in the
m =1 n =1 f =1 k =1
other set after the first year,

- b E the lower bound vector for the fraction of structures in the other
set,
- b E the upper bound vector for the fraction of structures in the other
set,
- α 1 and α 2 given constants.
The first condition means that amount of the car parts in the “Good” set must be
more or equal than a given value, in this case this is proportional with the starting
quantity. The second condition does not allow that after first year the amount of
the car parts in the “Bad” set can be more than a certain percent of the starting
amount. The third condition gives lower and upper bound for the other car parts
after first year.
For the further years similar inequalities (3.7) could be used
I J I J
Yijt R
i =1 j =1
Yi =1 j =1
ij(t +1) , t = 1,2,,T −1 (3.7)
where R could be one of the relations <, >, =, <=, >= and these relations could be
given in connection with each condition states (e.g. each rows could have different
relations).
Instead of (3.6) and (3.7) condition states could be applied for the end (3.8) of
the planning period (e.g. for t=T) :
M N F K I J
 (QmnfkZmnfkT)l ≥ α1 (bmnf )l , l ∈G
m=1 n =1 f =1 k =1 i =1 j =1
M N F K I J
 (QmnfkZmnfkT)l ≥ α2 (bmnf )l , l ∈R (3.8)
m=1 n =1 f =1 k =1 i =1 j =1
M N F K
(b E )l ≤  (QmnfkZmnfkT)l ≤ (b E )l , l ∈ E
m=1 n =1 f =1 k =1
Beside the condition for the states at all maintenance exercise is very important
the cost factor. As far as possible we must to work such a maintenance strategy,
which fulfills the conditions for the states and it has the lowest cost.
Let us denote by vector Cmnfk the unit cost vector of the maintenance policy k,
belongs to the n-th part of the m-th type car in case of f-th running efficiency. The
elements of vectors show that in case of certain qualification state how much
would it cost to make a unit of the maintenance action.
We can formulate more different conditions in connection with costs. One of
them is the yearly budget bound of each maintenance action:
M N F
   r ( t −1) C mnfk Z mnfkt = r ( t −1) M k , t = 1,  , T k = 1,  , K (3.9)
m =1 n =1 f =1
where r is the interest rate, Cmnfk is the unit cost vector of the maintenance policy
k, of the n-th part of m-th car type, belong to f-th running efficiency and Mk is the
budget bound available for maintenance policy k in the initial year.
Now the objective of the problem is formalized. The objective (16) is to minim-
ize the total cost of maintenance:
M N F K T
C = ZmnfktCmnfk → MIN ! (3.10)
m=1 n=1 f =1 k =1 t =1
If the yearly available B sum is given, then we can formulate two further condi-
tions in connection with the budget bound.
The budget bound condition for the initial year:
M N F K
Z
m=1 n=1 f =1 k =1
C
mnfk1 mnfk ≤ B. (3.11)
The conditions for further t = 2 , 3,  , T years are the following (3.12):

M N F K
r
m=1 n=1 f =1 k =1
(t −1)
ZmnfktCmnfk ≤ r (t−1) B . (3.12)
Besides the minimization of the maintenance costs we could aim to minimize the
user’s costs. The user’s costs depend on the type of car, on the car parts and on the
running efficiency. Let us denote this user’s cost-vector by Kmnf. The l-th coordi-
nate of the vector belongs to the l-th qualification state. ( 1 ≤ l ≤ L , in our exam-
ple it is 256).
The objective function of the minimization of users costs is the following
(3.13):
M N F T
C = VmnftKmnf → MIN ! (3.13)
m=1 n=1 f =1 t =1
Often it is practical to handle the two type of costs together and to write a com-
bined objective function. We can do this the following way (3.14):
M N F K T M N F T
C = α      Z mnfkt C mnfk + β     Vmnft K mnf → MIN (3.15)
m =1 n =1 f =1 k =1 t =1 m =1 n =1 f =1 t =1
Here we optimize in case α = 0 only the user’s costs, in case β = 0 only the
maintenance costs. So we can decide arbitrary objective functions with the differ-
ent value of two parameters. This model was demonstrated in [4].
4 Conclusion
The national economy importance of successful and efficient management of val-
uable road assets is evident. The relevant management decisions can be more ef-
fective in the long-term if reliable, accurate background information are based
upon. The pavement management models forecast the future road pavement per-
formance influenced, among others, by the actual traffic load (especially the num-
ber of heavy axle loads passed). The suggested advanced model considers the
usage of elements in order to take into account the more accurate deterioration
parameters every year of investigation period (planning horizon). As a conse-
quence, the managerial decisions can be more efficient and cost-effective contri-
buting to the better quality of the structures and reducing the life cycle costs.
References
[1] Ambrus-Somogyi, K., Bakó, A.: Performance Modeling of Structures. In: Proceedings
of IN-TECH-ED 2005 Conference BMF, Budapest, September 7-9, pp. 389–398
(2005)
[2] Ambrus-Somogyi, K.: Deterioration-based Maintenance Management Algorithm. Acta
Polytechnica Hungarica 4(1), 119–126 (2007)
[3] Ambrusné Somogyi, K.: Útburkolat-gazdálkodás változó feltételek melletti optima-
lizációs modelljei, Doktori (Ph.D.) értekezés, Széchenyi Egyetem (2009)
[4] Ambrus-Somogyi, K., Bakó, A.: Optimization model of maintenance of engineering
structures. In: Proceedings of 15th IEEE International Conference on Intelligent Engi-
neering Systems 2011, INES 2011, Poprad, June 23-25, pp. 73–76 (2011)
[5] Bakó, A., Szüts, I.: Optimal Road Maintenance and Rehabilitation Algorithm. In:
Proceedings of the International Conference on Pavement Engineering, Bhubaneswar,
India, pp. 74–80 (2009)
[6] Prekopa, A.: Personal communication
Course and Engineering Model Integration
for Teaching Automation
József Gáti1, Gyula Kártyás1, and Marián Bučko

2
1
Óbuda University, Budapest, Hungary
2
Technical University of Košice, Slovak Republic
Abstract. By changing environment of engineering activities to extensive and

highly integrated models in unlimited computer systems, more and more expert
consider integration of product modeling, higher education, and industrial practice
in a common environment in which recent shortage of problem solving based
education of engineers can be decreased. Joining to the efforts in order to achieve
this, the authors of this chapter revised and applied their earlier concepts and ap-
proaches and prepared a new approach to the above integration. This chapter starts
with an opinion for the future of integration of engineering higher education
processes and practice. Relevant works are cited to introduce important concepts
and approaches. Following this, the three computer system based technologies are
proposed to integrate in advanced higher education courses, communication of
processes in the proposed integrated modeling is discussed, and computer model
description of higher education system including information for the related areas
is explained. Finally, Possibility for implementation in teaching-learning environ-
ment, where understanding of real industrial problems motivates problem solving
is outlined as future plan.
Introduction
Role of computer systems based engineering increases with increasing demands
for highly engineered products, using less industrial resources, and achieving
shortened innovation cycles. Recently, enhanced industrial engineering technolo-
gy utilizes recent achievements in information technology including advanced
shape and other modeling techniques, Internet technology, virtual operation sys-
tems, cloud computing, and so on. Recent technology of highly integrated systems
offers new chance for the establishment of a new integration in engineering in
order to solve the problem of engineering communication shortage in higher edu-
cation systems. This shortage breaks realization of problem solving based engi-
neering education processes because the necessary methodology and information
communication tools are not available within higher education institutes.
In their earlier works, the authors of this chapter analyzed higher education
course models in order to apply them in higher education systems and programs
those have powerful connection with real industrial systems. This connection is
essential for the future because real problems are emerged in industrial environments

136 J. Gáti, G. Kártyás, and M. Bučko
and engineering practice related, information technology based problem solving

practice exists only there. In the scope of this effort, the authors of this chapter
analyzed solution variants for the integration of course modeling with well-proven
product modeling as a contribution to establishing future courses where increasing
part of education programs will be realized in engineering model space. They
widely published the results of their related earlier works [9-11].
In this chapter, a new concept of and approach to integration of higher educa-
tion, industrial engineering, and industrial automation is outlined. The authors
would like to contribute to current trends and efforts in this area by revision and
adaption of their earlier publisher results. The chapter is organized as follows.
Relevant works are cited to introduce important concepts and approaches. Follow-
ing this, three computer system based technologies are proposed to integrate in
higher education courses. Communication of processes in the proposed integrated
modeling is discussed. Next, description of higher education system including
information for the relevant areas is explained. Finally, implementation in a teach-
ing-learning environment is outlined as future plan. Understanding of real indus-
trial problem solving motivates this work.
1 Earlier Works and Preliminary Considerations
Engineering is inherently problem solving intensive. Because engineering model-

ing concentrates on problem solving, one of the main efforts in this area is include
problem solving in higher education processes. Conventional curriculum is topic
oriented. Problem solving based higher education needs integrated handling of a
relevant set of topics. Earlier separated subject matters must be integrated while
learning of them must be maintained in an enhanced level. The authors of this
chapter think this as a key problem of engineering education development for the
future.
In the following, relevant works are cited to introduce relevant concepts and
approaches. These works emphasize problematic of integrated model with six
dimensions, current limited formal academic use of information and communica-
tion technology, importance of computer-supported collaborative learning, charac-
teristics of problem-based learning environment, and the present situation with
lack of practical knowledge required for problem solving
A study is concluded in [1] using integrated model with six dimensions. These
dimensions are learners, instructors, courses, technology, design, and environ-
ment. Critical factors affecting learners’ satisfaction in computer assisted learning
are surveyed. The reported results revealed that learner computer anxiety, instruc-
tor attitude toward e-Learning, e-Learning course flexibility, e-Learning course
quality, perceived usefulness, perceived ease of use, and diversity in assessments
are the critical factors affecting learners’ perceived satisfaction. It is a critical
demand for course program, that student understands the teaching program, the
competences included, and the learning outcomes. Real world examples from the
industrial practice assist to achieve this.
Course and Engineering Model Integration for Teaching Automation 137
According to author of [2], many university students and faculties make only
limited formal academic use of information and communication technology. It is
emphasized that computer technology use is constructed in limited, linear, and
rigid terms far removed from the creative, productive, and empowering uses. Pa-
per [3] is a report about a study to investigate the effects of visualization of partic-
ipation during computer-supported collaborative learning. The results demonstrate
that visualization of participation can contribute to successful computer-supported
collaborative learning.
In [4] it is emphasized that problem-based learning is in the core of significant
developments in engineering education. In a problem-based learning environment
the problem drives the learning. Students realize they need to acquire new know-
ledge before the problem can be solved. Authors of [4] show the experience of a
structured problem-based learning approach to the teaching of an introductory
study module on heat transfer in a first year graduate program.
Authors of [5] state that most e-learning platforms offer theoretical knowledge
content but not practical knowledge required for problem solving. The paper pro-
poses a problem-based e-learning model which incorporates the problem-based
learning theory, social constructivism, and situated learning theories to assist regu-
lar and special education teachers in effectively developing knowledge for ma-
thematics teaching for students. Authors applied clustering and information re-
trieval techniques to construct context and content maps for case-based reasoning
with the capability of semantics identification.
One of the most recent areas in engineering covers integrated application of in-
telligent technologies for the purpose of process modeling and equipment control.
Paper [6] introduces an application that includes methodology that is also impor-
tant for the concept and approach proposed in this chapter.
Development of information technology and its accessibility for every day
teaching work quickened development of advanced computational method and
model based teaching environments, programs, and processes. A model based
classroom system is introduced in [7] and [8]. In [9], a course modeling is ex-
plained in which definition of information in the product model was controlled by
human intent and other human intent based content entities. Methods were intro-
duced for definition of human control on product model information, modeling of
human intent, and human intent based definition of human decision background at
problem solving.
In their earlier and present works, the authors of this chapter studied the model
in [7] and [8], revealed practical issues, and developed methods as published in
[10] and [11], and this paper. In [7], issues are discussed regarding a feature based
virtual university model including systematic description of course information,
handling of relationships, and flexible creation of student profile based course.
Well-proven principles from product modeling were applied at modeling of virtual
course. Resource based aspect of modeling was emphasized. Entities were defined
as generic, domain, and practice resources for course models. Paper [8] introduces
some virtual classroom related concepts and an approach to virtual university.
Feature driven associative model of virtual course as developed by the authors is

explained. Application oriented virtual course features are discussed.
In one of their previous works, the authors of this chapter analyzed and dis-
cussed cooperating managers and managing techniques for virtual classrooms in
[10]. They published practice oriented concepts concerning classroom objects and
processes considering higher education practice. They also discussed virtual teach-
ing procedures in Internet environment, local and global views of virtual environ-
ments, and main advantages of virtual education. They surveyed an earlier
published structure and construction of classroom model and possibilities for the
integration of modeled classrooms with engineering systems as it was mentioned
above. They also recognized that one of the most important concerns towards a
better organized course description is definition of organized activities for course
handling systems [11]. They analyzed the definition and application of organizing
activities for the control of course definition and course program execution
processes. Typical processes were discussed including sequences of activities in
the course of teaching and learning in a virtual classroom environment. In [12], a
novel approach was proposed to integration of industrial problem solving in high-
er education courses for teaching of engineers.
2 Equipment, Control, and Course Models
Three computer system based technologies are proposed to integrate in a higher

education course environment. Course processes are to be integrated with teaching
and learning specific product modeling and product model based equipment con-
trol processes. In the meantime, specific teaching materials, problem solving
drills, and other resources can be integrated in higher education courses from in-
dustrial or semi industrial environments.
Integration includes course model, equipment model, and controlled system
(Fig. 1.). Equipment under development is the environment of product modeling.
Product development and modeling competences are included in the education
program. Product model based programming communicates with controlled sys-
tem, for example in a flexible manufacturing system (FMS). FMS is a well proven
and representative industrial technology and includes so many possibilities to
define competences. Course system communicates with equipment and control
model by using of relationships those interconnect product and course models.
The main task in establishing the planned integration is definition of course ob-
jects and their connections. Any controlled and computer model developed can be
applied such as integrated higher education matter.
Course model structure includes, course, track, module and topic entities as it
was defined in [6] and analyzed by the authors in [9] and [10] (Fig. 1). Equipment
is structured according to the actual product model and includes product, unit,
product object, and relationship level entities. The controlled system includes
Course and Engineering Mod
del Integration for Teaching Automation 1339
Fig. 1. Equipment, control, and

a course models
purpose related equipmeent. For example, an FMS structure includes serveed

equipment, robot system,, and control subsystems. Control system includes com m-
puter system with controol purposed sensing and commanding facilities, and it is
typically is connected witth local and wide area network of computers both in joob
shop and engineering areaa.
140 J. Gáti, G. Kártyás, and M. Bučkko
Fig. 2. Integration of equipm

ment, control, and course models
Integration of course model,

m equipment model, and controlled system operatees
as it is sketched in Fig. 2.. Teachers, students, and engineers are in interaction witth
course and equipment modeling
m processes. Human interaction uses standarrd
processes from industriaal practice. Teaching functionality may be based upoon
special but connectable processes.
p Course and equipment models are connecteed
with the related processes through knowledge definitions in order to appropriaate
explanations for teaching purpose. Controlled system modeling is connected witth
both course and equipmeent modeling process considering appropriate functionns
for education and productt development tasks. In the meantime, process integratioon
is facilitated by special relationships in models.
3 Processes and Co
omponents
In the proposed integrated d modeling, industrial, project, problem solving, commuu-
nication, and course proceesses communicate through an integrated model (Fig. 33).
Integration of the three coommunicating subsystems is done by the communicatioon
both of relevant modeling g processes and relevant model entities. Participant induus-
trial instructors and storeed industrial methods help problem solving in industriial
processes.
Fig. 3. Communicating processes

Fig. 4. Course related compo

onents
Problem solving proceesses are back grounded by problem solving methodds.

Participant students and teachers
t are in connection with industrial, problem solvv-
ing, and project processees through communication processes. They are also in
direct connection with coourse processes. Course, problem solving, and industriial
processes are assisted by course management and teaching, problem solving, annd
industrial methods, respecctively.
Fig. 5. Course related activitties
As for the course relatted components, course processes are in connection witth
course model entities thrrough application of course management and teachinng
methods (Fig. 4). Conseq quently, grounding of development for these methods is
one of the most urgent ressearches in the next future.
Course processes consist of teaching processes and quality management, whiile

quality management inclu udes audit and accreditation. Course model includes onn-
going projects, organizeed contact hours, teaching programs, and teachinng
materials. In industrial problem
p solving oriented courses, contact hours includde
lectures, basic laboratoriies, problem solving laboratories, and problem solvinng
seminars. Teaching progrrams in the course model represent accredited coursees,
plan of programs, and course
c descriptions. A teaching material may be ow wn
prepared, description of an experience, linked, or acquainted. In practice, thesse
methods are mixed by usiing of hyperlinks.
In the proposed modeeling, course administration is undergone of the projeect
administration. In this coontext, project refers to integrated teaching and learninng
purposed group of activ vities. Course administration consists of course objeect
placing and course activitties (Fig. 5). Project administration is supported by casse
studies for projects whille course activities are supported by accredited coursse
requirements verified plann of program structures.
The product-course system is outlined in Fig. 6. In the integrated model, prodd-
uct objects and their struccture are constructed by product definition activities annd
controlled by product deffinition tasks. At the same time, course objects and theeir
structure are executed by y course execution activities and controlled by produuct
definition and course execcution tasks. Special functionality serves relating produuct
and course functionalitiess. At the same time, integrated model includes relating oof
model objects.
Fig. 6. Product-course system

m
4 A Description of Higher Education System

Using basic consideration ns and entities from course modeling in description oof
higher education system published in [8], and considering own experiences iin
course information structtures, the authors of this chapter concluded a classroom
modeling.
Fig. 7. Description of higherr education system

The descriptions appliied at this modeling are summarized in Fig. 7. Higheer

education system includees information for model description in teaching, coursse,
resource, and industrial arreas.
Fig. 8. Functions in the integ

grated system
Teaching area includess information for accreditation requirements, quality conn-

ditions, program contentss, and student profiles. In the practice, student profile foor
teaching and learning can n be configured in a free way restricted by issues from m
accreditation requirementts, quality conditions, and program contents in order tto
assure quality of the courrse. Course area covers administration, schedule, privacy,
and surfaces for communiication. In this case, teaching and learning is a communni-
cation in which schedulee, privacy restrictions are taken into consideration in aan
administration frame. Resource area describes availability of academic personnel,

teaching and engineering materials, laboratory facilities, and projects and assign-
ments. Industrial area depends on the actual connected engineering process, mod-
el, method, knowledge, and experience supports.
Functions in the integrated system are related in Fig. 8. Functionality manage-
ment controls four groups of functions for connection, teaching, solving, and man-
agement. Connection functions handle connections for personal, knowledge, and
company relations. Participant students and teachers, and experts and consultants
from industry are in communication with connection functions. This is essential to
establish outside knowledge, problem solving navigation, and explanation sources.
Management functions handle data, processes, and interoperability affairs. Engi-
neering problem solving functions handle tools, methods, and examples for this
purpose. Teaching functions handle student profiles, contact hours, distance hours,
laboratory practices, project activities, and industrial experts. Engineering sys-
tems, industrial tasks, and industrial engineering activities are in connection with
engineering problem solving functions.
Consistency of the education system is emphasized in order to utilize this im-
portant characteristic of current computer systems at course related applications.
Consistency analysis reveals entities, parameters, and relationships those required
but not included in the education system.
5 Implementation Issues
The concept and approach that is proposed in this chapter is devoted as a contribu-
tion to future methodology in the integrated industrial-academic systems. These
systems will be theoretically grounded, problem solving based, and industrially
oriented. In order to achieve this, a teaching-learning environment will be neces-
sary where understanding of real industrial problems motivates problem solving
and where deep knowledge is applied as a result of deep theoretical and methodo-
logical teaching at a higher education institute.
Advanced and comprehensive product lifecycle management systems have model
development functionality to accept human intent [9]. Characteristics of recent engi-
neering systems support advanced accommodation of knowledge and knowledge
based functionality. At the same time, open surface of these systems makes it possi-
ble to extend them to new application oriented model entities and entity handling
procedures. Integration of higher education and industrial systems is made possible
by group work, product data management, and interoperability characteristics of
recent industrial modeling systems. Consequently, appropriate configuration of a
new generation product lifecycle management systems is proposed to be in the cen-
tre of a future system having the integration proposed in this chapter.
6 Conclusion
This chapter introduces in a contribution to recent efforts in including problem
solving in higher education processes. This contribution is a new concept of and
approach to integration of higher education, industrial engineering, and industrial

automation. Consequently, three computer system based technologies are pro-
posed to be integrated in higher education courses. This implies integration of
lifecycle management of product information, control of equipment in flexible
industrial systems, and virtual classroom technologies. The utmost aim is a power-
ful computer based system for problem solving intensive higher education by
using of engineering, industrial automation, and computer based course manage-
ment systems. Connections in industrial, project, problem solving, communica-
tion, and course processes are discussed and description of higher education
system including information for model description in teaching, course, resource,
and industrial areas is introduced.
In order to contribute to achievements for the above aim, future plans include
further studies of processes and entities in the proposed modeling. Existing higher
education and engineering systems are to be analyzed in order to find method for
implementation where representative systems can be integrated. Future research in
modeling and implementation of the proposed concept requires understanding of
real industrial problem solving and application of deep knowledge.
References
[1] Suna, P.-C., Tsaib, R.J., Fingerc, G., Chend, Y.-Y., Yeha, D.: What drives a success-
ful e-Learning? An empirical investigation of the critical factors influencing learner
satisfaction. Computers & Education 50(4), 1183–1202 (2008)
[2] Selwyn, N.: The use of computer technology in university teaching and learning: a
critical perspective. Journal of Computer Assisted Learning 23(2), 83–94 (2007)
[3] Janssen, J., Erkensa, G., Kanselaara, G., Jaspersa, J.: Visualization of participation:
Does it contribute to successful computer-supported collaborative learning. Comput-
ers & Education 49(4), 1037–1065 (2007)
[4] Montero, E., Gonzalez, E.M.J.: Student Engagement in a Structured Problem-Based
Approach to Learning: A First-Year Electronic Engineering Study Module on Heat
Transfer. IEEE Transactions on Education 52(2), 214–221 (2009)
[5] Chua, H.-C., Chenb, T.-Y., Linc, C.-J., Liaod, M.-J., Chenc, Y.-M.: Development of
an adaptive learning case recommendation approach for problem-based e-learning on
mathematics teaching for students with mild disabilities. Expert Systems with
Applications 36(3), Part 1, 5456–5468 (2009)
[6] Madarász, L., Andoga, R., Főző, L.: Intelligent Technologies in Modeling and
Control of Turbojet Engines. In: Proc. of the New Trends in Technologies: Control,
Management, Computational Intelligence and Network Systems, Rijeka, Croatia,
pp. 17–38 (2010)
[7] Horváth, L., Rudas, I.J.: A Model-based Approach to Virtual University. In: Proc. of
the 4rd International Conference on Information Technology Based Higher Education
and Training, Marrakech, Morocco, pp. 777–780 (2003)
[8] Horváth, L., Rudas, I.J.: Modeling Course for Virtual University by Features. Acta
Polytechnica Hungarica 1(1) (2004)
[9] Horváth, L., Rudas, I.J.: Human Intent Representation in Knowledge Intensive Prod-
uct Model. Journal of Computers 4(9), 954–961 (2009)
[10] Preitl, S., Gáti, J., Kártyás, G.: A Practice Oriented Approach to Intelligent
Computing Assisted Distance Education for Engineering. In: Proc. of the 10th Inter-
national Conference on Intelligent Engineering Systems, London, United Kingdom,
pp. 233–237 (2006)
[11] Gáti, J., Kártyás, G.: New Aspects in Modeling of Internet-based Courses for
Engineers. In: Proc of the 14th International Conference on Intelligent Engineering
Systems, Las Palmas of Gran Canaria, pp. 123–126 (2010)
[12] Gáti, J., Kártyás, G.: Integrated Approach to Course and Engineering Model for
Automation Related Topics. In: Proc. of the INES 2011, 15th IEEE International
Conference on Intelligent Engineering Systems 2011, Poprad, Slovakia, pp. 85–89
(2011)
Dynamics and Control of a Snake Robot
Rectilinear Motion
Alexander Gmiterko, Michal Kelemen, Ivan Virgala, Róbert Surovec,

Martina Vacková, and Erik Prada
Technical University of Košice/Faculty of Mechanical Engineering/Department of Applied

Mechanics and Mechatronics
Abstract. The paper deals with a snake robot rectilinear motion on the flat sur-
face. At first, biological snake rectilinear motion observed in the nature is
introduced. Based on this the transformation of biological snake body to the me-
chanical system is established. After pattern gait design and motion conditions
introduction the mathematical model of snake robot rectilinear motion is estab-
lished. Subsequently an average velocity of motion is derived. From the equation
of average velocity the optimal number of masses is derived in order to maximum
system velocity. The average velocity courses for different pairs of materials in
the graph are shown. In the next section a feedback control system for masses
displacement is introduced. For this purpose PD regulator is used. In conclusion
the summary of study and simulation are done.
1 Introduction
The biological inspired snake robots are able to perform a stable motion in areas
where other kinds of mechanisms are ineffective or not very stable. Their locomo-
tion is on high level of stability because of that most parts of their body are in
contact with the surface on which they move. The snake robots are usually com-
posed from many same segments. On the one hand their body structure enables
them to perform a lot of desired functions. On the other hand their structure is
difficult from the view of control.
A rectilinear motion is one of the four basic biological snake locomotion mod-
es. This pattern gait is usually used by the snakes with heavy bodies which are not
able to move by undulation. This kind of motion is not very effective but it is usa-
ble in environments where gait as lateral undulation or sidewinding are not suita-
ble because of their high amplitudes of yaw. During the motion the abdominal
scales are used for providing the motion through the anchoring. The scales are
alternately smoothly lifting up from the surface and drawing forward and then
lowering down. The parts of the abdomen skin are drawing forward and so the
abdominal scales are joining in the bunch. This part of the body is then pushing
down and the sides of the abdomen go down on the surface. This motion enables
the snake going ahead in one line. So that to enable this mode of motion the snake
fixes several points of longitudinal lower part and moves the body parts among

152 A. Gmiterko et al.
them. The points are called as static or fixed points. The propulsive force of the
snake is primary the friction force and it is the force between the snake and the
surface. [1]
In practice applications there are often necessary mechanisms which can move
through the hard to reach areas, narrow spaces or man dangerous environments. In
some situations may be used rectilinear motion of a snake robot to achieve desired
objects. There are not many functional models concerning the rectilinear motion.
Most works are done only in theoretical level. For motion variables determination
there were used different kinds of mathematical models. [2]
Within this study the model consisting of only identical masses will be consi-
dered. There will not be considered passive bonds between the masses only the
linear actuators.
At first the paper deals with static friction models, namely stiction, Coulomb
friction model, viscous friction model and Stribeck effect. The mechanical system
investigated in this paper will obey isotropic Coulomb friction model behaving
according to Figure 1a. In the chapter 3 the mathematical model is established. By
means of Newton´s dynamics the external forces affecting i-th mass are described
and subsequently the average velocity of N mass system is derived. In order to
achieve the maximum velocity, the optimal number of masses N is derived. The
results are in the graphs shown as well. In the chapter 4 a feedback control system
is introduced by reason of set point tracking of mass displacement.
2 Friction Models
Over the past few decades friction was studied extensively in classical mechanical
engineering. Friction is a tangential reaction force that is present during the rela-
tive motion of surfaces in contact. These reaction forces are the result of many
different mechanisms, which depend on properties of the bulk and surface materi-
als of the bodies, displacement and relative velocity of the bodies, contact geome-
try and topology and the presence of lubrication. [3]
A friction in the mechanical parts of the moving systems causes failures espe-
cially during precise position regulation. Its compensation can be reaches by de-
sign solutions however they do not eliminate friction course nonlinearity within
low velocities.
Within this study a friction is necessary in order to mechanisms motion. The
motion of i-th mass is based on the friction of other masses. In the next section we
will discuss various kinds of friction models namely stiction, Coulomb friction
model, viscous friction model and Stribeck effect. In conclusion of this chapter
different friction models courses depending on velocity are shown in the graphs.
2.1 Stiction
Static friction models only have a static dependence on velocity. This kind of fric-
tion occurs if a mass velocity equals zero and it acts against a relative mass
Dynamics and Control of a Snake Robot Rectilinear Motion 153
motion. The force required to overcome the static friction and initiate motion is
called the breakaway force and can be expressed as
 Fe , v = 0 ∧ Fe < FS 
F =  (2.1)
 FS sgn( Fe ), v = 0 ∧ Fe ≥ FS 
Where Fe and FS is external force affecting the mass and static friction force, re-
spectively. After overcoming static friction force FS, the mass starts move.
The main disadvantage of static friction model is discontinuity at zero velocity.
The discontinuity at zero velocity also leads to numerical difficulties. To over-
come this problem the discontinuous behavior near zero velocity is approximated
by means of a continuous function like arctan function with a very steep slope.
However, a very steep slope around zero velocity can result in very short integra-
tion time steps which slow down or even stop simulation.[4]
2.2 Coulomb Friction Model

A Coulomb friction model does not depend on a velocity magnitude but only on
velocity direction – sgn(v). Coulomb friction coefficient fC is usually lower than
static friction coefficient fS for the same materials. Coulomb friction is also known
as kinetic friction or dynamic friction and can be expressed as
FT = FC sgn(v) (2.2)
FC = fC FN (2.3)
For zero velocities Coulomb friction depends upon the signum function. A com-
mon use of the switching function is:
+1∀v > 0 
 
sgn(v) =  0∀v = 0  (2.4)
 −1∀v < 0 
 
2.3 Viscous Friction Model

A viscous friction occurs in the cases when there is an oil between the contact
surfaces, which reduces friction coefficient f. This coefficient is lower than Cou-
lomb friction coefficient fC. Viscous friction is represented as a linear function of
velocity
FT = fV FN v (2.5)
2.4 Stribeck Effect

A Stribeck effect is a kind of friction which occurs when a liquid or solid oils is
used for the contact surfaces of moving mechanical parts. This causes decreasing
of friction FT with an increasing velocity v until to so-called Stribeck velocity vS.
From this velocity starts affect especially viscous friction. A Stribeck effect is a
function of used oil. Stribeck curve is a continuous drop in the friction force for
small velocities, which originates from the transition of boundary lubrication to
full fluid lubrication through partial fluid lubrication.[5]
The Stribeck effect can be described as:
 F (v), v ≠ 0 
 
FT =  Fe , v = 0 ∧ Fe < FS  (2.6)
 F sgn( F ),otherwise 
 S e 
Fig. 1. Friction models
There are different ways to describe F(v) but the most common form of the
nonlinearity is:
v δS
−
F (v ) = FC + ( FS − FC )e + Fv v
vS
(2.7)
Where vS is called the Stribeck velocity. The function F is easily obtained by mea-
suring the friction force for motions with constant velocity. The curve is often
asymmetrical.
It should be note that above mentioned friction models are idealized and they
are commonly used for simulations. In the cases of accurate positioning regula-
tions the friction causes many problems and it is usually compensated.
In our study propulsive force of a snake robot is based on stiction of static
masses and for motion it is necessary. The problem occurs with the moving mass
friction because of non exactness of friction force course.
3 Modeling of a Snake Robot Rectilinear Motion

In introduction a rectilinear motion of biological snake is introduced. Now it is
important to transform biological snake body to system which can be described by
tools of mechanics.
If we want to analyze a snake body we have to see its biological body as the se-
ries of N identical consecutive elements where each element performs some ac-
tivity. On the Figure 2 the model of the snake body is shown from the view of
mechanics. The following mechanical model is based on study of two-masses
system.[6]
In this case will be investigated only the motion on the flat surface. Between
masses are only linear actuators.
Fig. 2. Simplified model of a snake body
The mechanical system will behaves according to following points: [7]

• All masses have the same weight.
• One motion cycle consists of N phases.
• Only one mass moves while other stay at rest.
• Each phase of a motion is divided into two sections. During the first sec-
tion is i-th mass attracted to the (i+1)-th mass by propulsive force. During
the second section is i-th mass decelerated. A velocity course of each
phase behaves according to Figure 3.
• The actuators of the static masses affect so that they maintain their rela-
tive positions and these masses behave as one mass.
• On the moving mass affects total external force which is the same in each
phase.
• Each phase lasts the same time.
Fig. 3. Course of velocity during one phase
On the Figure 4 the free-body diagram of i-th mass is shown (i = 2, 3 , …, N-1).
Fig. 4. External forces affecting i-th mass
The sum of total external forces affecting i-th mass during the first section of
phase can be expressed through the matrix:
 FP1   ( N − i ) mg 0 
  
= −   fS 
F ( i 1) mg 0  f 
  
P 2
(3.1)
 F f   0 − mg   C 
 
Where FP1 and FP2 are propulsive forces and Ff is a friction force, m and g are
weight of mass and acceleration of gravity, respectively. The sum of total external
forces affecting i-th mass during the second section of phase can be expressed
through the matrix:
 FP1   −( N − i )mg 2mg 

  
= − −   fS 
F ( i 1) mg 0 f 
  
P 2
(3.2)
 Ff  
  0 − mg   C 
The force affecting the 1-th or N-th mass equals:
FP(1, N ) = ( N − 1) f S mg (3.3)
From the Figure 4 we can obtain following formulas:

The duration of each phase tTi equals:
s
tTi = 2 (3.4)
g ( N − 1) f S − f C 
Where s is maximum distance between two masses. The average velocity of
N-mass mechanical system can be expressed as:
1
v A( N ) = sg ( N − 1) f S − fC  (3.5)
2N
If friction coefficients fs and fc are almost identical (or identical for low velocity)
the average velocity equals:
1
v A( N ) = fsg (3.6)
ξ
Where ξ is coefficient which depends on the number of masses N.
2N
ξ= (3.7)
N −2
On the Figure 5 the velocity courses for different materials are shown.
Fig. 5. Velocity courses for various materials

From the (3.5) we can obtain formula for determining the optimal number of
masses N for maximum average velocity depending on the friction coefficients fS
and fC.
2 ( f S + fC )
N= (3.8)
fS
Considering friction coefficients fs and fc as the same we can derive from the
(3.6) the optimal number of masses for the maximum velocity according to the
Figure 6.
Fig. 6. Average velocity dependence on the number of masses N for the same friction
coefficients (fS and fC)
As can be seen in the case of the same friction coefficient (fS = fC) the mechani-
cal system will reaches the maximum average velocity using four masses with the
same weight. It can be also derived from the (3.6).
In theoretical level the snake robot should move by above equations. In practic-
al level there will be some differences. For example as was mentioned the propul-
sive force is based on static friction coefficient of static masses. Achievement of
the same static friction coefficient on the same material is unreal. Small changes
as small piece of oil or impurity cause different friction coefficient. Therefore
above mentioned equations are softly limited and it is possible use them only in
laboratory conditions for now.
4 Snake Robot Rectilinear Motion Feedback Control

In the chapter 3 the snake robot rectilinear motion analysis was done and for this
purpose an open-loop control was used. In other words, the system does not use a
feedback to determine if its output achieves the desired goal of input. The open-
loop control can be used only in the well-known or precise described environment.
Unfortunately, there are a lot of factors which vary and affect a robot during per-
forming the tasks. For example rising inclined plane, changing friction coefficient,
etc. Because of this, it is necessary to use a feedback control system.
In chapter 4 will be controller designed. For verification the simulation will be
done.
4.1 Controller Design

For controller design the pole placement methodology was used. The requirements
for control are:
• Aperiodic course of masses displacement
• Asymptotic set point tracking of required placement
The controller design is based on mathematical model of the snake robot rectili-
near motion. On the figure 7 the block diagram of feedback control system is
shown.
Fig. 7. Block diagram of the feedback control system
Where P(s), C(s) and F(s) are transfer function of plant, controller and sensor.
In our case we will assume F(s)=1, Z(s)=N(s)=0. For plant we can write
B(s)
P(s) = (4.1)
A( s )
After Laplace transformation we yield:
F
P( s) = (4.2)
ms 2
Where F represents propulsive and friction forces affecting the mass. From the
(4.2) it is obvious that we can use PD controller for set point tracking.
n=2 (4.3)
nl = n p = n − 1 = 1 (4.4)
ncl = 2n −1 = 3 (4.5)
Where n, nl-np and ncl are polynomial degree of controlled system, controller and
required system, respectively.
From the (4.3) can be written a transfer function of controller:
p1 s + p0
C (s) = (4.6)
l1 s + l0
Where p1, p0, l1, l0 are coefficient of controller. From (4.4) can be written required
characteristic equation
Acl ( s ) = ( s 2 + 2ξω n s + ω n2 )( s + κ ) (4.7)
The expression ( s 2 + 2ξω n s + ω n2 ) determines dominant poles (red color) of

system and ( s + κ ) determines pole which is less dominant pole of system (blue
color). Therefore it is necessary to place κ so far from the roots of the first ex-
pression in order to the least affects the system behavior. From the system stability
theory it is obvious that κ has to be place to the left from the imaginary part as
can be seen on the figure 8. Value of κ is usually 5 – 10 times higher than domi-
nant pole.
Fig. 8. Poles placement
From the equation (4.2) it is obvious that B(s) and A(s) are known. Our goal is
find the p1, p0, l1, l0 in order to characteristic equation equals Acl(s).
The characteristic equation of feedback control system
1 + P ( s )C ( s ) F ( s ) = 0 (4.8)
Gives us characteristic polynomial

A( s ) L ( s ) + B ( s ) P ( s ) (4.9)
From the (4.9) we can get

ms 2 (l1 s + l0 ) + F ( p1 s + p 0 ) = ( s 2 + 2ξω n s + ω n2 )( s + κ ) (4.10)
From the equation (4.10) we can obtain controller constants

1
11 = (4.11)
m
κ + 2ξωn
10 = (4.12)
m
2κξω n + ω n2
p1 = (4.13)
F
κωn2
p0 = (4.14)
F
By means of equations (4.11) – (4.14) we can calculate controller constants. To
determine these constants it is necessary to determine coefficients ξ and ωn. De-
termination of theses constants is based on the requirement of the system control,
i.e. the course of displacement has to be without overshoot.
This requirement will be reaches by
ωn 1 − ξ 2 = 0 (4.15)
F
ωn = (4.16)
Km
Where K is the maximum displacement of the mass during a one phase.
4.2 Simulation Conditions and Results

For the simulation isotropic Coulomb friction model was used. Maximum dis-
placement of mass is K=0.05 m. On the figure 9, four courses of mass displace-
ment are shown. In the legend various values of ξ are shown.
It should be note that the lower value of ξ is, the higher displacement error is. If
ξ=5 the mass moves more than allowed displacement (red color). If ξ=30 the dis-
placement is nearly K=0.05 m.
At first sight the problem can be reached by choosing high value of ξ, but it is
not entirely true. For the higher value of ξ is the higher actuator requirements are.
If chosen value of ξ is too high it can cause saturation of actuator and control will
be not effective.
Fortunately, in our case we do not require accurate positioning of masses there-
fore we can choose lower value of ξ. For the simulation Matlab / Simulink was
used.
The simulation is softly idealized because of non exactness of friction course.
As was mentioned the snake robots motion is based mainly on friction forces
between snake and surface but on the other hand just that causes the biggest
problem. One of the solutions is experimental determination of friction course for
specific materials.
Fig. 9. Different courses of mass displacement during one cycle of motion
On the figure 10 detail of figure 9 is shown.
Fig. 10. Detail of the mass displacement courses

5 Conclusion
In the paper snake robot rectilinear motion was investigated. For the main prin-
ciple of snake motion is friction between snake body and a surface on which it
locomotes, at first the various kinds of static friction models are discussed. Name-
ly stiction, Coulomb friction model, viscous friction model and Stribeck effect.
After analysis of friction model the mathematical model of snake robot rectili-
near motion was established. Subsequently the cycle time and average velocity
were derived. In the graph different courses of average velocity with different
kinds of friction coefficients are shown. In the case when static friction and kinetic
friction coefficient are the same the mechanism will has maximum velocity with
the four masses. Since open-loop control is not very reliable in every situation,
feedback control system was investigated. For controller constants determination
was used pole placement methodology and subsequently the simulation were done
using PD regulator. From the simulation it is obvious that the higher ξ is the more
accurate control is. On the other hand when the higher ξ is the higher actuator
requirements are. Since our case does not require accurate positioning of masses
we can use lower value of ξ.
The mainly problem in this study is investigation of friction course during the
motion. Described friction models in the chapter 2 are idealized and they are par-
tially different from real friction courses. It should be note that there are a lot of
challenges in this field of research mainly control, and therefore snake robots only
in laboratory conditions are done.
Acknowledgments. This contribution is the result of the project implementation: Centre for
research of control of technical, environmental and human risks for permanent development
of production and products in mechanical engineering (ITMS:26220120060) supported by
the Research & Development Operational Programme funded by the ERDF.
References
[1] Zimmermann, K., Zeidis, I., Steigenberger, J., Behn, C., Bohm, V., Popp, J., Kolev, E.,
Naltova, V.A.: Worm-like locomotion systems (WLLS) – Theory, Control and Proto-
types. Climbing & Walking Robots, 429–456 (October 2007) ISBN: 978-3-902613-
16-5
[2] Dowling, K.J.: Limbless locomotion: Learning to crawl with a snake robot. PhD thesis.
Carnegie Mellon University, Pittsburgh (1997)
[3] Olsson, H., Åström, K.J., Canudas de Wit, C., Gäfvert, M., Lischinsky, P.: Friction
models and friction compensation. Eur. J. Control 4(3), 176–195 (1998)
[4] Iurian, C., Ikhouane, F., Rodellar, J., Grinó, R.: Identification of a system with dry fric-
tion. University of Barcelona (September 2005)
[5] Wang, Y.F., Wang, D.H., Chai, T.Y.: Modeling and control compensation of nonlinear
friction using adaptive fuzzy systems. Mechanical Systems and Signal Processing
(2009)
[6] Chernousko, F.L.: The optimum rectilinear motion of a two-mass system. J. Appl.
Maths Mechs. 66(1), 1–7 (2002)
[7] Gmiterko, A., Kelemen, M., Virgala, I., Surovec, R., Vacková, M.: Modeling of a
snake-like robot rectilinear motion and requirements for its actuator. In: IEEE 15th In-
ternational Conference on Intelligent Engineering Systems, Poprad (2011)
Part III
Computational Intelligence, Web
Technologies and Data Mining
Smart and Cooperative Neighbourhood for
Spatial Routing in Wireless Sensor Networks
Jan Nikodem1, Zenon Chaczko2, Maciej Nikodem1, and Ryszard Klempous1

1
Wrocław University of Technology, Institute of Computer Engineering, Control and
Robotics, Wybrzeże Wyspiańskiego 27, 50-370 Wroclaw, Poland
{jan.nikodem,maciej.nikodem,ryszard.klempous}@pwr.wroc.pl
2
Faculty of Engineering and IT, University of Technology Sydney (UTS),
Bld.Broadway 1, Ultimo 2007 NSW, Australia
zenon.chaczko@uts.edu.au
Abstract. The outcome of applying relations and set theories instead of functions
when staging the proposed routing algorithm is an appearance of choice in Wire-
less Sensor Network nodes. This research presents a model that is universal, prac-
tical, and mathematically sound. In its essence, the model promotes the emergence
of a smart and cooperative neighbourhood. Central, to a rise of emergent proper-
ties in WSN, is the deployment of nodes within neighbourhood that are equipped
with some form of collective intelligence. While defining the concept of spatial
routing, we unfold the rules of smart neighbourhood for relaying data in a dy-
namic and evolving environment. The most interesting aspect of the proposed
concept of neighbourhood smartness is its support for a certain form of stigmergy
that allows neighbourhoods behavioural flexibility in WSN. The network routing
paths adapt to the level of interference and thus avoiding areas of increased distur-
bance. By using Link Quality Indicator parameters, it is possible to choose the
best consecutive relay. This allows adjusting our choices to the disturbance levels
and thus circumventing the area of interference. In the end; this ensures a success-
ful transmission even in the most adverse environmental conditions.
1 Introduction
The Wireless Sensor Network (WSN) is a network of RF devices that are deployed
in some area. These autonomous devices, called nodes, communicate with each
other and cooperatively relaying the data through the network towards a base station
(BS). Traditionally, WSNs monitor physical or environmental conditions, such as:
light, pressure, temperature, motion or chemical pollutants. Separately, the WSN
components are insignificant; however, they became important when they are inte-
grated into the type of network where an emergent cooperative neighbourhood is
created. In such neighbourhood it is possible to developed acquisition applications
that are limited only by the creativity and vision of the WSN designer.
The outcome of applying relations and set theories instead of functions when
staging the proposed routing algorithm is an appearance of choice in Wireless

168 J. Nikodem et al.
Sensor Network nodes. Hence, we tackle the challenge global-local using neigh-
bourhood abstraction. On one hand, communication activity of each node (locally)
within neighbourhood can be determined by actions. The Action should be con-
sidered as the property of every network element and can be defined as a ternary
relation:
Act : Nodes × States → States (1)
On the other hand, globally, we shaping concept of Behaviour, an external attrib-
ute which can be considered either as an outcome of actions performed by the
whole WSN or its subset. Thus, based on this we can construct the quotient set
Behaviour, elements of which are called equivalence classes linked to certain rela-
tion R and here denoted as:
Beh : Act / R = {act [ x ]∈ Act | act [ x ]R x} (2)
Since, Behaviour can be expressed by rules encoded in a form consequent on Ac-

tion performed in nodes, we should provide a model that is universal, practical,
mathematically sound and results in emergence of a smart and cooperative neigh-
bourhood (as a native node’s vicinity). Central, to a rise of emergent properties in
WSN, is the deployment of nodes within neighbourhood that are equipped with
some form of collective intelligence. This enables the WSN to realize behaviour
that a single node cannot afford to perform.
Shaping the concept of spatial routing we unfold rules of smart neighbourhood
available to relay data in an ever changing and evolving environment. A changing
dynamic within the environment causes several limitations, so in order to over-
come them, we should extend cooperation model by allowing it to vary in time.
One of the most striking aspects of proposed neighbourhood smartness is a kind of
stigmergy, which in this issue is about adoption of some principles not just copy-
ing the nature. This stigmergy provides neighbourhood behavioural flexibility:
routing paths adapt to the level of interference avoiding areas of disturbance.
2 Basic Dyadic Relations

The proposed method for multi-hop routing utilizes three basic relations, arranged
into abstractions manipulable by dyadic notation. These relations are defined on a
set of actions (Act) that describes communication activities in the WSN. The
names of the three algebraic relations are: subordination (π), tolerance (ϑ) and
collision (χ) as originally coined by Jaroń [7]. These fundamental relations are
found very useful in describing and visualizing various activities as well as quali-
tative relationship between elements of WSN [5, 8, 9]. Each of the three above
mentioned relations can be represented as a set of ordered pairs <x, y>. In order to
support the concepts presented in this paper and for further discussion on the di-
rected routing in the WSN, we define these relation as follows:
π = {< x, y >; x, y ∈ Act | xπ y}. (3)
Smart and Cooperative Neighbourhood for Spatial Routing 169
The expression xπy - defines the action x which is subordinated to the action y or
in other words, action y dominates over action x,
ϑ = {< x, y >; x, y ∈ Act | xϑ y}. (4)
The xϑy denotes that the actions x and y tolerate each other, and
χ = {< x, y >; x, y ∈ Act | xχ y} (5)
expresses that the actions x and y are in collision.
The basic properties of these relations could be defined [7] as:
π ∪ ϑ ∪ χ ⊂ Act × Act ≠ ∅ (6)
and
(π  π ) ⊂ π . (7)
Formula (6) states that all three relations are dyadic on non-empty set of Actions.
Formula (7) states that subordination is transitive. Further
π ∪ ϑ −1 ∪ (ϑ  π ) ⊂ ϑ (8)
means that:
• subordination implies tolerance – if π holds for some x,y ∈ Act then ϑ
also holds for these,
• tolerance is symmetric–if xϑ y yϑ x,
• subordinated action tolerate all actions tolerated by the dominant – if (xπ
y ∧ yϑ z)  xϑ z.
For collision relation we have that
χ −1 ∪ {π  χ } ∪ (ϑ  π ) ⊂ ϑ l (9)
′
where ϑ is the complement of ϑ:
ϑ l = {< x, y >∈ X × Y |< x, y >∉ ϑ}. (10)
Formula (9) states that collision is symmetric (χ-1 ⊂ χ), disjoint to tolerance (χ-1 ⊂
ϑ′), and subordinated action must be in collision with any action being in collision
with dominant ((π◦χ) ⊂ χ). The paper [8] presents how to use the dyadic relations
in order to model spatial communication. Subordination relation caters for multi-
hop path determination in routing, while tolerance increases a plurality of routing
paths and collision relation forms certain restrictions for the communication space.
3 Ordering of WSN Communication Activity

Dealing with relational multi-hop communication in the WSN, firstly we need to
determine general principles of node cooperation and describe the connection
between neighbouring nodes. For this purpose, let us focus on the relation of
subordination π. Out of all three relations (3)-(5), this one alone is transitive,
which allows us to model the retransmission paths.
The π relation, that is both transitive and reflective, forms a pre-order in the set
of Actions (Act) but further investigation requires a stronger order of this set. In-
troducing a partial order does not appear difficult. In real time applications, nodes
are distributed more or less randomly over a given area (i.e. they may be dispersed
out of a plane). In the case where two network nodes are found very close to each
other, one of them becomes tacit and stays in reserve. In this way, a singular
communication node of substantially greater robustness and survivability is
formed. In mathematical terms, such binding of two elements can be expressed as
the subordination becomes asymmetric. This, in turn, leads to a partial set order
(asymmetric pre-order). Therefore, the set of Actions which is partially ordered
(poset) represents a stronger relationship than the one based on an asymmetric
relation indicted above. The subordination relation is of anti-symmetric nature,
hence this is equivalent to irreflexivity as every relation that is anti-symmetric, is
both asymmetric and irreflexive. Indeed in the WSN, a situation when a sensor
node transmits to itself does not belong to a category of logical behaviour. So,
finally irreflexivity put together with transitivity provides a strict partial order.
The set of actions being finite and partially ordered can be represented in many
ways as any two argument relation can be represented in a form of a directed
graph or a diagram. For such a graphic representation we can use the Hasse dia-
grams which can help us to show the subordination relation between pairs of ele-
ments and the whole structure of partial classification of the set of actions.
Although, the Hasse diagrams are simple and very intuitive tools for dealing with
finite posets, it turns out to be a difficult task to draw clear diagrams for more
complex situations when an attempt is made to represent all possible communica-
tion links in the structure of the WSN. In most cases, when the Hasse technique is
applied, first drawing a graph with the minimal elements of an order and then
incrementally adding other missing elements, one may end-up producing diagrams
that are rather unclear. In these cases, the internal structure of the order is no
longer present due to a large number of connections. Henceforth, a better solution
is required. A new approach that uses the relations can lead to a more viable solu-
tion for the representation of connectivity in the WSN.
In multi-hop WSN, the subordination relation that reflects communication as-
pects of the network is not cohesive and finite. Hence, there are elements in WSN
for which this relation does not take place. However, it is possible to select subsets
of the WSN nodes that are linearly ordered; consequently the partial order will
additionally meet the condition of cohesion. In multiplicity and partial order theo-
ries, order subsets for which the order relation is found to be cohesive are called
chains. To form the chain we shall define the subordination relation setting by the
induction:
π 1 = π , π 2 = π 1  π , ... , π n = π n −1  π , (11)
π n = {< x, y >|< x, y >∈ Act n −1 × Act}, (12)

where x shall be called the successor of y or on the other hand, y will be called the
predecessor of x.
Forming the communication activities in multi-hop WSN is the fundamental
problem because there is a question whether messages from the network area can
be passed onto the base station. On a global scale (this involves the whole WSN)
to build suitable structure that allows us to find the answer for this question we
could borrow a concept from the theory of multiplicity - transitive closure of dy-
adic relation of subordination π on the set Act. However, in this chapter, the prob-
lem of solving the communication activity is perceived from the local level (node
neighbourhood). Therefore, we shall consider a case when a packet is transmitted
from the node y, which then after certain number of retransmissions should reach
the base station (BS). Applying (11), (12), for each sensor node y we define sets of
its ascenders - Asc and descenders - Des using the following expressions:
Asc( y ) = {z ∈ Act | (∃n ∈ N )( y π n z )}, (13)
Des ( y ) = {z ∈ Act | (∃n ∈ N )( z π n y )}, (14)
Expressions (13), (14) determine full communication space of the node y. One of
the main aims, however, is to find an answer to the question ‘to which node in a
communication space send a packet?’ hence we need to pay more attention to the
set Des. It is worth to notice, that for a packet which should arrive from the sensor
node y to its destination at the node BS, it is necessary (but not sufficient) for the
base station to be one of the elements of the set Des(y). Additionally, we could
form many subsets of the Des(y) and some of these subsets can help us to manage
communication activity in the WSN. Among the subsets of Des(y) we can distin-
guish two types:
• four sets that are partially ordered, and
• families of well-ordered chains (linearly ordered sets).
The selected, partially ordered sets can be defined as:
Des min ( y ) = {x ∈ Des ( y ) | ( BSπ y )}. (15)
The subset (15) contains the selected nodes that are the predecessors of the base
station (BS). Hence, only the retransmission path that involves one of these, allows
the packets sent from the node y to reach the BS. The cardinality of this set usually
determined the maximal throughput that can be obtained between node y and BS.
The second subset:
Des max ( y ) = {x ∈ Des( y ) | ( xπ y )} (16)

contains the nodes that are successors of the y node, in other words, these are the
nodes that are required to execute the retransmission of the packet from the node
y. The cardinality of this set usually determines the maximum number of packets
that could be sent from the y node.
The third subset:
Des mis ( y ) = {x ∈ Des( y ) | (~ ∃n ∈ N )( BSπ n x)} (17)

contains nodes that become the dead-ends on the paths to the base station. A
packet that arrives at such a node does not have even a chance to reach the BS.
The last subset
Des pfex ( y ) = {x ∈ Des( y ) | Card (Pr e( x ) > 1}, (18)
where Pre(x) is set of x predecessors, is made up of nodes called pontifices that are
located at intersections of the packet routes. These nodes become the ‘bottle-necks’
on the routing path from node y to the BS. Skilful shaping of the communication
activity allows for the best use of these elements. The cardinality of the set of pon-
tifices determines the capability of packet relaying from one routing path onto an-
other during the re-transmission to the BS.
Fig. 1. Multi-hop routing determined by rules encoded as functions
From the perspective of shaping the communication activity in the WSN, an-
other most interesting subset of Des(y) is a family of chains Chn(y) that consist of
linearly ordered subsets. When building each chain Chn(y) ∈ Chn(y) the following
condition applies:
∀Chn i ( y ) ∈ Chn ( y ) ⊂ Des ( y ) | i ∈ I )() BS =⊥ ∧ y = T ), (19)
where the symbol `⊥` denotes the last and the symbol `⊤` denotes the first element
of the chain Chni(y)
4 Relations Based Spatial Routing

Wireless Sensor Network is a distributed system in which resources and the activ-
ity of decision-making are dispersed territorially. Traditional approaches that have
been reported in the literature to describe a system activity used the functional
apparatus. The activity in network nodes was defined globally applying the
function:
f : X →Y (∀x ∈ X )(∃y ∈ Y )( f ( x ) = y ) (20)
which makes a projection of domain X onto co-domain Y, where X, Y are sets.

From a local perspective (for any WSN nodes x0), function f(x0)) determines ex-
actly one solution y0. By defining f(x) for entire WSN, locally (in each node x0) we
do not leave an opening for any choice (Fig.1), which occurs when the set of feasi-
ble solutions has cardinality greater than 1. This is why it is so hard to achieve the
effect of a territorially dispersed decision making activity when we use functions.
In the above sections, we introduced three relations; π, ϑ, χ, which we use to
describe the communication activity in sensor networks. Close study of such
Fig. 2. Multi-hop routing determined by rules encoded as relations
findings is well reported in several works [10], [8]. Below, we present only a
general idea of how to use relations and spatial routing concepts that are the result
of a such approach.
Let consider the following relation:
R: X →Y (∀x ∈ X )(∃y ⊂ Y ∧ ∃i ∈ I )(R ( x) = y = { y1 , y 2 ,..., y i }) (21)
This relation mapps a set X onto set Y. Any element x0 of domain X is mapped
onto subset {y1, y2,…, yi} of co-domain Y. Defining R(x) for entire WSN, locally
(in each node x0) constitutes a choice. Now, the set of feasible solutions {y1, y2,…,
yi} has cardinality i >1 and this allows to refine network activity on local level
according to current local requirements ( Fig.2).
In relations and spatial routing based method, the source of packets and its sink
(destination) node is determined globally, only. This is depicted as the origin,
node s and the destination (BS) as show in Fig.2. Until the packet arrives from a
given source to the sink (BS), we are not able to precisely indicate the route of the
packet. However, we can provide a collection of possible routes that belong to a
given area of the network (see S in Fig.2 or Fig.3). Here an important question
arises: how many of such chains can be actually chosen? After taking a closer
look at the Fig.3, we should be able to see that Card(x) for each node in the rout-
ing area is equal 5, and from here, in each consecutive node on the routing path
there are always 5 possible choices of next element. The total number of all possi-
ble routes for 6 re-transmitters is equal 56 = 15625 different paths! (see Fig.3).
Fig. 3. Number of possible paths in spatial routing
Therefore, it is possible to influence the size and shape of the area where the
spatial routing is realized and hence define the number of different routing paths
available. For this purpose, we can apply (3) - (5) relations and for any of the se-
lected, the intensity quotient within a given radio link range of a node k can be e
defined as:
IR(k ) = Card (R (k ) / Card ( IS (k ), (22)
where IS(k) is a set of all possible relationships within radio link range of node k.
For the case shown in Fig.3, the intensity quotient of R(k) =πk is equal 5 and
produces a huge number 15625 of different paths, however, if it is reduced to 3,
the number of available paths will be reduced to 729 paths. Taking one step fur-
ther, if the intensity quotient will be set to 1, there will be only one routing path
remaining (see Fig.1). The intensity quotient of subordination π determines the
number of different possible routes towards BS, whereas the intensity quotient of
tolerance ϑ allows shaping interior and the intensity quotient of collision shapes
perimeter of the routing area.
Another consequence of replacing the functional representations is the require-

ment for selection of the next route element at each of the consecutive nodes par-
ticipating in the packet re-transmission. In this way, a truly dispersed territorially
activity of decision making is obtained. The executed routing path is then a result
of making subsequent choices that are successively taken at consecutive re-
transmitter nodes of the routing path chain (19).
Fig. 4. Spatial routing area in WSN
A problem of directed communication towards BS needs to satisfy the condition

BS=⊥, see (19). This offers a trivial solution at nodes which are in the BS radio
communication range. Unfortunately, other nodes, at the time of decision, are not
able to take into account the entire WSN. They have limited information about
their vicinity only, and these ‘nodes’ entire world‘ results from their wireless
communication range. A restricted radio link range decides that a node knows
nothing about WSN topology and can’t see the entire network. Thus, traditionally,
in order to provide directed communication towards BS, network nodes gather all
the information that allow them align the transmission flow towards a preferred
direction. In spatial routing, in order to provide the information about preferred
(towards BS) direction globally, we use the concept of drainage surface [9].
Drainage surfaces ( Fig.5), ( Fig.6) provide a very restricted information about the
whole network, but it is sufficient enough for efficient creating of chains (19).
Fig. 5. Drainage surface based on node to BS euclidean distances
Fig. 6. Drainage surface based on node to BS relay distances

5 Smart and Cooperative Neighbourhood

There are various different definitions of neighbourhood available in the related
literature. Therefore, let us provide a general overview of the concept first, and
then we provide its proper definition using mathematical terms. Vicinity (V) of a
node describes all what is placed in the radio link range of that node. There are
two main groups of objects that are related to the term:
a) Nodes, defined as various different components that belong to the
WSN infrastructure.
b) Indirect elements, these are various network objects that do not belong
to WSN infrastructure, but play an important role in the behaviour of
the network.
A collection of objects from group (a) can be called neighbourhood (N), and a
collection made of objects from the group (b) is defined as environment (E). The
relationship between these two terms can be expressed as:
V = N ∪ E. (23)
Being an essential activity in a multi-hop, Wireless Sensor Networks [1, 4, 6, 16],
the term of cooperation is closely linked with the concept of neighbourhood. In
WSN, most of nodes do not to communicate directly with the base station (BS),
since they are out of the radio communication range. Hence, for nodes in the WSN
field, to transmit data packets to the BS, they are required to cooperate. In conse-
quence, a selected neighbourhood and method of WSN organisation should ensure
that the key property of cooperativity is guaranteed.
The second element of (23) - environment (E) is not only an area that is moni-
tored by WSN but also an area within which WSN is able to effectively operate.
Therefore, the sensor nodes that perform their measurement tasks characteristic to
their environment also need to adapt to changing and often adverse conditions in
their working surround. Some of these conditions, (i.e. environmental electromag-
netic disturbances), can have a very negative impact on operations in WSN
network.
At times, events in the environment conditions are impossible to predict and we
have no means to influence them in any practical way. Therefore, our aim was to
search for a very special kind of network neighbourhood that we could organise in
a specific way. As a result, we have obtained the neighbourhood that is able to
adapt to changing environmental conditions; the neighbourhood that is cognitive,
capable to perceive various levels of disturbance(s) in the environment, and thus
able to react and redirect data packets to `quieter’ regions, avoiding regions of
poor quality of radio transmission. Considering special properties of the proposed
model of neighbourhood, we use the term smart neighbourhood.
The WSN can be considered a really distributed system where nodes provide
necessary computations, communicate with each other within some neighbour-
hood; and maintain the state of the whole network infrastructure. There are several
WSN algorithms which take advantage of some concept of a neighbourhood (e.g.
multi-hop, reliable, bi-directional or geographic) [2, 15, 16]. This is the reason
why, we need to introduce the definition of neighbourhood.
Let us define Map(X;Y) as a set of mapping functions from X onto Y (surjec-
tion), where Sub(X) is defined as a family of all X subsets. In presented model, the
neighbourhood N can be mathematically expressed as follows:
N ∈ Map( Nodes, Sub( Nodes )). (24)
Thus, N(k) is the neighbourhood of node k , and N(C) is the neighbourhood of C

(set of nodes) defined as:
N ( k ) |k∈Nodes := {y ∈ Nodes | y RN k }, (25)
N (C ) |C ⊆ Nodes := {y ∈ Nodes | ( ∃x ∈ C )( y RN x )}, (26)
Fig. 7. Splitting neighbourhood N(k) into three subsets N<(k), N=(k), N>(k)
where RN is ‘to be a neighbour’ relation.

Paper [10] analyses different definitions of neighbourhoods that can be found in
the literature. It also presents advantages and drawbacks of dividing network into
clusters (clustering), single routing path or neighbourhoods. The neighbourhood ab-
straction that is determined by native (mostly technical) constraints such as is the
radio link range, has been proposed and more accurately defined as the native
neighbourhood.
Whereas, clustering represents a kind of simplification, that facilitates computa-
tion [3] however, at the same time, it reduces the set of possible solutions. There-
fore, the native neighbourhood was chosen as the most suitable form for the local
radio range. There are many native neighbourhoods within WSN network which
form an indexed family of sets N={Ni|i∈I} for which following properties hold:
(∀i ∈ I) (N i ≠ ∅) ∧ N = Nodes,
i (27)
(∃max i, j ∈ I | i ≠ j) (N  N ≠ ∅)
i j (28)
which relate to local (at any node) condition:
(∀y ∈ Nodes) (∃max i ∈ I | y N i ≠ ∅), (29)
Fig. 8. Decomposition of intensity quotients within neighbourhood N (27)-(29)
where ∃max means ‘exist as many as feasible’. It means that native neighbourhoods
cannot partition a set of WSN nodes onto mutually exclusive subsets. After having
well defined neighbourhood (27)-(29), let us try to decompose globally defined;
drainage surface and intensity quotients of π, ϑ, χ relations, into identical local
task assign to each node. It is not an easy task to cast all global dependencies from
network area to the neighbourhood. It is even harder if one considers that
neighbourhood conditions for the network nodes might be (and usually are) quite
dissimilar. In papers [9, 12], we described, how using a drainage surface (based on
hop distance), split neighbourhood (27)-(29) into three subsets N<(k), N=(k),
N>(k)( Fig.7). Each subset corresponds respectively to one of π, ϑ, χ relations.
Successively, in paper [13] we described decomposition of globally defined inten-
sity quotients of π, ϑ, χ relations, into these three neighbourhood N(k) subsets
( Fig.8).
We use 2-dimensional matrix N to represent the neighbourhood in WSN. Each
value of ‘1′ in cell n of row k in this binary matrix represents a membership
n∈N(k) (Fig.8). In order to model [π, ϑ, χ] space, we exploit three additional ma-
trices presented on Fig.8. The real number in any cell of these matrices expresses
the intensity quotient of relation. Elements [rk,k] represent the required intensity
quotient within N, while [ri,j]related to particular i-j nodes interactions. Next, we
need to consider for each node k, the distance from the base station (BS) to a node
position, to ensure that the routing activity within N(k) is partially ordered. The
relational state for selected N(k) now can be identified with ordered sequence of
node indices as follows:
< π k , ϑ k , χ k >= ceiling (< 0.2,0.54,0.07 >) =< 2,6,1 > . (30)
Fig. 9. Neighbourhood vector for N(k) space
6 A Simple Model of Adaptive Neighbourhood

The implementation of adaptive neighbourhood can be made possible by the adop-
tion of so-called stigmergic variables (i.e. variables that contain the information
used by nodes to indirectly communicate). These variables are specifically defined
and used by nodes to adaptively change the way they build routing path chain
(19). Any change in a stigmergic variable value determines a change in the pre-
ferred direction towards BS.
Fig. 10. Neighbourhood vector for N(k) space

In cases, when WSN needs to adapt to changing levels of signal disturbance or

noise conditions for the radio transmissions, we have considered to use Link Quality
Indicator (LQI) or Received Signal Strength Indicator (RSSI) parameters. These
parameters form the foundation of final selection algorithm for nodes over which
data will be eventually transmitted. In the end, we decided to apply LQI and thus
we recognised the fact that the information about the quality of a communication
link means something much more than just mere information on the strength of the
received signal that RSSI is able to provide.
The described algorithm was implemented in Matlab and simulation tests that
used different methods of noise generation were performed. In all cases, a circular
shape of signal disturbances with a radius larger than that of the radio link range
itself were form. The following two main types of noise were considered:
a) Constant disturbance remaining within the spatial routing area.
b) Mobile disturbance that moves across the spatial routing area.
Fig. 11. Neighbourhood N(k) vector modified by LQI
The results obtained when running the test (a) are shown in Fig.12. Infrequently,
when the relation of subordination (π) was used only, and when consecutive
nodes were too close to the disturbance zone, routes were falling out and not being
able to find any consecutive re-transmitter in its range. I cases, when a route
reached its factual ‘dead end’ it is moved to the set (17) called Desmis. Although,
that particular data packet is lost, the following packets are no longer directed to-
ward the marked node. The situation would persist until the node is able to im-
prove its LQI.
Fig. 12. Adaptive behaviour for constant disturbance
By using the relation of tolerance (ϑ), it is possible to improve sensor network ca-
pability to avoid the disturbance or noisy areas. Let us remark here that according
to (8) (π⊂ϑ). Therefore, applying ϑ means also using the relation π.
When using the full range of relations (π, ϑ, χ) for improving networks capabil-
ity to circumvent disturbance areas, it is also possible to prevent continuous packet
losses when they arrive at the node Desmis (17). This is particularly important in cases
of non-circular disturbance zones. In such situations, a packet can be passed onto
re-transmitters that are in certain distance from the disturbance zone. The draw-
back of such an approach is a noticeable increase in the length of routing paths.
Fig. 13. Adaptive behaviour for mobile disturbance, crossing spatial routing area
The results of the simulation tests, for cases when signal disturbances moves
across the spatial routing area for each of the three discussed relations are shown
in Fig.13. The currently executed routing paths are marked with a thicker line. The
spatial routing method tested in these experiments, has indicated persistent and
adaptive tendencies toward avoiding the noisy areas of the WSN field. Such a
desirable behaviour was observed when relatively moderate rates of a disturbance re-
gion movement were applied. This is due to the fact, that the direct neighbourhood
of disturbances requires some time for actualisation of LQI values in order to
reflect the changing propagation (anisotropy) conditions in the network.
7 Conclusion
In this chapter, we have investigated enhancement methods for the route selection
in WSN. Specifically, an adaptive, multi-hop algorithm based on a spatial routing
concept was proposed. The performance of spatial routing algorithm can be im-
pacted by the selection of various routes over which data is being transmitted. How-
ever, the spatial routing algorithm that determines the energy consumption within
neighbourhood is not able to ensure the adaptive network performance in wireless
environments. Henceforth, the approach taken is based on use of the existing spatial
routing features and combining them with LQI indicators in order to aid the route
selection within neighbourhood.
The ZigBee standard [17] was used, as it enables a measurement of the link
quality between neighbouring nodes in the network. This measurement is based on the
Link Quality Indicator (LQI) value that is reported with each of the received packet.
By applying the ordering of neighbours’ mechanism according to the value of
LQI, a choice of the consecutive relay with the best LQI can be ensured. Addi-
tionally, this approach allows an adjustment of our choices to the disturbance
levels and an ability to mitigate interferences, thus ensuring successful radio
transmission even in the most adverse environmental conditions.
At present, our objective is to investigate further enhancements of our model that
would increase the overall adaptability properties in the WSN. It is anticipated;
that by investigating various techniques of cooperation among nodes within neigh-
borhood we would be able to improve the adaptability of route selection.
Acknowledgments. This chapter has been partially supported by the project entitled: Detec-
tors and sensors for measuring factors hazardous to environment modeling and monitoring of
threats. The project financed by the European Union via the European Regional Development
Fund and the Polish state budget, within the framework of the Operational Programme Inno-
vative Economy 20072013. The contract for refinancing No. POIG.01.03.01-02-002/08-00.
References
[1] Burmester, M., Le, T.V., Yasinsac, A.: Adaptive gossip protocols: Managing security
and redundancy in dense ad hoc networks. Ad Hoc Networks 5(3), 13–32 (2007)
[2] Cerpa, A., Estrin, D.: ASCENT: Adaptive Self-Configuring Sensor Networks Topol-
ogies. IEEE Transactions on Mobile Computing 3(3) (July-September 2004)
[3] Cohn, A.G., Bennett, B., Gooday, J.M., Gotts, N.M.: Representing and Reasoning
with Qalitative Spatial Relations about Regions. In: Cohn, A.G., Bennett, B., Gooday,
J.M., Gotts, N.M. (eds.) Spatial and Temporal Reasoning. Kulwer, Dordrecht, pp. 97–
134 (1997)
[4] Chaczko, Z., Ahmad, F.: Wireless Sensor Network Based System for Fire Endan-
gered Areas. In: ICITA 2005, Sydney (2005)
[5] Chaczko, Z.: Towards Epistemic Autonomy in Adaptive Biomimetic Middleware for
Cooperative Sensornets, PhD thesis, UTS, Australia (2009)
[6] Feng, C., Yang, L., Rozenblit, J.W., Beudert, P.: Design of a Wireless Sensor Net-
work Based Automatic Light Controller in Theater Arts. In: ECBS 2007, pp. 161–170
(2007)
[7] Jaroń, J.: Systemic Prolegomena to Theoretical Cybernetics, Scient. Papers of Inst. of
Techn. Cybernetics, vol. (45). Wrocław Techn. Univ., Wrocław (1978)
[8] Nikodem, J., Klempous, R., Chaczko, Z.: Modelling of immune functions in a wire-
less sensors network. In: 20th European Modeling and Simulation Symposium,
EMSS 2008. Campora S. Giovanni, Italy (2008)
[9] Nikodem, J., Klempous, R., Nikodem, M., Woda, M.: Multihop Communication in
Wireless Sensors Network Based on Directed Cooperation. In: 4th International Con-
ference on Broadband Communication, Information Technology and Biomedical Ap-
plications, BroadBandCom 2009, Wroclaw, Poland (2009)
[10] Nikodem, J.: Designing Communication Space in Wireless Sensor Network Based on
Relational Attempt. In: Moreno-Díaz, R., Pichler, F., Quesada-Arencibia, A. (eds.)
EUROCAST 2009. LNCS, vol. 5717, pp. 83–90. Springer, Heidelberg (2009)
[11] Nikodem, J., Nikodem, M., Woda, M., Klempous, R., Chaczko, Z.: Relation-based
message routing in wireless sensor networks. In: Chinh, H.D., Tan, Y.K. (eds.) Smart
Wireless Sensor Networks, pp. 127–145. InTech (2010)
[12] Nikodem, J., Nikodem, M., Klempous, R., Chaczko, Z.: Multi-hop and Directed
Routing Based on Neighborhood Cooperation in WSN. In: 15th IEEE International
Conference on Intelligent Engineering Systems (INES), Poprad, pp. 221–227 (June
2011) ISBN: 978-1-4244-8954-1, doi:10.1109/INES.2011.5954748
[13] Nikodem, J., Chaczko, Z., Nikodem, M., Klempous, R., Wickramasooriya, R.: Com-
bating Security Threats via Immunity and Adaptability in Cognitive Radio Networks.
In: Fodor, J., Klempous, R., Suárez Araujo, C.P. (eds.) Recent Advances in Intelligent
Engineering Systems. SCI, vol. 378, pp. 221–242. Springer, Heidelberg (2012)
[14] Pichler, F.: Modeling Complex Systems by Multi-Agent Holarchies. In: Kopacek, P.,
Moreno-Díaz, R., Pichler, F. (eds.) EUROCAST 1999. LNCS, vol. 1798. Springer,
Heidelberg (2000)
[15] Stojmenović, I. (ed.): Handbook of Sensor Networks Algorithms and Architectures.
John Wiley and Sons Inc. (2005)
[16] Younis, O., Fahmy, S.: HEED: A Hybrid, Energy-Efficient, Distributed Clustering
Approach for Ad Hoc Sensor Networks. IEEE Transactions on Mobile Compu-
ting 3(4) (October-December 2004)
[17] ZigBee Standards Organisation, ZigBee Specification, Document No 053474r17,
Sponsored by ZigBee Alliance, San Ramon, CA, USA (January 2008)
Elimination of Dynamic Errors of
Thermocouples in Aircraft Engines Using
Neural Networks
František Adamčík1, Rudolf Andoga1, Ladislav Madarász2,

and Peter Krajňák3
1
Technical University of Košice, Faculty of Aeronautics, Department of Avionics,
2
{frantisek.adamcik,rudolf.andoga,ladislav.madarasz}@tuke.sk
3
AFB Sliač, Slovakia
peter.krajnak@gmail.com
Abstract. The article is focused on diagnostics in the field of aviation turbo-jet

engines. Evaluation of their technical status, and potential failures and pre-failure
processes can be improved by monitoring their thermally loaded components. Part
of the research in the given area also involves mathematical modeling of the chan-
nel measuring the exhaust gases temperature, using neural networks to compen-
sate for dynamic errors of slow thermocouples used in such measurements.
Keywords: turbojet engines, diagnostics, artificial intelligence, neural networks.
1 Introduction
Production costs, operational costs come today into forefront of constructer inter-
ests of aviation technicians together with high reliability. The increase of output
and efficiency of the newly constructed aircraft turbo-compressor engines is given
by construction and materials used in “hot parts” of such engines and quality of
control processes. Regarding these facts, efficient methods of technical state ob-
servation are being researched [5].
The present systems of flight parameters save a big amount of data and parame-
ters, based upon which the technical state of an aircraft can be assessed. The pos-
sibilities of the present computational technologies allow creating big databases of
data about parameters of individual blocks, aggregates, nodes and systems of
aircraft. This can assess the technical state of different systems onboard of the
aircraft.
The parameter fully reflecting the load of hot parts of an engine is the tempera-
ture of gasses in-front of the turbine (t3*). Such measurement is done very seldom
in real operation because the placement of sensors in-front of the high pressure
turbine (HPT) considerably decreases reliability of an engine. The only way to ob-
tain this temperature is to solve the problem of dynamic error of the output

186 F. Adamčík et al.
temperature t4* and create a model of this temperature from other known and rec-
orded parameters by means and approaches of system identification. The article
deals with modeling of the temperature channels and relation between real and
measured temperatures in-front and behind high pressure turbine, as well as with
possibilities of recalculation of these temperatures for purposes of diagnostics and
assessment of the engine’s state in-flight and after flight [1].
2 Modeling of the Temperature Channels

The physical essence of temperature measurement using thermo-elements (ther-
mocouples) gives the channel of temperature measurement of gases aft of the tur-
bine the characteristics of a dynamical, non-linear system, defined mainly by vari-
able the delay of the output signal of the thermo-element with variation depending
on the direction in the thermal change (increase, decrease). Different approaches
can be taken to eliminate this dynamic error that can hide rapid temperature
changes that can considerably damage engine and are not taken normally into
account.
The dynamic error of t4* temperature behind the HPT is caused by the process
of gas flow to the thermally sensitive element of the theromocouples – the hot
junction. The dynamic error is generally characterized by a time constant given by
the formula [11]:
cm
τ= , (1)
αQ S
where :c – thermal capacity of the hot junction of the thermocouple,
m – weight of the sensitive hot junction,
S – surface of the hot junction,
α – the coefficient of heat transfer of the hot junction material[11]..
* *
t 4S K t 4 mer
τp + 1
Fig. 1. The general model of the thermocuple
In this approach the description represents a-periodic inertial element (fig. 1)

that can be described in operator form by the following equation:
K
t 4* = t* , (2)
mer τ p +1 4S
Elimination of Dynamic Errors of Thermocouples in Aircraft Engines 187
where t4*mer – is the measured value,

t4*S – the real value of the temperature,
p – Laplace operator [11].
To compensate the influence of the dynamic delay in measurement of the t4* tem-
perature a derivative element can be applied (Fig. 2).
The correction signal value Δt4*mer can be expressed in the operator form as
follows
K kor Р
Δ t 4* mer = t 4* S , (3)
τ kor P + 1
where K kor – is the gain coefficient of the correction circuit,
τkor – time constant of the correction circuit.
The equation (3) can be transcribed into the following form shown in the figure 2.
t*4 sk t*4 mer t* 4 mer kor

k
Tp + 1
Δt4*mer
k kor p
Tkor p + 1
Fig. 2. The general compensation model of the thermocouple
In practice, the coefficient K kor and its time constant τ kor of the correction
circuit is set in the way so that the resulting signal t4*kor mer = t4*mer + Δt4*mer
would be as close as possible to the real temperature t4*S behind the turbine:
t4*S ≈ t4*kor mer = t4*mer + Δt4*mer. (4)
3 Relation between Temperatures and Its Modeling

The equilibrium of output and work of compressor and turbine, the amount of the
absolute temperature in-front of the turbine t3*can be described by the following
formula:
T3* = T4* +
GV 1 c рв
⋅ ⋅
G D η T c рГ
(
⋅ T2* − T1* , ) (5)
where: GV – air mass flow on the inlet of the engine, GG – air mass flow per seconds
through the outlet of the engine, ηT – pressure ratio of the turbine, c рв – mean
measurable temperature capacity of air by T2, temperature c рГ - mean measura-

ble temperature capacity of air by the temperature T3, T2 – tetemperature of air be-
hind the compressor, T1*– temperature on the inlet of the engine [11][12].
The formula (5) is correct for normal regimes and afterburning regimes.
Regarding [1] it is known that turbo-reactive engines can be described by formula
(6):
GВ 1 c рв
⋅ ⋅ ≈ 0,87 . (6)
G Г η m c рГ
Temperature t2* can be calculated by the relation if it is not measured
T2* = T1* 3 π K , (7)
PK
where: πK = – is the pressure ration behind the compressor and on the inlet
PВ
of the engine. In this way the formula 5 can be transformed into the following
form to calculate the temperature t3*
 P 
t3* = t 4* + 0.87 ⋅ t1* ⋅  3 K − 1 (8)
 PB 
4 Neural Modeling of the Temperature Channgel

The design of a neural model can be based on a fundamental architecture result-
ing from the experiments with modeling of complex systems as well as pilot-
experiments on the basis of a selected data set for the construction of
a correctional unit to eliminate the dynamic error of the channel for temperature
measurement (t4c) [16,17]. The selected architecture contained two hidden layers
with thirty neurons in the first and twenty neurons in the second hidden layer.
Innovative approach in this field is the application of input delay window of the
network. Then such a network is learning how to approximate the sequential da-
ta in the selected time frame. In the given case the time frame, as the result of
the experiments, was set to three and four samples. The width of the time frame
corresponds to the degree of the modeled system- i.e. to the thermo-element. It
means that this dynamic system could be described by a non-linear, third or
fourth-order differential equation [3]. The basic structure of the neural network
of the inverse model of the channel of temperature measurement is illustrated in
the figure 3.
Fig. 3. The designed neural network to model the temperature channel
(T99) previously used to measure the temperature T3c (marked as t3c_mer) and
the high pressure compressor rpm, n2. The required output for the inverse model at
the training, was the true temperature T3c measured by a high-speed thermo-
element (marked as t3c_m). The predicted output of the neural network
representing the true value of the temperature is marked as t3c_kor. The network is
trained by the SCG („Scaled Conjugate Gradient“) algorithm on the basis of the
error signal, computed in each step as A difference between the true temperature
and the predicated temperature of the neural network.
t3c_m t3c_mer
Temperature
measurement
Inverse neural t3c_kor

n2 model
e (t)
Fig. 4. The training architecture used to model the temperature channel
For training and validation of the inverse neural model two data sets of real-
world data from an engine RD-33 were used. The first set designated as FSM-1 is
shown in the figure 5 and the second data set was designated as FSM-2 and is
shown in the figure 6. Two approaches were tested, the set FSM-1 for training and
the set FSM-2 for validation and their reversal. The training of different neural
networks with either single input (t3c) or double inputs (t3c and n2 – speed of the
compressor) is summarized in the table 1. Mean absolute percentage error
(MAPE) and maximum absolute percentage error (MAAPE) have been selected to
assess performance. The table shows that networks that had an additional input
from the speed of high pressure compressor n2 generally better results than net-
works with a single input of temperature. Also networks that were trained using
the first data-set FSM-1 generally performed better. The best network that is
proposed and further presented in the article is designated as network INM3a
trained on the dataset FSM1.
Fig. 5. The training set FSM 1
Fig. 6. The training set FSM 2

Table 1. The training results on both training sets with cross validation
The inverse neural model trained on measurements of temperatures T3 can also

be used for temperature T4, as both temperatures can be recalculated. The corres-
ponding simulation scheme of the inverse neural model and the structural diagram
of the mathematical correction for the dynamic error of the channel of temperature
measurement T4c is illustrated in Fig. 7.
Fig. 7. The structural scheme of mathematical correction of dynamic error of T4c mea-
surement channel
Fig. 8. The implemented system for dynamic error elimination in Matlab environment
Behavior of the inverse model at its verification on a testing data set of meas-
ured temperatures T3c is obvious form Fig. 9. The red curve (t3c_m) represents the
true temperature T3c undistorted by the dynamic error of the thermo-element. The
green curve (t3c_mer) represents the temperature characteristics with the dynamic er-
ror of the sensor. Prediction of the output of the inverse neural model is
represented by the blue curve (t3c_kor). Comparison of the of the curve characteris-
tics leads to a conclusion that at the given training set, the inverse neural model is
eliminating the dynamic error of the thermo-element when measuring temperature
t3c. The presented comparison also points out the fair dynamical characteristics of
the inverse model, which is capable of eliminating the dynamical error of the
thermo-element at rapid changes of the input variable.
Temperature (oC)
time (0.1 sec)
Fig. 9. Testing results shown compared to real-world data

Temperature (oC)
time (0.1 sec)
Fig. 10. Testing results shown compared to real-world data expanded to show elimination
of the dynamic error.
The presented approach allows us to be used in many engine applications where
slow thermal sensors are used to measure temperature. The approach using a neur-
al network can bring sufficient quality and perspective real world application with
enough robustness and adaptability to solve the problem for different thermo-
couples used in extreme temperatures measurements. The author have found a
suitable architecture of the neural network that can be expanded for further expe-
riments and further engineering applications.
No. 1/0298/12 – “Digital control of complex systems with two degrees of freedom”. The
work presented in this paper was also supported by KEGA under Grant No 018TUKE-
4/2012 - Progressive methods of education in the area of control and modeling of complex
systems object oriented on aircraft turbo-compressor engines. This support is very grateful-
ly acknowledged.
References
[1] DeCastro, A.J.: Rate - Based model predictive of turbofan engine control. QSS
Group, Inc., Cleveland Ohio (2006)
[2] Harris, C., Hong, X., Gan, Q.: Adaptive Modeling, Estimation and Fusion from Data,
p. 323. Springer (2006) ISBN: 3-540-42686-8
[3] Hlavička, J., Racek, S., Blažek, T.: Digital Fault tolerant systems, p. 330. ČVUT,
Praha (1992) (in Slovak language) ISBN 80-01-00852-5
[4] Jonathan, S.L., Turso, J.A., Shah, N., Sowers, T.S., Owen, T.S.: A Demonstration of
a Retrofit Architecture for Intelligent Control and Diagnostics of a Turbofan Engine.
NASA/TM -2005- 214019 (2005)
[5] Kelemen, M., Lazar, T., Klecun, R.: Ergatic systems and safety in aviation (Ergatické
systémy a bezpečnosť v letectve: edukácia a inteligencia zručností v leteckej
prevádzke: vysokoškolská učebnica), vol. 1, p. 316. Akadémia ozbrojených síl ge-
nerála M.R. Štefánika, vyd - Liptovský Mikuláš (2009) (CD-ROM) ISBN: 978-80-
8040-383-6
[6] Krajňák, P.: The utilization of the recorded flight data for the determination the tech-
nical state of the aircraft jet engine. Dissertation thesis, Faculty of Aeronautics,
TUKE, 105 p. (2010)
[7] Kreidl, M.: Diagnostics systems, p. 314. ČVUT, Praha (1997) (in Czech language)
ISBN: 80-01-01670-6
[8] Lazar, T., et al.: Development tendencies and modeling of avionics systems: MoSR,
Bratislava, 160 p. (2000) (in Slovak language) ISBN: 80-88842-26-3
[9] Linke-Diesenger, A.: Systems of commercial turbofan engines: an introduction to
systems functions. Springer (2008) ISBN: 978-3-540-73618-9
[10] Madarász, L., Andoga, R., Főző, L.: Intelligent Technologies in Modeling and Con-
trol of Turbojet Engines. In: Er, M.J. (ed.) New Trends in Technologies: Control,
Management, Computational Intelligence and Network Systems, Sciyo, pp. 17–38
(2010) ISBN: 978-953-307-213-5,
http://www.intechopen.com/articles/show/title/intelligent
-technologies-in-modeling-and-control-of-turbojet-engines
[11] Považan, J.: Konštrukcia matematických modelov leteckých turbo-kompresorových
motorov. VLA M.R.Š. v Košiciach (1999) ISBN: 80-7166-030-2
[12] Sanjay, G.: Fundamentals of Aircraft Engine Control Design Course, Lecture. NASA
Glenn Research Center (September 15, 2007)
[13] Simon, S.G., Guo, T.-H., Mercer, C., Millar, R., Behbahani, A., Bajwa, A., Jensen,
D.T.: A survey of inteligent control and health management technologies for aircraft
propulsion systems. NASA (2005)
[14] Simon, D.L., Simon, D.: Constrained Kalman filtering via density function truncation
for turbofan engine health estimation. Glenn Research Center, Cleveland Ohio (2006)
[15] Yang, L., Ioachim, I.: Adaptive estimation of aircraft flight parameters for engine
health monitoring system. Delta Air Lines, Atlanta (2002)
[16] Várkonyi, T.A., Tar, J.K., Bitó, J.F., Rudas, I.J.: Simple Noise Reduction in the
Adaptive Synchronization of Coupled Neurons by Robust Fixed Point Transforma-
tion. In: Proc. of the 15th IEEE International Conference on Intelligent Engineering
Systems (INES 2011), Poprad, Slovakia, June 23-25, pp. 297–302 (2011) ISBN: 978-
1-4244-8954-1
[17] Várkonyi-Kóczy, A.R.: Observer-Based Iterative Fuzzy and Neural Network Model
Inversion for Measurement and Control Applications. In: Rudas, I.J., Fodor, J.,
Kacprzyk, J., et al. (eds.) Towards Intelligent Engineering and Information Technolo-
gy. SCI, vol. 243, pp. 681–702. Springer, Heidelberg (2009)
Special Aspects of Teaching in Virtual Learning
Environments
Andrea Tick
Budapest Business School, Faculty of International Business and Management, 1165

Budapest, Diosy L. u.22-24, Hungary
tick.andrea@kkfk.bgf.hu
Abstract. In the Information Society in the 21st century the ever evolving
information technological innovations and challenges force education to introduce
these innovations in the teaching and learning processes. Students expect stimuli
and challenges and are not satisfied any more with the traditional and conventional
ways of learning. Thus the launch and spreading of 3D virtual learning
environments is a must in public and higher education as well and will expectedly
boom in the next decade. This paper outlines the special aspects to be considered
when developing and introducing virtual learning environments considering both
the student and the teacher sides.
1 Introduction
Educational institutions, teachers and tutors especially in higher education all face
the challenges of the information society and meet the products of the
continuously and fast developing information technologies which alter, modify
and ultimately shape the structure of teaching and learning. Students and teachers,
especially institutions have to constantly raise such questions as how to adapt to
the new situations, how to offer such courses and certain additional alternatives
that can support effective learning, attract students to take on these courses, and at
the same time can be organically integrated in the existing educational structure
and effectively co-live with the already used methodologies. Life-long learning is
not a question any more, blended or distributed learning are such ways of teaching
and learning that are affected by the physical and social surroundings of an
institution. The absorption of telecommunication and mass-media technologies
enhanced the birth of the Global village as proposed by Marshall McLuhan in the
1960s. [1] According to him, with the help of electric technology connections can
be built from any point of the world thus contracted the world into a small village.
He presupposed the internet as an “extension of consciousness” [1] thirty years
before its invention. As he stated, a computer as a research and communication
instrument could enhance retrieval, obsolesce mass library organization, retrieve
the individual's encyclopedic function and flip into a private line to speedily
tailored data of a saleable kind [1]. The instantaneous movement of information
from every quarter of the world to every point of the world at the same time calls

196 A. Tick
forth a radical change in the learning process as well as in the visualization of the
learning environment. A new, more effective form of knowledge acquisition
becomes a central role player in the evolving knowledge-based society, which is
characterized by constructivist pedagogy, emphasizing life-long learning and the
training of competences. Knowledge becomes the primary source of the added
economic value [2]. The question again is how to create such open, active,
creative, collaborative, and primarily motivating learning environments that would
fit both the students’ and the tutors’ needs and requirements. By now students
have turned into active participants of the learning process most of them belonging
to the digital generation. On the other hand, teachers have adapted to the situation
and the majority of them participate in in-service training in order to meet the
challenges. Changes in the learning environment have gone through not only due
to radical technological developments but also because of the pressure from the
student side. Learner-centered actual virtual classrooms come to existence in the
global village in which the student-student as well as the student-teacher
relationships alter and shape radically. In this process that is in developing virtual
learning environments and classrooms the fundamental transformation of
communication technologies, the usages of web 2.0 applications as well as the
proliferation of eLearning technologies focusing on 3D visual solutions are the
key role players.
2 Virtual Learning Environment – Extension of Classrooms

The radical social and technical-technological changes in the industry and in the
society, namely the appearance of the digital generation, the emergence and
sudden boom of social networking, the proliferation of the mobile internet capable
devices triggered the appearance of such educational services, such extended
virtual classrooms that enable instant participation and can be reached anywhere
and anytime.
2.1 The Grown-Up “Digital-Generation”

Teachers and tutors in the 21th century must recognize that learners of this century
do not learn like their predecessors. They must find balance between past
methodologies and ways of teaching-learning and their students' future, closing
the gap between the digital native learning and the conventional teaching
environment. Different generations have diverse attitude to internet use and
lifelong learning. The technologies the older generations do not understand or
simply take for granted have become a part of students’ identities. This digital
bombardment is shaping who they are and who they will become [3]. We must
realize they live in a different world, and speak a different language. The oldest
generation, the so-called veterans born before the baby boom regard the internet
and the use of computer as a challenge since they met the internet late in the life.
The baby boom generation born up to the middle of the 60s’ uses the computer
and the internet for work and mainly at their workplace. The next generations
Special Aspects of Teaching in Virtual Learning Environments 197
called XYZ are the children of the digital generation. They gain information they
want quickly medium-independently and share and re-interpret this information
with their virtual peers. The Internet is present in the private life of the Generation
X (born between the middle 60s’ and the early 80s’) but it does not influence and
change their lifestyle to a great extent. The Millennium Generation or Net
Generation as the Generation Y is also called was born from the early 80s’ to the
early years of 2000 and it is the first wave of the digital generation. They are the
children of the baby boomers They saw the rise of instant communication
technologies and the use of web 2.0-based websites is part of their life. They
already look for such applications and eager to join social networking and study in
peer-oriented environments. This generation knows what it wants, when, where
and how it wants it. The youngest generation, Generation Z, called also as
Generation I or Internet generation, was born within the use of the internet. They
are exposed to the World Wide Web as early as on their birthday. The use of
instant and text messages, mobile internet devices, cloud computing justifies their
nickname “digital natives”. The children of this generation usually have an online
identity.
Consequently, the generations grown up in the last two decades is not the
“Nintendo-generation” that grew up on small manual games, but the “PlayStation-
generation” grown up on a much more developed HD quality technology. This is
the generation that learnt to use the computer well before they learnt to write, who
handles the DVD player better than their parents. This generation lives “on-line”,
handles the virtual worlds as the extension of the real one. This generation is the
one that if goes home from school, logs in the social network and chats with the
classmates, while in another window plays a 3D videogame with someone from
the other end of the world. Consequently, these students browse much more
comfortably in the virtual space, since it is entirely natural to them [4].
2.2 Characteristics of Virtual Learning Environments

The term virtual learning environment (VLE) has been defined by many authors
and in several ways. The concept deserves a little explanation. What shall we call
a virtual learning environment? The term has been used in a very broad sense by a
lot of authors including learning and teaching possibilities and facilities ranging
from simple web pages through Learning Management Systems or Learning
Content Management Systems to as far as 3D virtual learning environments like
Second Life. Some might call static hyperlinked web pages with less sophisticated
user interfaces and graphical display a virtual learning environment while others
restrict the concept to three-dimensional VLEs where a special world is created
with an avatar taking on different characteristics and behaviors [5]. A set of web
pages, however, does not constitute a virtual learning environment unless there is
social interaction about or around the information [6]. A few examples of virtual
social interactions may be instant messaging, discussion boards, emails, blogs, and
podcasts. A widespread definition of VLE defines it as “a set of teaching and
learning tools designed to enhance a student's learning experience by including
computers and the Internet in the learning process”[7]. Some others state that with
198 A. Tick
the help of web-based possibilities real world conventional education can be

modeled and be supplemented by virtual classes, groups, classrooms including
tests, homework, tasks and assignment options with the aim to foster learning. The
creation of self-organizing groups of students enables the formation of classes and
study groups even at a very simply technical level. Interactivity, community
forming opportunities, thus, create a new layer in the study environment. The
umbrella term VLE thus includes all types of information technology-based
learning environment that are built on intranet or internet grounds ranging from
the use of a software product via registering and completing online courses to
becoming a member of a virtual world and participating in virtual seminars and
discussions. However, some argue that the physical teaching environment cannot
be eliminated from the learning process and the VLE must be well and accurately
designed and integrated in the curriculum or, as a further possibility it should
provide an extracurricular option to widen the scope of learning. The question is
where the focus is. In case of higher education there is a tendency to offer courses
in VLE because distance learning as well as the nature of lifelong learning
deserves it. This way in a wider sense an open virtual educational learning
environment is born in which virtual learning becomes general. However, personal
face-to-face communication cannot be avoided [2], for instance a kick-off meeting
of a course, that helps identify participants or to clear details in a very short
amount of time. In addition, virtual learning environments should incorporate all
the functions that exist in physical learning environments and must be set up in a
way that it gives the framework for formal as well as informal communication,
administration, notice boards, virtual libraries, learning and assessment facilities
etc. The question here is not to build a fence between virtual and physical
environments but to integrate them to form an effective and comfortable learning
environment. The following table lists some tools that can be combined to create
an optimal learning environment for students:
Table 1. Learning Strategies and Integrated Tools based on NIIT and Rossett- Douglis-
Frazee categories [2g]
Virtual Live/Non-virtual
LMS Phone
Information transfer
Email Notice board
Instruction
Online bulletin boards Face-to-face meeting
e-books
Library
online resource links
Books
web-based tutorials
Demonstration/Self study Printed material
EPSS
Workbooks
Web-learning modules
Conventional classroom
Knowledge databases
Face-to-face, traditional
Webinar1
Emulation session classroom
email
Workshops,
1
Webinar is short form “for Web-based seminar, a presentation, lecture, workshop or seminar that is
transmitted over the Web. A key feature of a Webinar is its interactive elements -- the ability to give,
receive and discuss information.” [11].
Table 1. (continued)
Online chat forums

Discussion forums
Workshops
Discussion/queries Instant messenger
Group meetings
Email
Online communities
Online self assessment
Assessment Print test
Web-based test
Email
Print report
Support/feedback LMS
Personal tutorial
e-mentoring
According to Pillenbourg [5] VLE is a designed information space, where

information is designed particularly for educational purposes to be used in
educational interactions, where citing and authoring as well as storing,
maintaining and sharing information are key terms. He argues that a VLE
becomes a social space immediately that it is populated. The members with the
activities of sharing views, objects, commenting and discussing create a learning
environment since there is social interaction arising. Information exchange that is
communication can be formal and informal, synchronous and asynchronous in a
virtual learning environment each supporting different purposes [8].
Table 2. Computer-based Communication Tools [9]
Communication Tools Synchronous Asynchronous:

Formal Video conferencing Discussion board
Audio conferencing Streaming video, audio
Web conferencing narrated slideshows
web-based training
digital libraries
databases
e-books
survey and polls
shared calendars
web site links
Informal Chat Blogs
Instant messages email
Skype Facebook
Facebook forums
An information space enriched by social interaction turns into a social place

states Munro et al [10] namely “settings in which people interact”. As Dourish
argues, “while spaces take their sense from configuration of brick, mortar, wood
and glass, places take their sense from configurations of social actions. Places
provide what we call appropriate behavioral framing” [3]. On projecting the
student-student and student-teacher interactions into this framework both teachers
and students take on the roles and characteristics of their society in an entirely
natural way. They do this without being conscious about it, namely the process is
not consciously controlled. Web 2.0 applications, the booming of Facebook,
Second Life and other virtual communities and social networking applications
200 A. Tick
force both teachers and students to change their attitude and approach to teaching
and learning, respectively. As a result, an entirely new, interactive authentic study
environment can be created that to a great extent supports an autonomous,
individual type study form, nevertheless enables the development of group and
class level courses, and for students it gives the feeling of belonging to a coven.
This community regarding its aims and objectives is an entirely self-organizing,
self-cleaning, homogenous community that to a great extent might back the study
process through the evolving positive team spirit.
A further element in defining VLE is that the above mentioned virtual space is
explicitly defined and represented where the focus is on the activities and tasks
that students conduct in this representation. The balance between the content and
the graphical design is to be found not to distract students’ attention from the core
objective of the existence of the virtual learning environment namely the task of
learning and knowledge gain.
Registering in a VLE and enrolling in a course implies the acceptance of the
game rules. The participants – students as well as teachers - are also actors in the
“creative theatre” where students need to learn more about acting. With the
activities and the active participation students as well as teachers fill the
framework with content and build a special virtual environment existing for their
special purpose. The methodologies like task-based, project-based or problem-
based learning can be excellently applied in such virtual learning environments.
A developed VLE package usually include curriculum mapping (structured
sections of the curriculum that are assigned and assessed), student registration and
tracking, online support for both teacher and student and electronic
communication (e-mail, discussions, chat). The package also offers extra Internet
links to outside curriculum resources. Teachers and students are identified,
teachers see what students see, but teachers have additional user rights to create or
modify content and track student performance.
2.3 An Extended VLE Option – The 3D VLE
Developments in computer hardware and software resources especially in

graphical design and computer animation can enrich a virtual learning
environment by creating an illusion of being immersed in a 3D space, being
surrounded by objects, giving the feeling of being part of the virtual world. Virtual
Learning Environments (VLE) thus creating the possibility for the emergence of
3D Personalized Learning Environments.
This further VLE development is a three-dimensional online space where a
virtual person, a so-called avatar represents a student. The term virtual refers to an
online, internet, or web-based component, the learning component differentiates
VLE for educational purposes from other virtual environments and the term
environment is simply a location/place in which users can gather together in a
social context. The avatar that can be parameterized determines the personality,
the presence, the place and the interaction of a student in the virtual space within
the VLE: students can communicate with each other, with the teacher/tutor and
with Computer-based Agents in real time. They can use the digital documents,
videos and audio files loaded in the system.
In the 3D VLE students can travel around 3D worlds, walk on the campus,
enter the library, look around, and enter a classroom, or even follow a physical
experiment. A special characteristic of 3D virtual systems is the use of the avatar,
which helps users represent themselves or be creative and hide behind a character
to remain unknown, but which helps to project users’ wishes, so we can create and
form such a character that we would like to be, and we can define and decide what
kind of environment and circumstances we would like to live in. Each avatar must
have its own personal account that maintains personal information for assessment
purposes while students’ personal identification is not shown to everyone in the
game world. The colorization of the everyday life, the access of an arbitrary
virtual environment gives the success of Second Life. Thus the idea comes
naturally that if either mountain climbing without any danger or even going to
different places is possible then why couldn’t we go to university and explore the
world of knowledge and gain new and interesting knowledge?
Students i.e. the avateers must sign for some courses and the tutors authorize
these avateers to join and complete the courses. The problem arises at
identification, namely, how to identify the person behind the avatar. Here again
the personal face-to-face meeting cannot be avoided at least at the launch of a
course and testing of identification periodically is a must. Should exams or tests
start at a time but not at the same place in a virtual learning environment,
identification can be solved on the spot for instance with the help of video
connection or some extra authorization process. However, if students reveal their
identity than the key feature of the avatar disappears the avatars will be known to
the tutor and students cannot hide behind them. Another question arises whether
the tutors should reveal their identity or remain unknown to the students. A further
question again is whether students following a virtual course are equally trained
and skilled in computers whether their computer literacy is high enough to be
successful in online courses. It might happen that not all the enrolled students feel
comfortable in online courses, just think of extrovert or introvert personalities [2],
which means that tutors should provide an alternative way of teaching thus turning
back to the conventional classroom and supplement the offered online courses.
The first 3D type highly successful initiative was the development of the
system at the University of Texas using Second Life. On the virtual area bought
by the university all the 16 campuses of the university were realized. These
“private islands are a great way to maintain a safe and controlled learning
environment within a world with almost infinite possibilities”. [6] See Figure 1.
Other well-known universities also decided to enter Second Life and build virtual
universities. See Figure 2.
202 A. Tick
Fig. 1. 3D Virtual Education Centre at Texas University in Second Life
Fig. 2. Island Map of Open University in Second Life
There is always a learning management system behind each 3D virtual learning

environment since educational, administrative and application requirements such
as student registration, course administration, content display must be maintained.
Such systems are capable of educational process automation. Competence
management, knowledge level analysis, progress plan, the possibility of virtual
classes and resource assignment (place, room, book, teacher/tutor, etc) are all
among the services of the most comprehensive learning management systems
providing the highest number of services [12]. Most of the systems are self-service
ones at the student interface, they enable self-registration thus ensures course
accessibility. On the one hand a learning management system is capable of
managing registered students, on the other hand it can handle contents and
information, furthermore, it is capable of assigning these two to each other, thus it
can display learning materials and provide courses in 3D graphical design.
3 The Effects of Virtual Learning/Teaching

After getting to know some features and characteristics of virtual learning
environments let us examine whether these Virtual Learning Environments have
added value and if they do what kind of added value they might have or what
pitfalls might be hidden in such VLEs. Certainly, in order to conduct exact
examinations, long time data collection and control groups are needed based on
which valid and reliable statistical analyses can be pursued. These facts are still
not available due to the young age of the initiatives; however, some conclusions
can be drawn from experience.
The previous paragraphs have already listed some benefits and drawbacks of
virtual learning environments however there are several questions that can be
raised linked to the situation. It can be stated that due to the appearance of 2D and
by now 3D learning environments the methods of teaching and learning are
changing, skills and competences that were taught separately some decades ago
are now taken for granted. Computer literacy is highly evaluated and is a must in a
virtual learning environment. In 2D as well as 3D virtual learning environments
the visual effects of movies and simple pictures have regained their place in the
cognitive processes, since a written or printed text is not interactive; there is no
visual metaphoric support [13] while, on the other hand, it was already proved that
people think first in pictures then in words. As a result, virtual learning
environments can highly support cognitive processes and conduce a lot to
learning. Simple texts only record and describe relationships between facts but
visual representations can even present how to deal with situations and/or issues
[2]. Nyíri argues that pictures in a printed text are always in a subordinate role
compared to the text itself implying that theoretical knowledge dominates, i.e.
presides over practical knowledge. In this century where skills and competencies
are highly appreciated and theoretical knowledge must be backed by practical
skills and competences the use of 3D virtual learning environment is highly
advantageous and is welcome. In such circumstances practical knowledge would
be highly trained while providing theoretical knowledge with the help of e.g. e-
books and digital documents. In this way of learning human visual as well as
audio senses will be called in the learning process. The length of a digital text also
raises some doubts. How long can a text be that it can be taken in? In any virtual
environment only a chunk of a text can be displayed while in a printed version
students can have a grasp of the whole text. As a result the development of digital
learning materials should be done in a way that it avoids lengthy and complicated
texts. Illustrations, well designed diagrams, synopses incorporated in well-
structured written e-documents can highly support understanding and learning
since they deliver meaning and understanding to the students. Furthermore, the
digital generation’s computer literacy is highly developed; its members easily
navigate in the world of hypertexts and hyperlinks.
Another consideration would be that the conventional pedagogical principles
lose their value, the conventional pedagogical techniques like classroom
management, student handling, personal contacts seem to disappear which means
that new ways and techniques must be found and introduced in virtual classroom
204 A. Tick
management. In the conventional way of classroom management the motivation,

the question of participation, discipline problems can be solved on the spot but
how students can be motivated from a distance? Even if video conferencing is
used in a virtual course, if there is no real world meetings and communication,
motivation will be lost. Certainly classroom registration, recording of students’
progress are much more documented and convenient in virtual learning
environments. In conventional classrooms teachers can monitor and control
student progress and their “concentration” on the task and beware of distraction.
In 3D virtual environments such learning environments should be created that
students do not get distracted due to the brilliant graphical effects and will
definitely concentrate on the task they are doing. So such rich contents must be
developed that can attract students and they will be involved in learning
seamlessly and with enjoyment. When developing learning materials, not only the
content must be focused on but the outcomes and definitely authenticity as well.
Virtual learning environments are called a social place but does digital
communication have all the added value that a personal face-to-face
communication has? Does the electronic form of communication ensure all the
social, emotional and communicative features that personal communication does?
Communication in virtual environments can ensure the social informative and
emotional background of communication only if it leans on earlier personal
meetings and communication [2] and personal face-to-face meetings happen
regularly. This also justifies that effective learning is pursued if conventional
(including personal face-to-face meetings) and virtual learning environments are
formed into an organic learning environment each having its place and tasks in the
learning process. How can we deliver the traditional values and morals when
electronic communication enhances alienation, individualism and avoids personal
contacts? How can the rules and models of social behavior be taught in 3D virtual
learning environments? We should not forget that individual and social learning
have a complex and necessary interdependence [14]. Or should we adapt to the
new situation and accept that students will take on behavioral patterns that they
learn in the virtual environment?
4 Conclusion
Considering higher education as the target group of our interest it must be stated
that “at least 300 universities around the world teach courses and conduct research
in Second Life.” as Claudia L'Amoreaux at Linden Lab states in [6]. Should we
consider higher education institutions as sites of knowledge management and as
knowledge intensive service providers in parallel with the fact that the role of
knowledge-based services is growing [15] the appearance of these institutions
seem reasonable in 3D virtual learning environment where apart from knowledge
transfer from the higher education institutions and students, students themselves
can share and distribute the knowledge gained. This implies that developing 3D
VLEs in higher education might be advantageous and gives an added value to the
existing delivery of courses. In higher education the learning environment and
progress must not be very closely and strictly monitored and controlled since
participants are grown up people whom supposedly enroll for university courses
for the sake of learning. 3D virtual learning environments extends the social place
and ensures such learning environments where students can form their learning
environment which deserves computer literacy but trains it at the same time.
Virtual environments compared to traditional learning support environments might
have a much deeper impact on students since visual representation is a primary
source of information gaining. This means that these systems bring much higher
added value and benefit for the future generation in the learning process. It can be
concluded that “the 3D VLE experience is designed to serve as a reinforcement of
the concepts learned in the traditional lessons” [6] and not instead. 3D virtual
learning environments will not displace the conventional form of teaching but will
extend it.
References
[1] McLuhan, M.: The Gutenberg Galaxy, The Making of Typographic Man. University
of Toronto Press, Toronto (1962)
[2] Nyíri, K.: Virtuális Pedagógia – A 21. század tanulási környezete (June 17, 2009),
http://www.ofi.hu/tudastar/iskola-informatika/
nyiri-kristof-virtualis (accessed January 20, 2012)
[3] Dourish, P.: Where the footprints Lead: Tracking down Other Roles for Social
Navigation. In: Munro, A., Höök, K., Benyon, D. (eds.) Social Navigation of
Information Space, pp. 15–34. Springer, London (1999)
[4] Längin, D.W., Ackerman, P.A., Lewark, S.: Internet-based learning in higher forestry
education. Unasylva 55(216), 39–46 (2004)
[5] Dillenbourg, P.: Virtual Learning Environments. In: EUN Conference 2000 (2000),
http://tecfa.unige.ch/tecfa/publicat/
dil-papers-2/Dil.7.5.18.pdf (accessed February 10, 2009)
[6] Barkand, J., Kush, J.: GEARS a 3D Virtual Learning Environment and Virtual Social
and Educational World Used in Online Secondary Schools. Electronic Journal of e-
Learning 7(3), 215–224 (2009), http://www.ejel.org
[7] http://whatis.techtarget.com/definition/0,sid9_gci866691,
00.html (accessed January 25, 2012)
[8] Hrastinski, S.: Asynchronous and Synchronous E-Learning Educause Quarterly 31(4)
(2008)
[9] http://www.asaecenter.org/Resources/articledetail.cfm?ite
mnumber=13572 (accessed January 10, 2012)
[10] Munro, A., Höök, K., Benyon, D.: Footprints in the snow. In: Munro, A., Höök, K.,
Benyon, D. (eds.) Social Navigation of Information Space, pp. 1–14. Springer,
London (1999)
[11] Valiathan, P.: Blended Learning Models (2002),
http://www.learningcircuits.org/2002/aug2002/
valiathan.html (accessed September 12, 2006)
206 A. Tick
[12] Gábor, A.: e-learning oktatasmenedzsment keretrendszerek (Learning Management

Systems) (2003), http://www.oktopusz.hu/domain9/files/modules/
module15/271CFD97067E893.pdf#search=%22ny%C3%ADlt%20forr%
C3%A1sk%C3%B3d%C3%BA%20tanulm%C3%A1nyi%20keretrendszer%22
(accessed September 22, 2006)
[13] Platon: Phaidrosz in Platon összes művei, 2nd edn., Budapest (1984)
[14] Stiles, M.J.: Effective Learning and the Virtual Learning Environment. In:
Proceedings EUNIS 2000 – Towards Virtual Universities, Poznan, Poland (2000),
http://www.staffs.ac.uk/COSE/cose10/posnan.html
(assesses January 15, 2012)
[15] Farkas, F., Király, Á.: What makes Higher Education Knowledge-Compatible? Acta
Polytechnica Hungarica 6(3), 93–104 (2009)
[16] Tick, A.: A new direction in the Learning Processes, the Road from eLearning to
vLearning. In: Proccedings of 6th IEEE International Symposium on Applied
Computational Intelligence and Informatics, Timisoara, Romania, pp. 359–362
(2011)
[17] Jones, S.: The Internet goes to College. In: Pew Internet and American Life Project
2002, Washington D.C (2007),
http://www.pewinternet.org/pdfs/PIP_College_Report.pdf
(accessed September 20, 2007)
[18] Gáti, J., Kártyás, G.: Virtual Classrooms for Robotics and other Engineering
Applications. In: Proceedings of 19th International Workshop on Robotics in Alpe-
Adria-Danube Region, Budapest (2010)
[19] Gáti, J., Kártyás, G.: New Trend in Teaching and Learning Environments for
Distance Higher Education. In: Proceedings of 5th International Symposium on
Applied Computational Intelligence and Informatics, Timisoara, Romania (2009)
[20] Lim, C.P.: Spirit of the game: Empowering students as designers in schools? Br. J. of
Educ. Technol. 39, 996–1003 (2008)
[21] Rossett, A., Douglis, F., Frazee, R.V.: Strategies for Building Blended Learning
(2003), http://www.learningcircuits.org/2003/
jul2003/rossett.htm (accesses September 01, 2006)
Weights Aging in Social Networks
Gabriel Tutoky and Ján Paralič
Dept. of Cybernetics and Artificial Inteligence, Faculty of Electrical Engineering and

Informatics, Technical University of Košice, Letná 9, 040 01 Košice, Slovakia
{gabriel.tutoky,jan.paralič}@tuke.sk
Abstract. This paper describes basic approaches for modeling collaboration social
networks. Some of the new extensions are described – new approach for weight-
ing of the ties among event participants; and temporal based method for modeling
of network evolution – aging of the ties among actors in passing time. We eva-
luate proposed methods by two experiments with the real data set and we compare
our modeled network with the reference data gathered from people’s opinions
expressed in targeted inquiries.
Keywords: social network analysis, collaboration networks, networks modeling,

network projection.
1 Introduction
In recent years, many “social networks” have been analyzed like various Internet
communities, email networks, peer-to-peer networks, telephone call graphs or
train routers [1]. All of these networks are interesting in some of their specific
aspects and they provide a notable data source for network analysis. There are
usually large-scale networks with thousands of nodes and edges. Analysis of these
networks, usually based on global properties, can lead to interesting and helpful
results. Nevertheless, there exist many different situations in network analysis
where data used for analysis of these networks did not carry sufficient informa-
tion, e.g. temporal information is often neglected in these analyses. In this paper
we describe a new approach how to model and analyze one particular type of so-
cial networks, affiliation networks, making use of more strands of additional in-
formation, including the temporal one.
One of interesting types of social networks are affiliation networks. An affilia-
tion network is a network of actors connected by common memberships in
groups/events such as clubs, teams, organizations or other activities. Affiliation
networks are special type of two-mode social networks [2] where one mode is a
set of actors, and the second mode is a set of events which are affiliated to the
actors. The tie between actors and events are created, if actor is member of a
group/participates on particular event. Affiliation networks describe collections of
actors rather than simply ties between pairs of actors. Based on such an affiliation
network we are able to derive connections among members of one of the modes
based on linkages established through the second mode [2].

208 G. Tutoky and J. Paralič
Affiliation networks were studied in past, e.g. studying attendance of women in

social events [3], movies and their actors [4] or co-authorship network of scientists
and their papers [1]. Whereas in the first two examples, the authors used un-
weighted representations of the networks, in the last work, the author introduced
interesting approach for building of collaboration network where he used weighted
ties between authors of the same paper. The weight of the tie between collaborated
authors of a single paper is derived from count of the paper collaborators, and
final weight of two collaborated authors is a sum of weights over all papers where
authors collaborated. This approach allows finding the “most connected” scientists
in the whole collaboration network.
In our work we build collaboration network of teenagers (the necessary back-
ground details are described in next section 2) based on their participations on
educative-pedagogic workshops for evaluating the “most important” persons in
desired time stamp. In network building process we used two different approach-
es: 1) modified weighting of the ties between collaborators based on weighting
used by Newman in [1, 5] and 2) our weighting described in details in section 2.2.
Both of these weightings are next evaluated taking into account also the temporal
aspects. We proposed time dependent weights decreasing over time (see section
2.3). We assume that weight of two collaborators is dependent on the number of
events’ participants and on the time interval between events where these actors
collaborated together. Weight decreases with increasing number of event’s partic-
ipants and also with increasing length of the time interval between two common
events.
2 DAK Collaboration Network

1
DAK – community network is a collaboration network of members (usually tee-
nagers) of a non-profit organization dealing with organizing educative-pedagogic
workshops for young people [6]. Usually there are organized around 10 work-
shops annually with 150 – 700 participants on single workshop. The number of
participants depends on workshop’s type and duration. All participants of a single
workshop are partitioned into smaller groups, usually with 8 – 12 members. Each
group member cooperates with other group members and so there are established
new social relations between group members. Generally there are two types of
group members – group participants and leader(s) of a group. Participants are
spending most of the time inside the group (i.e. they usually do not establish social
relations outside the group), but leader cooperates with other leaders and so they
create another type of social relations. Additionally we recognize two types of
groups – participants’ group and organizers’ group. In summary we have two
types of groups and four types of group members: base participant, base organiz-
er, leader of participants’ group and leader of organizers’ group.
Compositional attributes in DAK data set are available for both, actors and
events. Actors are described by attributes such as date of birth (age) and gender; as
well as by geographical attributes – city or area of living. Events are described by
1
http://www.zksm.sk/
Weights Aging in Social Networks 209
their type. We can recognize two main types of events – events for base partici-
pants and events for organizers. Events for organizers are next categorized by their
types of activity like registration, security or accommodation event.
Moreover, temporal attributes are available together with compositional
attributes, e.g. start and end of events/workshops. From these data we can derive
several other attributes, such as “length of event” or “time of first visit” for partic-
ular actor. In our case it means the moment when the actor visited any event for
the first time.
3 Collaboration Network Modeling

Collaboration network described above can be expressed for single workshop by
a bipartite graph as representation of two-mode network. The first mode is a set of
actors, and the second mode is a set of events which affiliate the actors. We
represent each group on the single workshop as a single event, so for one work-
shop we obtain several events. Additionally we added two more events for repre-
sentation of cooperation between leaders of participants and leaders of organizers.
One of the advantages of DAK data set is availability of temporal information
in the data. We are able to track events in the time and recognize which events
were organized in parallel and which sequentially. Also we are able to track par-
ticipation of single actors on particular events.
2.1 Base Projection Onto One-Mode Network

Affiliation networks (two-mode representation) are most often projected onto one-
mode networks (see figure 1) where actors are linked to one another by their affil-
iation with events (co-membership or co-operation), and at the same time events
are linked by the actors who are their members (overlapping events). This proper-
ty is called duality of the affiliation network [2].
a1 e2 a1
e3 a4
e1
a3 e5 a3 a6
e4 a2
e6
a5 a5
Fig. 1. Projection of two-mode network ( – actors, – events) onto one-mode network

from actors view
Usually, weights in both, affiliation (two-mode) and also in projected (one-

mode) networks have binary values. The ties in the networks exist or not. In the
step of projection of two-mode networks onto one-mode networks we can use
different measures for weight definition, e.g. the ones summarized by Opsahl in
[7]:
- Weight is determined as a count of participations (co-occurrences) – e.g. the

count of events were two actors participated together, formalized expression
is
∑ 1, (1)
where is the weight between actors (nodes of the first mode) and , and
are events (nodes of the second mode) where and participated together.
- Newman in [1, 5] proposed extended determination of weights while working
with scientific collaboration networks. He supposes that strength of social
bonds between collaborators is higher with lower count of collaborators on
a paper and vice versa social bonds are lower with many collaborators on a
paper. He proposed formula (see formula 2) for defining the weights among
collaborators where is the count of collaborators on paper (event) .
∑ (2)
Till now we considered only binary two-mode networks and their projection to
weighted one-mode networks. However, there exist also weighted two-mode net-
works, such as networks of online forums (weight is determined as count of posts
or posted characters) or collaboration network described above and also in [8, 9].
So, both just presented measures for weight definition could be extended for
weighted two-mode networks as follows:
- ∑ , , (3)
where , is the weight of th actor to th event where and participated to-
gether. This method differentiates how two particular actors interact with the
common event, and projects it onto a directed weighted one-mode network.
[7].
- In a similar way, the Newman’s method can be extended for projecting of
two-mode networks. The weights are calculated by the following formula:
,
∑ . (4)
This formula would create a directed one-mode network in which the out-strength
of a node is equal to the sum of the weights attached to the ties in the two-mode
network that originated from that node [7].
2.2 Extension of One-Mode Projection

In the next two sections we describe our extensions of projection of two-mode
collaboration networks onto one-mode networks. This step – projection of two-
mode networks, has strong impact on analysis of collaboration networks. It is
important step for creation of the most suitable network model by projection onto
one-mode network.
At first, we propose new, more general weighting of the ties created among
event participants as Newman’s weighting method. The reason is that Newman’s
weighting method results in fast decreasing value with just a small increase of
event participants (more than two). This can be good in some cases, but not in
general for any collaboration network. We suggest using an exponential curve:
N
w α (5)
The weights are also here decreasing with increasing number of event participants.
But parameter can be adjusted with respect to particular type of collaboration
network (and in such a way influence the shape of the exponential curve). Parame-
ter depends on collaboration type and it should be estimated by a domain expert
e.g. with the following formula:
β
α √2. (6)
This formula enables easier set up of an optimal value of the parameter for par-
ticular type of collaboration network because is the number of participants
(event size) when weight of collaboration ties decreases 50%. For example in
scientific collaboration network, the strength of collaboration ties among
8 scientists should by weaker (by Newman it is 0,14286), but e.g. in collaboration
network of students or in the DAK network described above, the strength of the
ties among 8 collaborators participating on the same event should be higher, e.g.
0,62. Number 2 used in the index of radical represents an “ideal” number of event
participants, when the strongest ties are created among event participants (this is
analogical to the Newman’s method).
2.3 Time Based Network Modeling

Various collaboration networks contain time series data – usually the time of the
event is known. It is reasonable to assume that the weight of the ties created be-
tween participants of a common event will decrease over the time. So, we propose
time dependent weights in our representation of one-mode projected affiliation
network – a kind of aging of the ties. This should be considered as similar ap-
proach to the one presented in [10, 11] where authors considered aging of the
nodes in the context of citation networks. They describe node’s age as influence to
the probability of connecting current node to new nodes in the network.
Our proposed weight aging method is based on assumption that past collabora-
tions among network members are less important than lately created collabora-
tions. These past collaborations after passing sufficient long time have no more
influence in the present and they are next removed from the network – old ties
(without refreshing) among collaborators are than “forgotten” in such a way. From
the social network analysis point of view our proposal of aging of the edges can
lead to new opportunities in network analysis:
2
We consider interval <0, 1> for weight values where 0 represents weakest and 1
represents strongest tie.
- Tracking collaborations over the time – i.e. tracking of collaboration strength

with passing time among selected actors of the network. This should provide
detailed information describing evolution of cooperation among desired
actors.
Fig. 2. Tracking of single collaboration over the time between actors I and j for two various
ageing factors
- Creation of network snapshots in given time – it allows us to obtain actual

network state in desired time and consequently to analyze e.g. strongest colla-
borations in the network. It can lead to different results of network analysis
because we do not consider older collaborations so high like last created. In
collaboration network we are able to “view” still actual and (by our confi-
dence) important collaborations among network members.
We have investigated humans’ forgetting curve described by Ebbinghaus in [12]
which has exponential progress for possibility of using it for aging of edges
(i.e. for decreasing of their weights) with passing time. Forgetting curve can be
described as
, (7)
where is memory retention, is the relative strength of memory, and is time.
By [12, 13] the forgetting curve decreasing rate depended on repetition of memory
– in collaboration network context it is repetition of collaborations among actors.
If actors collaborate together frequently (the collaboration strength grows quickly
in short time) the edge aging is slower than aging in case of just one (or very spo-
radic) occurrence of mutual collaboration. Also we have investigated similar
works with aging of nodes where in [11] authors studied aging of the nodes in
scientific collaboration networks and they derive exponential progress of network
aging – similar to the forgetting curve (formula 7).
In network modeling process we propose the use an exponential curve for mod-
eling of edges aging described by formula:
∆ ∆
, if
∆ (8)
0, otherwise
where w t is weight of the collaboration in the time and w t ∆t is the

weight after ∆t time left after the last collaboration in time among actors and .
Value is threshold value of minimal collaboration weight and factor is called
ageing factor which designates “ageing speed”. It is described by , where
/
/ is the time when weight of the ties decreases to 50% by aging process.
2.4 Network Modeling Process

We decompose the network modeling process into the following steps:
- Creation of two-mode network – from available real data we created affilia-
tion network.
- Projection of two-mode network onto one-mode network using the follow-
ing alternative weighting schemes:
- Simple weighting – each collaboration on an event has value 1
- Newman’s weighting – collaboration strength is derived from number of
event participants by formula (see equation 2).
N
- Our proposed weighting – collaboration strength is derived from number
of event participants by means of equation 5; for we used value 1,04.
- Solitary network modeling over the time:
- Simple summation over all collaborations – collaborations computed in
step before are now summed – see equation 1 for simple weighting case
and eq. 2 for Newman’s weighting.
Aging of edges – simulation of network edges aging, we created 24 network snap-
shots, each one for the time of a workshop and we derived collaboration strength
before and after the workshop. Collaboration strength between two selected actors
is depicted on the figure 2 for different ageing factors.
3 Evaluation
We implemented all methods for projection of two-mode networks onto one-mode
networks presented above and we evaluated them for both variations of weighting of
the ties – for simple summing of all collaboration weights; and also for aging of ties
(collaborations) with passing time. We used two different experiments. In the first
one we looked for most important persons within given collaboration network.
3.1 Evaluation of the Most Important Persons in the Network

In order to evaluate which of implemented approaches models best the reality in
considering most important persons, we used for comparison data gathered by
means of targeted inquiry from 16 respondents who are actual members of the
analyzed collaboration network. We selected such members of the network who
know the network structure very well since a longer time period and follow activi-
ties of its members trough organized workshops. Each of these respondents had to
create a list with 30 most important persons in the network by his/her opinion. The
goal was also to sort created list from most important to less important persons. As
a result of this inquiry we obtained from all respondents altogether a list of 90
distinct actors which were mentioned 1 – 15 times by our respondents. For our
evaluation we filtered this list for persons who were mentioned at least 4 times and
next we ordered this list by average value of actor’s position assigned to him/her
by respondents.
On the other side, for each particular model of collaboration network we ob-
tained list of top 35, 30, 25 and 15 actors by Hubs and Authorities analysis [14,
15]. We compared these lists with results from our inquiry. For each list size we
evaluated the quality of estimation of most important actors by particular model of
collaboration network (see figure 3). We first simply computed intersections be-
tween these lists (one gathered as a result from the inquiry process described
above and the other one by calculation of Hubs and Authorities in case of particu-
lar collaboration network model) and expressed it in percentage of the whole. The
results are graphically presented in figure 3.
This experiment confirmed our assumption that projection of two-mode net-
work based on weighting with constant value 1 (formula (1)) cannot provide suffi-
cient model of collaboration network (see dark columns in figure 3). On the
second hand, this experiment showed unexpectedly high precision of results for
Newman’s weighting of collaboration ties for simple summing of weights over the
time (see all left middle gray columns). In this case we expected better results for
Fig. 3. Evaluation results of the most important actors in the network. Three weighting
types – simple, Newman’s and our proposed weighting are distinguishable by gray tone.
Left columns of the same tone displaying result for simple summing of weights over the
time; whereas rights columns displaying results with aging of ties over the time.
Newman’s weighting than constant weighting, but we also expected higher preci-
sion of our proposed weighting. Our expectation was validated in case of aging of
the ties where it has better results than Newman’s weighting (see lightest gray
columns). In case of aging of the ties, our proposed weighting has better results
than Newman’s weighting, especially for identifying 15 most important actors.
3.2 Evaluation by Graph Isomorphism

In the second experiment we evaluated implemented network models making use
of different inquiry data obtained from 46 members of analyzed collaboration
network. Each of the respondents expressed his/her collaboration strength to mi-
nimally thirty other network members. The structure of inquired network members
was the following: members with potentially strong or strongest collaboration,
members with potentially middle or weak collaboration and finally members
without reciprocal collaboration. In such a way we have obtained real picture in
form of a balanced collaboration network with 2278 ties between 828 distinct
actors with their collaboration strength. Strength was partitioned into five nominal
categories: No collaboration, Weak collaboration, Middle collaboration, Strong
collaboration, Very strong collaboration.
This real collaboration network created from data gathered in the inquiry was
used as a reference network. As next we compare this reference network with
equivalent parts of modeled networks gained from the DAK system. We norma-
lized weights of all networks into 0, 1 interval for correct evaluation and we also
transformed five categories of collaboration strength into five regular intervals:
0, 0.2 , 0.2, 0.4 … 0.8, 1 . By classification weights of modeled networks into
these intervals we obtained precision of estimated weights by modeled networks,
results are depicted on figure 4.
Fig. 4. Evaluation of modeled networks (F1 measure) in comparison to reference network

gathered by inquiry – the classification accuracy into 5 intervals of collaboration’s strength
is compared a) for simple summing with passing time; b) for aging
We are able to observe several interesting results: i) methods that use summing
of weights with passing time have different results for each used weighting
method. For simple weighting we achieved the worst results (as was expected,
because summing of weights takes high values and after normalization, all weights
falls into interval of strongest ties – we cover all of these collaborations (precision
is 100%), but with weak recall). Newman’s weighting method assigned relatively
small values to all identified collaboration weights (because average event size is
relatively high – 10 and more participant) and it resulted to good identification of
weak weights but very weak identification of strong weights. Proposed ratio
weighting resulted into most balanced results for all intervals of collaboration
weights, but with relatively weak values (from 11% to 41%). ii) In case of aging
we observed better results. For simple weighting we are able to identify also
weights of other intervals than is the interval of very strong collaborations; New-
man’s weighting achieved better precision to weak but to worst precision to strong
weights; and proposed ratio weighting resulted to the best balanced results again,
but now with higher values (from 18% to 54%).
By detailed analysis of results we noticed that in many cases, the distance be-
tween classification of weights from modeled network and the reference network
was pretty small – usually just one interval of collaboration strength (i.e. the method
classified the strength into neighboring interval). Therefore as next we counted the
classification accuracy once more taking into account this fact. All classifications
which had distance 1 and less in comparison to the reference classification we eva-
luated as correct classifications – so we used tolerance to the one neighbor interval.
In figure 5 we can observe significant improvement of our results in general,
which is obvious. In case of summing of weights with passing time, we are able to
predict collaboration strength better for each of collaboration strength. In case of
aging the results are similarly balanced as results without tolerance (figure 4) but
Fig. 5. Evaluation of modeled networks (F1 measure) in comparison to the network ga-
thered by inquiry – the strength of collaborations is compared. Difference in classification
up to one classification interval is considered as acceptable for simple summing with pass-
ing time and for aging.
with significantly higher values. Best balanced result was reached in case of pro-
posed ratio weighting where values of F1 measure for all categories were in the
interval (from 48% to 91%). Ageing of the proposed ratio weighting method re-
sulted in the both evaluation methods (without and with tolerance) to the best
balanced results with the best values in average.
4 Conclusion
In section 2.1 we described various methods for weighting of collaborations
among event participants in collaboration networks and in section 2.2 we de-
scribed one new method of weighting of the ties. In next sections 2.3 and 2.4 we
described our method for modeling networks with passing time, where we pro-
posed method for aging of the ties among collaborators. We next evaluated all
presented methods on data from DAK collaboration network. Both experiments
brought results comparable with existing weighting approaches. Experiments
showed that proposed type of weighting, especially in combination with aging
resulted in best balanced results among all categories.
In our future work we will extend evaluation of different collaboration network
models with considering of trend line in actors collaborations. By this trend line
we will be able to estimate whether collaboration strength between actors increas-
es/decreases or if it is constant.
Acknowledgments. The work presented in this paper was supported by the Slovak Grant
Agency of Ministry of Education and Academy of Science of the Slovak Republic under
grant No. 1/1147/12 (50%) and the Slovak Research and Development Agency under the
contract No. APVV-0208-10 (50%).
References
[1] Newman, M.E.J.: Who is the best connected scientist? A study of scientific coauthor-
ship networks. Complex Networks (2004)
[2] Wasserman, S., Faust, K.: Social Network Analysis. Cambridge University Press,
Cambridge (1994)
[3] Davis, A., Gardner, B.B., Gardner, M.R.: Deep South. A social Anthropological
Study of Caste and Class. University of Chicago Press (1941)
[4] Watts, D.J., Strogats, S.H.: Collective dynamics of ’small-world’ networks. Nature
(1998)
[5] Newman, M.E.J.: Scientific collaboration networks. II. Shortest paths, weighted net-
works, and centrality. The Amarical Physical Society 64 (2001)
[6] DAK - Collaboration Network, data set of non-profit organization (2011),
http://www.domcek.org
[7] Opsahl, T.: Projecting two-mode networks onto weighted one-mode networks (2009)
[8] Tutoky, G., Paralič, J.: Modelovanie a analýza malej komunitnej sociálnej siete. In:
5th Workshop on Intelligent and Knowledge Oriented Techniloties, Bratislava (2010)
[9] Tutoky, G., Repka, M., Paralič, J.: Structural analysis of social groups in collabora-
tion network. In: Faculty of Electrical Engineering and Informatics of the Technical
University of Košice, Košice (2011)
[10] Hajra, K.B., Sen, P.: Aging in citation networks. Elsevier Science (2008)
[11] Zhu, H., Wang, X., Zhu, J.-Y.: The effect of aging on network structure. The Ameri-
can Physical Society (2003)
[12] Ebbinghaus, H.: Memory: A Contribution to Experimental Psychology. Teachers
College, New York (1885)
[13] Savara, S.: The Ebbinghaus Forgetting Curve – And How To Overcome It,
http://sidsavara.com/personal-productivity/
the-ebbinghaus-curve-of-forgetting
[14] Kleinberg, J.M.: Authoritative Sources in a Hyperlinked Environment. In: ACM-
SIAM Symposium on Discrete Algorithms (1998)
[15] Batagelj, V., Mrvar, A.: Pajek - Program for Analysis and Visualization of Large
Networks, Ljubljana (2009)
[16] Tutoky, G., Paralič, J.: Modeling of Collaboration Social Networks Including Tem-
poral Attributes. In: Proc. of the 15th IEEE International Conference on Intelligent
Engineering Systems, Poprad, Slovakia (2011) ISBN: 978-1-4244-8955-8
Part IV
Methodologies, Methods
and Algorithms
The HOSVD Based Canonical Form
of Functions and Its Applications
András Rövid1, László Szeidl2, and Péter Várlaki2

1
Óbuda University, John von Neumann Faculty of Infomatics, 1034 Budapest,
Bécsi út 96/B, Hungary rovid
andras@nik.uni-obuda.hu
2
Széchenyi István University, System Theory Laboratory, Egyetem tér 1.,
9026 Győr, Hungary
{szeidl,varlaki}@sze.hu
Abstract. The paper deals with the theoretical background of the higher order
singular value decomposition (HOSVD) based canonical form of functions. Fur-
thermore in special case it describes the relation between the canonical form and
the Hilbert-Schmidt type integral operators. The described techniques have a
variety of applications, e.g. image processing, system identification, data compres-
sion, filtering, etc. As an example of application from the field of intelligent sys-
tems, a tensor-product based concept is introduced useful for approximating the
behavior of a strongly non-linear system by locally tuned neural network models.
The proposed approach may be a useful tool for solving many kind of black-box
like identification problems. The weights in the corresponding layers of the input
local models are jointly expressed in tensor-product form such a way ensuring the
efficient approximation. Similar concept has been used by the authors for approx-
imating the system matrix of linear parameter varying systems in state space
representation. We hope that the proposed concept could be an efficient compro-
mised modeling view using both the analytical and heuristic approaches.
1 Introduction
Many methods have been proposed to deal with multi input, multi output systems,
by the literature. There exist several types of nonlinear models, i.e. black box
models, block structured models, neural networks, fuzzy models, etc. [15]. The
effectiveness of these models may further be improved by combining the analytic
and heuristic approaches.
Linear parameter varying (LPV) structure is one by which non-linear systems
can be controlled on the basis of linear control theories. As another frequently
used approach to approximate dynamic systems the Takagi-Sugeno fuzzy
modelling can be mentioned. This interest relies on the fact that dynamic T-S
models are easily obtained by linearization of the nonlinear plant around different
operating points [12]. Beyond these nonlinear modelling techniques, the neural
network-based approaches are highly welcome, as well, having the ability to learn

222 A. Rövid, L. Szeidl, and P. Várlaki
sophisticated nonlinear relationships [13][16]. The tensor product (TP)

transformation is a numerical approach, which makes a connection between linear
parameter varying models and higher order tensors ([9],[8]). The approach is
strongly related to the generalized SVD, i.e. so called higher order singular value
decomposition (HOSVD) [5], [14]. One of the most prominent property of
the tensor product form is its complexity reduction support [11]. As described in
the upcoming sections, the HOSVD based canonical form is strongly related to the
Hilbert-Schmidt type integral operators, as well. It is shown that the determination
of the canonical form in two variable case is strongly connected to the classical
results of E. Schmidt related to the integral equations which in general form are
described by the theory related to the so called Hilbert-Schmidt type operators
defined on Hilbert spaces.
Furthermore we will describe how a set composed of multilayer perceptrons
(MLPs) may be jointly expressed in tensor product form and how such a
representation of MLPs may support the identification of parameter varying
systems [17].
The paper is organized as follows: Section 2 gives a closer view on how to
express a multidimensional function by using polylinear functions on HOSVD
basis and how the HOSVD based canonical form can be obtained, Section 3 shows
the special properties of the canonical form, i.e. its relation to the Hilbert-Schmidt
type integral operators. Section 4 shows how the described techniques may be
applied to jointly express multiple MLPs as local models in the parameter space in
tensor product form and finally future works and conclusions are reported.
2 HOSVD Based Numerical Reconstruction of n -Variable

Functions
The approximation methods of mathematics are widely used in theory and practice
for several problems. If we consider an n-variable smooth function
f (x), x = ( x1 ,..., xN )T , xn ∈ [ an , bn ] , 1 ≤ n ≤ N ,
then we can approximate the function f(x) with a series

I1 IN
f (x) = ...  α k ,..., k p1, k ( x1 ) ⋅ ... ⋅ pN , k ( xN ). (1)

1 n 1 N
k1 =1 k N =1
where the system of orthonormal functions pn,k ( xn ) can be chosen in classical

n
way by orthonormal polynomials or trigonometric functions in separate variables
and the numbers of functions In playing role in (1) are large enough. With the help
of Higher Order Singular Value Decomposition (HOSVD) a new approximation
method was developed in [7] in which a specially determined system of orthonor-
mal functions can be used depending on function f(x), instead of some other
The HOSVD Based Canonical Form of Functions and Its Applications 223
systems of orthonormal polynomials or trigonometric functions. Assume that the

function f(x) can be given with some functions v n,i ( xn ), xn ∈ [ an , bn ] in the form
I1 IN
f (x) = ...  α k ,..., k v 1, k1 ( x1 ) ⋅ ... ⋅ v N , kN ( xN ). (2)

1 n
k1 =1 k N =1
I ×...× I N
Denote by  ∈  1 the N-dimensional tensor determined by the elements
αi1,...,iN , 1 ≤ in ≤ I n , 1 ≤ n ≤ N and let us use the following notations (see [5]).
•  ×n U : the n-mode tensor-matrix product,

•  ×nN=1 U n : the multiple product as  ×1 U1 ×2 U 2 ... ×N U N .
Fig. 1. Three possible ways of expansions of a 3-dimensional array into matrices
Fig. 2. Illustration of the higher order singular value decomposition for a 3-dimensional
array. Here S is the core tensor, the U l -s are the l-mode singular matrices.
The n-mode tensor-matrix product is defined as follows: Let U be a K n × M n

matrix, then  ×n U is a M 1 × ... × M n −1 × K n × M n +1 × ... × M N tensor for which the
relation
def
( ×n U) m ,..., m
1 n −1 , kn , mn +1 ,..., mN
= 
1≤ mn ≤ M n
am ,..., m
1 n ,..., m N
Uk
n , mn
holds. Detailed discussion of tensor notations and operations is given in [5]. Using
this definition the function (2) can be rewritten as a tensor product form
f (x) =  ×nN=1 v n ( xn ), (3)
where v n ( xn ) = (v n,1 ( xn ),..., v n, In ( xn )) , 1 ≤ n ≤ N . Based on HOSVD it was

T
proved in [9] that under mild conditions the (3) can be represented in the form
f ( x) =  ×nN=1 v n ( xn ), (4)
r ×...× rN
where  ∈  1 is a special (so called core) tensor with the properties:
1. rn = rankn ( ) is the n-mode rank of the tensor  , i.e. rank of the linear space
spanned by the n-mode vectors of  :
{(ai ,...,i ,..., ai ,...,i )T :1 ≤ i j ≤ I n , 1 ≤ j ≤ N },

1 n −1 ,1, in +1 ,..., iN 1 n −1 , I n ,in +1 ,..., iN
2. all-orthogonality of tensor  : two subtensors i and i (the n-th indi-

n =α n =β
ces in = α and in = β of the elements of the tensor  keeping fix) orthogonal

for all possible values of n, α and β : i , i = 0 when α ≠ β . Here
n =α n =β
the scalar product i , i denotes the sum of products of the appropri-

n =α n =β
ate elements of subtensors i and i

n =α n =β
3. ordering i ≥ i ≥  ≥ i > 0 for all possible values of n

n =1 n =2 n = rn
( i = i , i denotes the Kronecker-norm of the tensor i ).

n =α n =α n =α n =α
4. Components vn ,i ( xn ) of the vector valued functions
v n ( xn ) = (vn ,1 ( xn ),..., vn , r ( xn ))T , 1 ≤ n ≤ N

n
are orthonormal in L2-sense on the interval [an , bn ] , i.e.

b
∀n :  n vn ,i ( xn )vn , j ( xn ) dx = δ i , 1 ≤ in , jn ≤ rn ,
an n n n , jn
where δ i , j is a Kronecker function ( δ i , j = 1 , if i = j and δ i , j = 0 , if i ≠ j )
The form (4) was called in [9] HOSVD canonical form of the function (2). Let us
decompose the intervals [an , bn ] , n = 1..N into Mn disjunct subintervals Δn,m ,
n
1 ≤ mn ≤ M n as follows:
ξn,0 = an < ξn,1 <  < ξn, Mn = bn , Δn, mn = [ξn, mn , ξn,mn −1 ).

Assume that the functions vn , k ( xn ), xn ∈ [ an , bn ] , 1 ≤ n ≤ N in the equation (2) are

n
piece-wise continuously differentiable and assume also that we can observe the
values of the function f(x) in the points
y i ,...,i = ( x1,i ,..., xN ,i ), 1 ≤ in ≤ M n , (5)
1 N 1 N
where
xn, m ∈ Δn,m , 1 ≤ mn ≤ M n , 1 ≤ n ≤ N.
n n
Based on the HOSVD a new method was developed in [9] for numerical recon-
struction of the canonical form of the function f(x) using the values
f (yi ,...,i ), 1 ≤ in ≤ M n , 1 ≤ in ≤ N. We discretize function f(x) for all grid points as
1 N
bm ,..,m = f (y m ,..,m ).
1 N 1 N
Then an N dimensional tensor  = (bm ,, m ) is constracted from the values

1 N
bm ,..,m . Obviously, the size of this tensor is M 1 × ... × M N . Further, we discretize

1 N
vector valued functions v n ( xn ) over the discretization points xn ,m and construct

n
matrices Vn from the discretized values as
 vn ,1 ( xn ,1 ) vn,2 ( xn ,1 )  vn , r ( xn ,1 ) 
 n

 vn ,1 ( xn,2 ) vn ,2 ( xn ,2 )  vn , r ( xn ,2 ) 
Vn =  n
. (6)
    
 v (x ) vn ,2 ( xn , M )  vn , r ( xn , M ) 
 n,1 n , M n n n n 
Tensor  can simply be given by (4) and (6) as

 =  ×nN=1 Vn . (7)
Consider the HOSVD decomposition of the discretization tensor

 =  d ×nN=1 U ( n ) , (8)
(
where  d is the so-called core tensor, and U ( n ) = U1( n ) U 2( n )  U M( n )
n
) is an
M n × M n size orthogonal matrix ( 1 ≤ n ≤ N ). Let us introduce the notation:
rn = rankn  , 1 ≤ n ≤ N and consider the r1 ×× rN size reduced version
d d d
 d = ( d
 m ,, m ,1 ≤ mn ≤ rn ,1 ≤ n ≤ N ) of the M 1 ×  × M N size tensor  . The
d
1 N
following theorems were proved in [9]. Denote

Δ = max max (ξ n , m − ξ n , m )
1≤ n ≤ N 1≤ i ≤ M n n −1
n n
and
N
ρ = ∏ρ n , ρ n = (bn − an ) / M n .
n =1
If Δ is sufficiently small, then rn = rn , 1 ≤ n ≤ N

d
and the convergence
 → D, Δ → 0 is true. If Δ → 0 then
ρ
d
bn
an
(vn ,i ( x ) − u n ,i ( x )) 2 dx → 0, 1 ≤ i ≤ rn ,1 ≤ n ≤ N
3 Special Properties of the Canonical Form

In this section the theoretical background of the applied HOSVD based algorithm
for two variable case is introduced, i.e.
f ( x, y ), 0 ≤ x ≤ T1 , 0 ≤ y ≤ T2
by noting that the theoretical results valid for the scalar case may be adapted (by
appropriate modifications) also for cases, where the function f is non scalar, but
vector or matrix valued function (e.g. in case of image processing by having three
color components a three element vector valued function may be considered) (see
[2] [3] 3rd chapter). The theoretical background helps to understand the characte-
ristics of the algorithm and to expound the results, which may be specific (e.g.
preprocessing of images, extracting characteristic features, etc.) for certain tasks.
The basis of determination of the canonical form in N=2 case is strongly con-
nected to the classical results of E. Schmidt related to the integral equations,
which in general form are described by the theory related to the so called Hilbert-
Schmidt type operators defined on Hilbert spaces. ([2][3] - 3rd chapter). We note
that such a determination can be considered as the continuous analogue to the
singular value decomposition of matrices. Let 0 < T1 , T2 < ∞ and let f ( x, y ) be a
continuous function on [0, T1 ] × [0, T2 ] . Let 1 and 2 denote over the intervals
[0, T1 ] and [0, T2 ] the Hilbert space of square-integrable functions. Let us consid-
er the A f : 2 → 1 Hilbert-Schmidt integral operator with core function f ( x, y ) ,
i.e.
T2
ϕ ( x) = ( Af ψ )( x) =  f ( x, y )ψ ( y )dy, ψ ∈ 2 .
0
∗ A
Let Af : 1 → 2 denote the adjunct operator to f , i.e.
T2
ψ ( x) = ( A∗f ϕ )( x) =  f ( x, y )ϕ ( x )dx, ϕ ∈ 1 .
0
∗ ∗
In this case the Hilbert-Schmidt integral-operators Af Af and Af Af are the conti-
nuous symmetrical core functions of K1 and K2, respectively, i.e.:
T1
K1 (r1 , r2 ) =  f ( z , r1 ) f ( z , r2 )dz , 0 ≤ r1 , r2 ≤ T2
0
and
T2
K 2 ( s1 , s2 ) =  f (s , z ) f ( s , z)dz,
0
1 2 0 ≤ s1 , s2 ≤ T1 .
∗ ∗
The operators Af Af and Af Af are compact positive having the same discrete
spectrum. Let λ1 ≥ λ2 ≥ ... > 0 denote the monotonically decreasing series of their
non-zero eigenvalues and let ϕ k , k = 1, 2,... represent the series of normalized
∗
eigenfunctions corresponding to the eigenvalues λk of the operator Af Af . In this
case
A∗f Af ϕk = λkϕk , k = 1, 2,...
The series of functions ψ k = Af ϕ k , k = 1, 2,... define the normalized eigenfunc-

∗
tions corresponding to the eigenvalues λk of the operator Af Af where
Af A∗f ψ k = λkψ k , k = 1, 2,...
According to the results connected to the Hilbert-Schmidt operators, function f can

be given in the form of
∞
f ( x, y ) = λk ϕk ( x)ψ k ( y ), 0 ≤ x ≤ T1 , 0 ≤ y ≤ T2
k =1
Furthermore, for the best bilinear approximation the following can be written:
N N
inf f ( x, y ) − uk ( x )vk ( y ) = f ( x, y ) − λk ϕ k ( x )ψ k ( y ) .
u k ∈H T , vk ∈H T , 1≤ k ≤ N
1 2 k =1 L2 k =1 L2
The error of the approximation depends on the asymptotic behavior of the eigen-
values λk . By the estimation of the error the condition for smoothness will be
held[4]. If the core function f ( x, y ) may be determined in the form of
N
f ( x, y ) = λ ϕ
k =1
k k ( x )ψ k ( y )
then the image space of the operator A f will be a finite dimensional Hilbert
space. If the eigenvalues λk are different, then this determination is ambiguous
and is equivalent to the canonical reconstruction of f.
4 System Approximation via HOSVD by Using Local MLP

Models
Let us consider a parameter varying system modeled by local MLPs representing
local "linear time invariant (LTI) like" models in parameter space. Suppose that
these local models are identical in structure, i.e. identical in the number of neurons
for the certain layers and in shape of the transfer functions. The tuning of each
local model is based on measurements corresponding to different parameter vec-
tor. In Fig. 4 a two parameter case can be followed. The architecture of local mod-
els is illustrated by Fig. 3. The output of such a local model can be written in ma-
trix form as follows
( (
y 3 = φ 3 W (3) φ 2 W (2) φ1 ( W (1) u ) ))
Fig. 3. The architecture of the local neural network models. ( R = S0 ).
 w11( j ) w12( j )  w1(Sj )j −1 b j1 

 ( j) 
 w21 w ( j)
 w ( j)
bj2 
=
22 2 S j −1
W( j) ,
   
 w( j ) wS( jj )2 wS( jj S) j −1 b jS j 
 S j1 
j=1..NL and NL stands for the number of layers, i.e. in our case N L = 3 ,
u = ( u1 u 2  u R 1)
T
stands for the input vector, while vector
( )
T
y = y31 y32  y3 S3 represents the output of the MLP in Fig. 3.
Fig. 4 Example of a two dimensional parameter space, with identified neural networks as
local models at equidistant parameter values
Let us assume that the behaviour of the system depends on parameter vector
p = ( p1 pN ) . Let Wi( j ) i
T
p2  represent the matrix containing the weights
1 N
for the jth layer of the local neural network model corresponding to parameter
vector pi ,...,i . Using the weights of the jth layer in all local models and the
1 N
I ×× I N ×S j ×(1+ S j −1 )
parameters pi , i = 1..N an N+2 dimensional tensor  ∈ℜ 1 can be
constructed, as follows:
1 N
{
Wi( j ) i = i i
1 N ,α , β
}
,1 ≤ α ≤ S j ,1 ≤ β ≤ (1 + S j −1 ) , Wi( j ) i ∈ ℜ
1 N
S j × (1+ S j −1 )
By applying the HOSVD on the first N dimensions of tensor  , a core tensor 

and for each dimension an n-mode singular matrix is obtained, which columns
represent the discretized form of one-variable functions discussed in (1). Starting
from the result of this decomposition the parameter varying model can be
approximated with the help of the above mentioned local models, as follows.
Tensor product (TP) transformation is a numerical approach, which can make
connection between parameter varying models and higher order tensors. The
weights corresponding to the jth layer of the parameter varying neural network
model can be expressed in tensor product form, as follows:
W( j ) ( p ) =  ×nN=1 vn ( pn ) ,
(
where the elements of v n ( pn ) = vn1 ( pn ) vn 2 ( pn )  vnI
n
( pn ) ) .
Finally, the output of the parameter varying model can be expressed via local
neural network models illustrated in Fig. 4 in tensor product form as follows:
( (
y 3 ( p ) = φ3 W (3) (p )φ 2 W (2) (p)φ1 ( W (1) (p)u ) , ))
where
W(1) ( p ) = 1 ×nN=1 v(1)
n ( pn ) ,
W(2) ( p ) = 2 ×nN=1 v(2)

n ( pn ) ,
W(3) ( p ) = 3 ×nN=1 v(3)

n ( pn ) .
By discarding the columns of the n-mode singular matrices corresponding to the

smallest singular values, model reduction can be performed. Further details about
HOSVD based model reduction are given in [11].
5 Future Work and Conclusion

In the present paper the theory behind the HOSVD based canonical form of
functions has been discussed. It has been shown how the core function and the
normalized eigenfunctions are related, what properties they have. Additionally we
showed how the HOSVD based canonical form is related to the Hilbert-Schmidt
type integral operators in case of two variable functions. Furthermore, the
HOSVD based upon the canonical form it has been shown how the output of an
MLP neural network may be expressed in tensor product form. Using this repre-
sentation of MLPs the approximation of a parameter varying system may be
performed based on locally tuned MLP models. Our previous researches showed
that the same concept can efficiently be applied to perform reduction in linear
parameter varying (LPV) systems in state space representation. Our next step is to
analyse the impact of the reduction on the output of the system, how the
approximation caused changes in weights of the NNs influence the output. We
hope that the proposed concept could be an efficient compromised modelling view
using both the analytical and heuristical approaches.
Acknowledgments. The project was realised through the assistance of the European Union,
with the co-financing of the European Social Fund TÁMOP-4.2.1.B-11/2/KMR-2011-0001.
References
[1] Schmidt, E.: Zur Theorie der linearen und nichtlinearen Integralgleichungen. 2.
Auflösung der Allgemeinen Linearen Integralgleichungen. Math. Ann. 64, 162–174
(1907)
[2] Balakrishnan, A.V.: Introduction to Optimization Theory in a Hilbert Space. Sprin-
ger, Heidelberg (1971)
[3] Balakrishnan, A.V.: Applied Functional Analysis. Springer, Heidelberg (1976)
[4] Chang, C.-H., Ha, C.-W.: Sharp inequalities of singular values of smooth kernels.
Integral Equations and Operator Theory 35, 20–27 (1999)
[5] De Lathauwer, L., De Moor, B., Vandewalle, J.: A multilinear singular value decom-
position. SIAM Journal on Matrix Analysis and Applications 21(4), 1253–1278
(2000)
[6] Rövid, A., Rudas, I.J., Sergyán, S., Szeidl, L.: HOSVD Based Image Processing
Techniques. In: Proc. of the 10th WSEAS International Conference on Artificial In-
telligence, Knowledge Engineering and Data Bases, Cambridge, UK, February 20-22,
pp. 297–302 (2011) ISBN: 978-960-474-273-8
[7] Szeidl, L., Várlaki, P.: HOSVD Based Canonical Form for Polytopic Models of Dy-
namic Systems. Journal of Advanced Computational Intelligence and Intelligent In-
formatics 13(1), 52–60 (2009) ISSN : 1343-0130
[8] Nagy, S., Petres, Z., Baranyi, P.: TP Tool-a MATLAB Toolbox for TP Model Trans-
formation. In: Proc. of 8th International Symposium of Hungarian Researchers on
Computational Intelligence and Informatics, Budapest, Hungary, pp. 483–495 (2007)
[9] Szeidl, L., Baranyi, P., Petres, Z., Várlaki, P.: Numerical Reconstruction of the
HOSVD Based Canonical Form of Polytopic Dynamic Models. In: 3rd International
Symposium on Computational Intelligence and Intelligent Informatics, Agadir, Mo-
rocco, pp. 111–116 (2007)
[10] Nickolaus, M., Yue, L., Minh, N.D.: Image interpolation using multiscale geometric
representations. In: Proceedings of the SPIE, vol. 6498, pp. 1–11 (2007)
[11] Harmati, I., Rövid, A., Várlaki, P.: Approximation of Force and Energy in Vehicle
Crash Using LPV Type Description. WSEAS Transactions on Systems 9(7), 734–743
(2010)
[12] Khaber, F., Zehar, K., Hamzaoui, A.: State Feedback Controller Design via Takagi-
Sugeno Fuzzy Model: LMI Approach. International Journal of Information and Ma-
thematical Sciences 2(3), 148–153 (2006) ISBN: 960-8457-10-6
[13] Chena, S., Billingsb, S.A.: Neural networks for nonlinear dynamic system modelling
and identification. International Journal of Control 56(2), 319–346 (1992)
[14] Mastorakis, N.E.: The singular value decomposition (SVD) in tensors (multidimen-
sional arrays) as an optimization problem. Solution via genetic algorithms and me-
thod of Nelder–Mead. SEAS Transactions on Systems 21(4), 6(1), 17–23 (2007)
[15] Mulders, A.V., Schoukens, J., Volckaert, M., Diehl, M.: Two Nonlinear Optimization
Methods for Black Box Identification Compared. Preprints of the 15th IFAC Sympo-
sium on System Identification, Saint-Malo, France, July 6-8, pp. 1086–1091 (2009)
[16] Babuska, R., Verbruggen, H.: Elsevier, Annual Reviews in Control 27(1), 73–85
(2003)
[17] Rövid, A., Szeidl, L., Várlaki, P.: On tensor-product model based representation of
neural networks. In: 15th IEEE International Conference on Intelligent Engineering
Systems, Poprad, Slovakia, June 23-25, pp. 69–72 (2011) ISBN: 978-142448956-5
Anytime Extension of Generalized Fuzzy Neural
Network Models with Non-singleton
Consequents
Annamária R. Várkonyi-Kóczy
Institute of Mechatronics and Vehicle Engineering, Óbuda University

Népszínház u. 8., H-1081 Budapest, Hungary
varkonyi-koczy@uni-obuda.hu
Abstract. Nowadays practical solutions of engineering problems involve model-

integrated computing. Model based approaches offer a challenging way to inte-
grate a priori knowledge into the procedure. Recently, Artificial Neural Networks
(ANNs) has become popular because they are able to learn complex mappings
from the input/output data and are relatively easy to implement in any application.
Generalized Neural Network (GNN) based models may have an exceptional role
at many fields, where the problem to be solved is complex, highly nonlinear or
when only partial, uncertain and/or inaccurate data is available, however their high
complexity, and in certain cases unknown accuracy, can limit the applicability,
especially in time critical situations. Combining GNNs with anytime techniques
may offer a solution to the complexity problem, but only in cases when the error
of the reduced models can be estimated. In this paper, author gives error bounds
for a new class of GNNs where the nonlinear transfer functions are approximated
by product-sum-gravity fuzzy systems with non-singleton consequents (NGFNN)
thus extending the range of possible anytime soft computing models. The model
complexity can flexibly be reduced to cope with the temporal, possibly dynamical-
ly changing resource, time, and data availability. It is shown that the accuracy of
the reduced models can be upper-bounded and thus, the error is always known,
and further, monotonously decreases parallel with the increase of the complexity
of the used model. These features make NGFNNs suitable for anytime use.
Keywords: artificial neural networks, fuzzy neural networks, anytime models,

situational models, fuzzy function approximation, adaptive systems, reconfigura-
ble systems.
1 Introduction
Today, Soft Computing (SC) approaches [17] are well known and wide spread
used techniques in complex, nonlinear, and/or time critical systems. Their advan-
tages include robustness, low complexity, and easy interpretability [14]. The con-
cept of anytime [15] and situational processing [5] fits into the idea of SC and if
combined with other: fuzzy [3], [6] and Neural Network (NN) [7], [4] methods, it

234 A.R. Várkonyi-Kóczy
extends the advantages of SC approach with the flexibility with respect to the
available input information and computational power.
Anytime systems ([15], [5]) are to provide continuous operation in cases of
changing circumstances and are to avoid critical breakdowns in cases of missing
input data, temporary shortage of time, or computational power. If a temporal
shortage of computational power and/or loss of some data occur, the actual opera-
tions are continued based on degraded configurations, i.e. algorithms and models
of simpler complexity (and less accuracy) thus resulting in less accurate, approx-
imate output.
Iterative anytime algorithms/models are popular tools, because their complexity
can easily and flexibly be changed. The accuracy of their output is always known
and it grows with the amount of calculations. Further, we don’t have to know the
time/resource-need of a certain configuration in advance. They can simply be
stopped when the results are needed.
If there is not any appropriate iterative method at hand, we still can use the
general modular architecture technique suggested in [11], however at the prize of
lower flexibility and a need for extra planning and considerations.
In this paper, special neural network structures are proposed for anytime use (see
also [13]). The transfer functions of the generalized NNs (where the non-linear
transfer functions are moved from the nodes into the links, see Fig. 1) are approx-
imated by simple SISO Product-Sum-Gravity (PSG) fuzzy systems (generalized
fuzzy-neural networks, GFNN) and the transfer matrix between two neighboring
layers is decomposed based on (Higher Order) Singular Value Decomposition
((HO)SVD) [16]. This offers a way for anytime-type operation of the NNs. Al-
though, we have to be able to give an estimation (upper bound) for the error of the
degraded models. This result is given in the literature only for NNs using Product-
Sum-Gravity fuzzy systems with Singleton consequents (PSGS GFNN, SGFNN).
This paper extends the anytime usability for generalized fuzzy-neural networks
with non-singleton consequents (PSGN GFNN, NGFNN) by determining the error
bounds for the non-exact evaluation of such models. In Section 2 a short introduc-
tion is given to anytime systems. Section 3 summarizes the main ideas of Singular
Value Decomposition. Section 4 is devoted to the reduction of generalized neural
networks with SVD, and Section 5 concludes the paper.
2 Anytime Processing
In resource, data, and time insufficiency, the so-called anytime algorithms, mod-
els, and systems [15], [5] can be used advantageously. They are able to provide
guaranteed response time and are flexible with respect to the available input data,
time, and computational power. This flexibility makes these systems able to work
in changing circumstances without critical breakdowns in the performance. Natu-
rally, while the information processing can be maintained, the complexity must be
reduced, thus the results of the computing become less accurate.
Anytime Extension of Generalized Fuzzy Neural Network Models 235
fl,1,1(x1) Yl+1,1
X1 Nl+1,1
Nl,1 Yl,1
fl,1,j(x1)
fl,1,i(xi)
Xi Yl,j Yl+1,j
Nl,i fl,j,i(xi) Nl+1,j
fl,d,i(xc)
fl,1,c(xc) Yl+1,d
Xc Yl,c Nl+1,d
Nl,c
fl,d,c(xc)
d=nl+1
c=nl
Fig. 1. Generalized Neural Network
Basically, two types of algorithms/models can be used in anytime systems. Iter-

ative algorithms/models always give some, possibly not accurate result and more
and more accurate results can be obtained, if the calculations are continued. A
further advantageous aspect of iterative algorithms is that we don’t have to know
the time/resource-need of a certain configuration in advance, the calculations can
simply be continued until the results are needed.
The general technique for the application of a wide range of other types of
models/ computing methods in anytime systems has been suggested in [11], how-
ever at the prize of lower flexibility and a need for extra planning and considera-
tions. In this case, an anytime modular architecture is used which is composed of
modules realizing the subtasks of a given problem. Each module of the system
offers several implementations (characterized by different attribute-values) for a
certain task having uniform interface (same set of input, output, and solving the
same problem) but differing in their computational need and accuracy. An expert
system is monitoring the actual circumstances (tasks to complete, achievable
time/resources, needed accuracy, etc.) in order to choose the adequate configura-
tion, i.e. the units to be used.
A significant drawback of dynamic anytime schemes is that they unavoidably
suffer from transients. Both parameter and structure adaptations generate tran-
sients. The nature of these transients depends not only on the transfer function of
the structures to be implemented, but also on the actual implementation of the
processing structure. According to our experience, fuzzy and GFNN models are
advantageous from this respect [12].
3 Singular Value Decomposition

Singular Value Decomposition (SVD) has successfully been used to reduce the
complexity of a large family of systems. The technique offers a formal measure to
filter out the redundancy (exact reduction) and also the weakly contributing parts
(non-exact reduction). The degree of reduction can be chosen according to the
maximum acceptable error corresponding to the current circumstances. In case of
multi-dimensional problems, the SVD technique can be defined in a multidimen-
sional matrix form, i.e. Higher Order SVD (HOSVD) can be applied.
SVD is serious candidate to overcome the complexity problems arising in mod-
eling of complex systems where we either have an analytical description of the
system (possibly too complicated to be handled), or the system is represented only
by input-output sample pairs. In these cases, we can build a model approximating
the system using local (linear) models. Such techniques include e.g. Takagi-
Sugeno (TS) fuzzy model approximation [10]. The method has theoretically a
universal approximation property; however, it can not really be exploited because
the complexity grows exponentially with the number of parameters. If the number
of the local units is bounded then the built model will only be an approximation of
the original system. Thus, we have to find a balance between the computational
complexity and the accuracy. For this, some kind of exact or non-exact complexity
reduction method like SVDR can be applied.
The SVD based complexity reduction algorithm is based on the decomposition
of any real valued matrix F:
F (n ×n ) = A1,(n ×n ) B(n ×n ) A2,(n ×n )

T
(1)
1 2 1 1 1 2 2 2
where A , k=1, 2 are orthogonal matrices ( Ak Ak = E ), and B is a diagonal matrix

T
containing the λ i singular values of F in decreasing order. The maximum num-

ber of the nonzero singular values is nSVD = min(n1, n2 ) . The singular values indi-
cate the significance of the corresponding columns of A k . The matrices can be
partitioned in the following way:
r
A k = A k ,( n
r d
A k ,( n
and B = B ( n ×n )
r r
0 ,
×nr ) × ( n k − n r )) d
k k
0 B (( n −n )×( n
1 r 2 − nr ))
where r denotes “reduced” and nr ≤ nSVD .

d
If B contains only zero singular values then B and Ak can be dropped:
d d
F = A1 B A2 . If
r r rT d
B contains nonzero singular values, as well, then the
r r rT
F'= A1 B A2 matrix is only an approximation of F and the maximum difference
between the values of F and F ' equals to

n SVD
E RSVD = F − F ' ≤ (  λ )1 i ( n1 × n 2 )
(2)
i = n r +1 .
A better error bound is given in [9] according to which the error of the matrix
reduction can be estimated by the maximum element of the error matrix:
n SVD
E RSVD = max( E ) , where E = F − F ' = λ
p = n r +1
p a 1, k a 2 , k .
T
The error estimation procedure can be applied both to matrix F and also on a sub-
matrix of E. These latter offers a possibility to give a better error bound for a sub-
space of the system, which can be advantageous in cases when the output of the
system may vary in a large domain and we need lower bounds for the relative
error of the system in the different domains.
For higher order cases Higher Order SVD (HOSVD) can be applied in a similar
way (see e.g. [2]).
If SVD based complexity reduction is applied to a two dimensional matrix then
it can be proved that the resulting matrix of lower rank will be the best approxima-
tion of the original matrix in least-squares sense (minimum L2 norm of the error,
i.e. the reduction is “optimal”). In case of higher dimension matrices where
HOSVD is applied, the minimum property does not hold anymore. We can only
state that the “significant” singular values will have the “lower” indices. However,
in the most cases HOSVD results in an approximation which is “very near” to the
optimal one.
4 Reduction of Generalized Fuzzy Neural Networks With SVD
4.1 Generalization of Feedforward Neural Networks

The classical multi-layer neural network can be generalized if the non-linear trans-
fer functions are moved from the nodes into the links. It results in neurons that
apply only a sum operation to the input values, and links that are characterized by
possibly non-linear weighting functions instead of simple constant weights (see
Fig. 1). A further advantage of GFNNs is that even different weighting functions
can be applied at the different connections.
Let us focus on two neighboring layers l and l+1 of a forward model. Let the
neurons be denoted as N l , i , i = 1 ..n l in layer l, where n l is the number of the
neurons. Further, let input values of N l , i be x l , i , k , k = 1..nl −1 and its output
y l , i . The connection between layers l and l+1 can be defined by the f l , j , i ( yl , i )
weighting functions ( j = 1..nl +1 ). Thus
x l + 1, j , i = f l , j , i ( y l , i ) (3)
and the output of neuron N l +1, j can be computed as
Bl,1,i,t
X1 μ μ
Nl,1 Nl+1,1 Yl+1,1
Antecedents Yl,1 Xl+1,1
Al,i,t Rule Bl,j,i,t

Xi μ μ
Nl i Yl+1,j
Nl+1,j
Xl+1,j
Antecedents Yl,j
Bl,d,i,t
Xc μ μ Yl+1,d
Nl,c Nl+1,d
Xl+1,d
Antecedents Yl,c
c=nl d=nl+1
Fig. 2. SGFNN approximation of the GNN
nl
y l + 1, j =  f l , j , i ( y l , i )
i =1 (4)
.
The weighting functions can also be changed during the training: the unknown
weighting functions are approximated with linearly combined known functions,
where only the linear combination must be trained (see also Fig. 2).
4.2 Reduction of Product-Sum-Gravity GNNs with Singleton

Consequents
In case of PSGS generalized fuzzy neural networks, for the approximation of the
weighting functions of the GNN, SISO PSGS fuzzy systems are used (see Fig. 2)
nl nl ml ,i
y l + 1, j =  f l , j ,i ( y l ,i ) =   μ l ,i , t ( y l ,i ) b l , j ,i , t (5)
i =1 i = 1 t =1 .
To reduce the size of a generalized neural network the SVD based complexity
reduction can be used. (5) can always be transformed into the following form
n lr+1 nl m lr,i
y 'l +1, j =  al , j , z   μ lr,i , t ( yl , i )b 'l , z , i , t (6)
z =1 i =1 t =1
where “r” denotes “reduced”, further n lr + 1 ≤ n l + 1 and ∀i : mlr,i ≤ ml,i .

The reduced form is represented as a neural network with an extra inner layer
between layers l and l+1 (see Fig. 3). Between the original layer l and the new
layer the weighting functions are approximated from the reduced PSGS fuzzy
systems, and layer l+1 simply computes the weighted sum ( a l , j , z ) of the output of
the new layer.
The reduction means the reduction of the B = [bl , j ,i ,t ] three-dimensional matrix
in two steps by applying the HOSVD reduction algorithms: In the first step, the
first dimension is reduced, and the a l , j , z values are determined while in the
second the third dimension is reduced, and the new membership functions are
determined. The detailed description of the algorithm can be found in [1].
The maximum error of the resulted NN can be computed from the discarded
singular values, considering that the singular values discarded in the first step
“count” n l times ([8]).
Bl,1,i,t
μ Yl+1,1
X1 μ
Nl,1 Nl+1,1
Al,i,t Rules Bl,j,i,t

μ Yl+1,j
Xi
Nl,i μ Nl+1,j
Antecedents Yl,j Xl+1,j
Bl,e,i,t
μ
Xc μ Yl+1,d
Nl Nl+1,d
Antecedents Yl,c Xl+1,e
c=nl
d=nl+1
e=nrl+1
Extra layer
Fig. 3. Reduced SGFNN
4.3 Reduction of Product-Sum-Gravity GNNs with Non-singleton

Consequents
In case of PSGN generalized fuzzy neural network (NGFNN), the approximation
of the weighting functions of the GNN is solved by a PSGN fuzzy system with
rules
If At then Bt .
Here the consequent fuzzy sets are non-singleton fuzzy sets with centers of gravity
of d t and areas of st (PSGN fuzzy inference). The result of the inference equals
μ (x )d s t
∗
t t
y∗ = t =1
m
(7)
μ (x )s
t =1
t
∗
t
In order to approximate the contribution of each neuron to the neurons of the next
layer by the PSGN technique, let (7) be substituted into (4). Based on this, a more
general form can be defined where all antecedent universes may have different
number of antecedent sets:
ml ,i
nl nl μ l ,i ,t ( yl ,i ) d l , j ,i ,t sl , j ,i ,t
yl +1, j =  f l , j ,i ( yl ,i ) =  t =1
ml ,i
=
 μ l ,i,t ( yl ,i )sl , j ,i,t
i =1 i =1
t =1
m l ,i
nl μ l ,i ,t ( yl ,i )bl , j ,i ,t nl
N l , j ,i
= =
t =1 (8)
ml ,i
Dl , j ,i
μ
i =1 i =1
l ,i , t ( yl ,i ) sl , j ,i ,t
t =1
Here ml ,i stands for the number of antecedent sets in layer l, bl , j , i , t = d l , j , i , t sl , j , i , t ,

N l , j , i denote the nominators and D l , j , i the denominators. Fig. 4 depicts the
NGFNN representing (8).
Theorem 1. (8) can always be transformed into the form of
mlr,i
n l +1 μ r
l ,i ,t ( yl ,i )bl' , z ,i ,t n l +1
N 'l , j , i (9)
y l +1, j = 
' t =1
mlr,i
=
D 'l , j , i
μ
i =1 r i =1
l ,i ,t ( yl ,i ) sl' , z ,i ,t
t =1
where “r” denotes “reduced”, and ∀ i : m lr, i ≤ m l , i , and the maximum error of the
reduction can be computed from the discarded singular values. The reduced form
can be represented as a neural network illustrated in Fig. 5.
Proof. Let μ i and M l ,i
be defined as:
[
μ i ( y l ,i ) = μ l ,i ,1 ( y l ,1 )  ]
μ l ,i , m ( y l ,i ) and M l , i = O l , i P l , i
l ,i
Bl,1,i,t
X1 μ μ
Yl+
Nl,1 Nl+1,1 11
Al,i,t
Rule Bl,j,i,t
Xi μ
Nl,i μ Yl+1
Nl+1,j ,j
Antecedents Yl,j
Xl+1,j
Bl,d,i,t
Xc μ Yl+1,d
μ
Nl,c Nl+1,d
Antecedents Yl,c Xl+1,d
c=nl d=nl+1
Fig. 4. NGFNN approximation of the GNN
where
bl ,1,i ,1  bl ,n ,i1 sl ,1,i ,1  sl ,n ,i1

l +1 l +1
.
O l ,i =    P l ,i =   
bl ,1,i ,ml ,i  bl ,n sl ,1,i ,ml ,i  sl ,n
l +1 ,i ,ml ,i l +1 ,i ,ml ,i
With this notations

[N l ,1 , i  ]
N l , nl +1 ,i = μ i ( y l ,i ) O l ,i and D l ,1 ,i [  ]
D l , n l + 1 , i = μ i ( y l ,i ) P l ,i (10)
Applying the above described singular value-based reduction to the M l ,i

matrices
one will get
M l ,i ≈ T l ,iU V = T l , i M 'l , i = T l , i O ' P 'l , i (11)

l ,i l ,i l ,i
The error of this reduction ( M l , i − T l , i M 'l , i ) will be the sum of the discarded
singular values, Ei, while the output of the reduced neural network equals
m lr,i
n l +1 μ r
l ,i ,t ( yl ,i )bl', z ,i ,t n l +1
N 'l , j ,i , (12)
y 'l +1, j =  t =1
m lr,i
=
D 'l , j ,i
μ
i =1 r i =1
l ,i ,t ( yl ,i ) sl' , z ,i ,t
t =1
where
[N 'l ,1,i ]
 N 'l , nl+1 ,i = μ i ( yl ,i )O' , and D'l,1,i  D'l,nl+1 ,i = μ i ( yl ,i )P'l ,i .
r
l ,i
[ ] r
Bl,1,i,t
X1 μ μ Yl+1,1
Nl,1 Nl+1,1
Al,i,t
Rules Bl,j,i,t
Xi μ Yl+1,j
Nl,i μ Nl+1,j
Antecedents Yl,j
Xl+1,j
Bl,d,i,t
Xc μ
μ Yl+1,d
Nl,c Nl+1,d
Antecedents Yl,c
c=nl Xl+1,d d=nl+1
Fig. 5. Reduced NGFNN
The antecedent fuzzy sets of the reduced network can be obtained as
μ r i ( y l ,i ) = [μ r l ,i ,1 ( y li 1 )  μ r l , i , m ( y l ,i ) ] = μ i ( y l , i ) T l , i
l ,i
(13)
The reduced consequent fuzzy sets of the reduced network will have centers of
gravity of d 'l , j ,i ,t = b 'l , j ,i ,t / s 'l , j ,i ,t and areas of s'l , j ,i ,t , where b'l , j ,i ,t and s'l , j ,i ,t are the
elements of O'l ,i and P'l ,i , respectively.
The errors of the numerators can be determined as
N l , j , i − N ' l , j ,i = μ i ( y l ,i ) O l ,i − μ i ( y l ,i ) O ' l , i
r
(14)
nl
= μ i ( y l ,i )( O l ,i − T l ,i O 'l ,i ) ≤ μ
i =1
l ,i , t ( y l ,i ) E i = E i
because the original antecedent fuzzy sets are in Ruspini-partition.

The error of the denominator can be estimated in the same way, and it is also
equal or less, then Ei. Let
Δ N l , j ,i = N 'l , j ,i − N l , j ,i and Δ Dl , j ,i = D 'l , j ,i − Dl , j ,i .
The overall error of yl +1, j can be given as
nl nl nl
N 'l , j ,i N l , j ,i N 'l , j ,i N l , j ,i
y 'l +1, j − yl +1, j =  D'
i =1
−
i =1 Dl , j ,i
≤
i =1 D'l , j ,i
−
Dl , j ,i
l , j ,i
(15)
nl
Dl , j ,i ( N l , j ,i + ΔN l , j ,i ) − N l , j ,i ( Dl , j ,i + ΔDl , j ,i )
=
i =1 Dl , j ,i D'l , j ,i
nl
ΔN l , j ,i N l , j ,i ΔDl , j ,i
= −
i =1 D'l , j ,i Dl , j ,i D'l , j ,i
Because ΔN l , j ,i ≤ E i and ΔD l , j ,i ≤ E i :
nl  Ei N l , j ,i Ei  nl  E  N  (16)
y 'l + 1, j − yl + 1, j ≤   +  =  i  1 + l , j ,i 
i = 1 D 'l , j ,i Dl , j ,i D 'l , j ,i  i = 1 D 'l , j ,i  Dl , j ,i 
   
Since the original antecedent fuzzy sets are in Ruspini-partition, the maximum of
Nl , j , i
Dl , j ,i
will be in one of the grid-points (points, where one of the original
μ l , i , t ( yl , i ) = 1 for every i):
Nl , j,i
≤ max(dl ,i, j,t ) . (17)
Dl , j,i t
If D 'l , j ,i ≥ 0 , i.e. Ei ≤ min(sl , j ,i,t ) , then the error is

t
nl  
 Ei  1 +  ,
y 'l +1, j − yl +1, j ≤   max ( d l , j , i , t )   (18)
i =1  min ( sl , j , i , t ) − E i  t 
 t 
where E i is the sum of discarded singular values, sl , j ,i , t are the areas and d l , j ,i ,t
are the centers of gravity of the original consequent fuzzy sets.
4.4 Improvement on the Reduction of NGFNNs: Loosing the

Fuzzy Logic Form
The effectiveness of the complexity reduction may be improved, if the fuzzy logic
form can be lost in the resulted network.
Theorem 2. (8) can always be written into the form of
n lr,i m lr,i
nl a μ l ,i , z
r
l ,i , t ( yl , i )b ' 'l , z , i , t nl
N ' 'l , j , i , (19)
y 'l +1, j =  z =1
r
t =1
r =
i =1
n l ,i m l ,i
i =1 D ' 'l , j , i
a μ
r
l ,i , z l , i ,t ( yl , i ) s ' 'l , z , i , t
z =1 t =1
where „r” denotes „reduced”, „ ” denotes values in the denominator correspond-

r r
ing to values in the numerator, further nlr+1, nl +1 ≤ nl +1 and ∀i : mlr,i , ml ,i ≤ ml ,i , and
the maximum error of the reduction can be computed from the discarded singular
values.
Proof. The numerator and denominator parts can be reduced separately. In the
following, first the reduction of the numerator is shown and the error bound is
given. The reduction of the denominator can be carried out in the same way, and
the error-bound can be determined similarly.
The reduction can be made in two steps. First, the a l , j , z values are computed,
while in the second the new μ r l , i , t ( y l , i ) functions are determined.
Step 1: Determination of the a l , j , z values: Let μ i and S l be defined as:
[
μi = μl ,i,1( yl,i )  μl,i,m ( yl ,i ) T , S l = H l ,1  H l ,n ,
l ,i
] l
bl ,1,i ,1  bl ,1,i ,ml ,i

H l ,i =   
bl ,nl +1 ,i ,1  bl ,nl +1 ,i ,ml ,i
where the μ vectors (lengths: ml ,i ) contain the values of the membership func-
i
tions in a given ( y l ,1 ,..., y l , n ) point, and S contains all the bl , j ,i ,t values. The size
l l
nl
of S is n ×  m .
l l +1 l ,i
i =1
With this notation:
[N l ,1 ,i  ]
N l , n l +1 , i = H l ,i μ i ( y l ,i ) . (20)
Applying the above described singular value-based reduction to the S l

matrix
one gets:
S ≈ Al Dl V = Al S 'l = Al H 'l ,1  H 'l ,nl , (21)

l l
The error of the reduction is the sum of the discarded singular values, EN,1:
nlr+1
bl , j ,i ,t −  al , j , zb'l , z ,i ,t ≤ E N ,1 , (22)
z =1
where b'l , j ,i ,t are the elements of S 'l . The numerators after the first step are:
[N ' l ,1 , i  ]
N 'l , n l + 1 , i = A l H ' l , i μ i ( y l , i ) . (23)
The difference between the current and the original numerators, using that the
original antecedent fuzzy sets are in Ruspini-partition (i.e. sum ( μ 1,i ( y l ,i )) = 1 for
every y l , i ):
(
Al H 'l ,i μ 1,i ( yl ,i ) − H l ,i μ 1,i ( yl ,i ) ≤ Al H 'l ,i − H l ,i μ 1,i ( yl ,i ) ≤ EN ,1 ) (24)
Step 2: Determination of the new membership functions: Applying the singular

T
value-based reduction to the H'l ,i matrices one gets:
H 'Tl ,i ≈ T l ,i Dl ,iV = T l ,i H ' 'Tl ,i . (25)

l ,i
The error of this reduction ( H'Tl ,i −T H' 'Tl ,i ) will be the sum of the discarded singu-
l ,i
lar values, EN,2,i (the error also depends on i). The numerators after the reduction
are
[N' ' l ,1,i

T
]
 N' 'l,nl +1,i = μi ( yl ,i )T l ,i H ' 'Tl ,i Al .
T T
(26)
The new μ r l , i , t ( y l , i ) functions can be obtained as
μr i ( yl ,i ) = [μ r l,i,1( yli1)  μr l ,i,m ( yl ,i )] = μTi ( yl ,i )T l ,i , l ,i

(27)
where the b ' 'l , j , i , t values are the elements of H ' 'l , i .
The error of the numerator in the second step:
[N ' l ,1,i ] [
 N 'l ,nl +1 ,i − N ' 'l ,1,i  N ' 'l ,nl +1 ,i
T
]T
=
= μ i ( yl ,i )T l ,i H ' 'Tl ,i Al − μ i ( yl ,i ) H 'Tl ,i Al = μ Ti ( yl ,i )(T H ' 'Tl ,i − H 'Tl ,i ) AT ≤ E N , 2,i

T T T T
l ,i l
(28)
For the last step, the sum of the rows of A l must be less or equal than one. This
can be ensured by a further transformation in the first step:
−1
S ≈ A'l D'l V 'l = ( A'l K l ) K l D'l V 'l = Al S 'l ,
l
−1
K l = 1 / max { a l , j , z } , and S 'l = K l D'l V 'l (29)
j
z
Thus, the error of the numerator will be altogether the sum of the errors originat-
ing from the first and second steps (from (24) and (28)):
EN ,i = Nl , j ,i − N ' 'l , j,i ≤ EN ,1 + EN ,2,i . (30)
Similarly can be obtained ED,i , the error of the denominator, as well. Let
Δ N 'l , j , i = N ' 'l , j , i − N l , j , i and Δ D 'l , j , i = D ' 'l , j , i − D l , j , i . The overall error of yl +1, j is
(using (15))
nl
N ' 'l , j , i nl
N l , j ,i nl
Δ N 'l , j , i N l , j , i Δ D 'l , j , i (31)
y 'l + 1, j − y l + 1, j =  D' '
i =1
−
i =1 Dl , j , i
≤
i =1 D 'l , j , i
−
Dl , j , i D ' 'l , j , i
l , j ,i
Because ΔN'l , j ,i ≤ EN ,i and ΔD'l , j ,i ≤ ED,i :
nl  
E N E
y 'l +1, j − yl +1, j ≤   N ,i + l , j ,i D ,i  (32)

i =1  D ' D D 'l , j ,i 
l , j ,i l , j ,i 
If D ' 'l , j , i ≥ 0 , i.e., ED,i ≤ min(sl , j ,i ,t ) then the error is, using (17):
t
nl  
 1  E + E  , (33)
y 'l +1, j − yl +1, j ≤   D , i max ( d l , j , i , t )  
i =1  min ( sl , j , i , t ) − E D , i  
N ,i
t
 t 
where sl , j ,i , t are the areas and d l , j ,i ,t are the centers of gravity of the original con-
sequent fuzzy sets, and E N , i and E D , i are the sums of the discarded singular
values during the reduction of the numerators and denominators, respectively.
5 Conclusion
In modern embedded signal processing, monitoring, diagnostics, measurement,
and control systems, the available time and resources are often not only limited,
but can also change during the operation of the system. In these cases, the so
called anytime models and algorithms can be used advantageously. While differ-
ent soft computing methods are widely used in system modeling, their usability is
limited, because of the lack of any universal method for the determination of the
needed complexity. It often results in huge and redundant neural networks/fuzzy
rule-bases. In this paper, author investigates a possible way to carry out anytime
processing of generalized fuzzy neural network models with non-singleton conse-
quents, with the help of the (Higher Order) Singular Value Decomposition based
transformation and complexity reduction. An error-bound is given for such models
thus the error of the reduction can always be estimated from the discarded singular
values. The given error is valid in case of the reduction of only one layer of the
neural network. If more layers are reduced then the effect of the reduction of the
layers must be summarized.
Acknowledgments. This paper was supported by the Hungarian National Fund for Scien-
tific Research OTKA T78576.
References
[1] Baranyi, P., Lei, K., Yam, Y.: Complexity reduction of singleton based neuro-fuzzy
algorithm. In: Proc. of the 2000 IEEE International Conference on Systems, Man, and
Cybernetics, Nashville, USA, vol. 4, pp. 2503–2508 (2000)
[2] Baranyi, P., Várkonyi-Kóczy, A.R., Yam, Y., Michelberger, P.: HOSVD based Com-
putational Complexity Reduction of TS Fuzzy Models. In: Proc. of the Joint 9th IFSA
World Congress and 20th NAFIPS International Conference, IFSA / NAFIPS 2001,
Vancouver, Canada, pp. 2482–2485 (2001)
[3] Klir, G.J., Folger, T.A.: Fuzzy Sets, Uncertainty, and Information. Prentice Hall Int.
Inc., USA (1988)
[4] Lovassy, R., Kóczy, L.T., Gál, L.: Function Approximation Performance of Fuzzy
Neural Networks. Acta Polytechnica Hungarica 7(4), 25–38 (2010)
[5] Madarász, L., Andoga, R., Főző, L., Lazar, T.: Situational Control, Modeling and
Diagnostics of Large Scale Systems. In: Rudas, I.J., Fodor, J., Kacprzyk, J. (eds.)
Towards Intelligent Engineering and Information Technology. Studies in Computa-
tional Intelligence, vol. 243, pp. 153–164. Springer, Heidelberg (2009)
[6] Perfilieva, I., Wrublová, M., Hodáková, P.: Fuzzy Interpolation According to Fuzzy
and Classical Conditions. Acta Polytechnica Hungarica 7(4), 39–58 (2010)
[7] Rojas, R.: Neural Networks, A Systematic Introduction. Springer, Berlin (1996)
[8] Takács, O., Nagy, I.: Error-bound of the SVD Based Neural Networks. In: Proc. of
the IFAC Symp. on Artificial Intelligence in Real-Time Control, AIRTC 2000,
Budapest, Hungary, pp. 139–144 (2000)
[9] Takács, O., Várkonyi-Kóczy, A.R.: SVD Based Complexity Reduction of Rule Bases
with Non-Linear Antecedent Fuzzy Sets. IEEE Trans. on Instrumentation and Mea-
surement 51(2), 217–221 (2002)
[10] Takagi, T., Sugeno, M.: Fuzzy Identification of Systems and Its Applications to Mod-
eling and Control. IEEE Trans. on Systems, Men, and Cybernetics 15, 116–132
(1985)
[11] Várkonyi-Kóczy, A.R., Ruano, A., Baranyi, P., Takács, O.: Anytime Information
Processing Based on Fuzzy and Neural Network Models. In: Proc. of the 2001 IEEE
Instrumentation and Measurement Technology Conference, IMTC/2001, Budapest,
Hungary, pp. 1247–1252 (2001)
[12] Várkonyi-Kóczy, A.R.: Methods of Computational Intelligence for Modeling and Da-
ta Representation of Complex Systems. DSc Thesis, Hungarian Academy of Science
(2009)
[13] Várkonyi-Kóczy, A.R.: Neural Network Models for Anytime Use. In: Proc. of the
15th IEEE Int. Conference on Intelligent Engineering Systems INES 2011, Poprad,
Slovak Republic, pp. 95–100 (2011)
[14] Yager, R.R.: Fuzzy thinking as quick and efficient. Cybernetica 23, 265–298 (1980)
[15] Zilberstein, S.: Using Anytime Algorithms in Intelligent systems. AI Magazine 17(3),
73–83 (1996)
[16] Yam, Y.: Fuzzy Approximation via Grid Sampling and Singular Value Decomposi-
tion. IEEE Trans. on Systems, Men, and Cybernetics 27(6), 933–951 (1997)
[17] Zadeh, L.A.: Fuzzy logic, neural networks, and soft computing. Communications of
the ACM 37(3), 77–84 (1994)
On Probabilistic Correlation Coefficients
for Fuzzy Numbers
Robert Fullér1, István Á. Harmati2, and Péter Várlaki3,4

1
Institute of Intelligent Engineering Systems, John von Neumann Faculty of Informatics
Óbuda University, Bécsi út 96/b, H-1034 Budapest, Hungary
fuller.robert@nik.uni-obuda.hu
2
Department of Mathematics and Computational Science, Széchenyi István University
H-9026 Győr,Egyetem tér 1., Hungary
harmati@sze.hu
3
Széchenyi István University H-9026 Győr, Egyetem tér 1., Hungary
4
Budapest University of Technology and Economics H-1118 Budapest,
Bertalan L. u. 1., Hungary
varlaki@sze.hu
Abstract. In this paper we introduce alternative definitions for the measure of

interactivity between fuzzy numbers by defining non-uniform probability distribu-
tions on the γ-level sets (γ-cuts) of their joint possibility distribution. These proba-
bility distributions are determined by the shape function of the joint possibility
distribution if we consider this as a probability density function (with an appropri-
ate constant multiplier), so we use only the information contained in the joint pos-
sibility distribution. We also show some detailed examples for the calculation
when the joint possibility distributions are defined by well-known t-norms, such as
Mamdani, Lukasiewicz and Larsen t-norms.
1 Introduction
Uncertain variables have great importance in life sciences, engineering sciences,
economy, mathematics, etc. Measuring dependence (or interactivity) between
uncertain quantities is also an important and difficult task.
In probability theory the expected value of functions of random variables plays
a fundamental role in defining the basic characteristic measures of probability
distributions. For example, the variance, covariance and correlation of random
variables can be computed as the expected value of their appropriately chosen
real-valued functions. In possibility theory we can use the principle of expected
value of functions on fuzzy sets to define variance, covariance and correlation of
possibility distributions. Marginal probability distributions are determined from
the joint one by the principle of 'falling integrals' and marginal possibility distribu-
tions are determined from the joint possibility distribution by the principle of
'falling shadows'. Probability distributions can be interpreted as carriers of incom-
plete information [1], and possibility distributions can be interpreted as carriers of

250 R. Fullér, I.Á. Harmati, and P. Várlaki
imprecise information. In 1987 Dubois and Prade [2] defined an interval-valued

expectation of fuzzy numbers, viewing them as consonant random sets. They also
showed that this expectation remains additive in the sense of addition of fuzzy
numbers.
In possibility theory we can use the principle of average value of appropriately
chosen real-valued functions to define mean value, variance, covariance and corre-
lation of possibility distributions. Namely, we can equip each level set of a possi-
bility distribution (represented by a fuzzy number) with a uniform probability
distribution, then apply their standard probabilistic calculation, and then define
measures on possibility distributions by integrating these weighted probabilistic
notions over the set of all membership grades [3, 4]. These weights (or impor-
tances) can be given by weighting functions.
Definition 1. A function: g : [ 0,1 ] →  is said to be a weighting function if g is
non-negative, monotone increasing and satisfies the following normalization con-
1
dition  g ( γ ) d γ = 1 .
0
Different weighting functions can give different (case-dependent) importances to

level-sets of possibility distributions. We should note here that the choice of uni-
form probability distribution on the level sets of possibility distributions is not
without reason. We suppose that each point of a given level set is equally possible
and then we apply Laplace's principle of Insufficient Reason: if elementary events
are equally possible, they should be equally probable (for more details and genera-
lization of principle of Insufficient Reason see [5], page 59).
Definition 2. A fuzzy number A is a fuzzy set  with a normal fuzzy convex and
continuous membership function of bounded support.
Fuzzy numbers can be considered as possibility distributions. A fuzzy set C in  2

is said to be a joint possibility distribution of fuzzy numbers A, B, if it satisfies the
relationships
max{ x | C ( x, y)} = B( y) and max{ y | C( x, y)} = A( x)
for all x, y ∈  . Furthermore, A and B are called the marginal possibility distribu-
tions of C. A γ-level set (or γ-cut) of a possibility distribution C is a non-fuzzy set
denoted by [C] and defined by
γ
[C]γ = {( x, y) ∈ X | C ( x, y) ≥ γ }
if γ >0 and cl(supp C) if γ= 0, where cl(supp C) denotes the closure of the support
of C.
On Probabilistic Correlation Coefficients for Fuzzy Numbers 251
2 Possibilistic Correlation Coefficient for Marginal Possibility

Distributions
Fullér, Mezei and Várlaki introduced a new definition of possibilistic correlation
coefficient (see [6, 7]) that improves the earlier definition introduced by Carlsson,
Fullér and Majlender in 2005 (see [3]).
Definition 3. The g-weighted possibilistic correlation coefficient of fuzzy numbers
A and B (with respect to their joint distribution C) is defined by
1
ρ g ( A, B ) =  ρ ( X γ , Yγ ) g ( γ ) d γ ,
0
where
cov ( X γ , Yγ )
ρ ( X γ , Yγ ) = ,
var ( X γ ) ⋅ var (Yγ )
and, where X γ and Y γ are random variables whose joint distribution is uniform
( )
on [C] for all γ ∈ [ 0,1] , and cov Xγ ,Yγ denotes their probabilistic covariance.
γ
In other words, the g-weighted possibilistic correlation coefficient is nothing

else, but the g-weighted average of the probabilistic correlation coefficients
ρ ( Xγ ,Yγ ) for all γ ∈ [ 0,1] .
We should note here that there exist several other ways to define correlation
coefficient for fuzzy numbers, e.g. Liu and Kao [8] used fuzzy measures to define
a fuzzy correlation coefficient of fuzzy numbers they formulated a pair of nonli-
near programs to find the γ-cut of this fuzzy correlation coefficient, then, in a spe-
cial case, Hong [9] showed an exact calculation formula for this fuzzy correlation
coefficient.
3 Alternative Definitions of Possibilistic Correlation Coefficient

Using Non-uniform Distributions
In standard definition of possibilistic correlation coefficient (according to Lap-
lace's principle of Insufficient Reason) each γ-level set is equipped with uniform
probability distribution [6]. From this the correlation coefficient is the weighted
average of correlation coefficients of random variables X γ and Y γ whose joint
probability distribution is uniform on [C] . A natural generalization of this ap-

γ
proach to define non-uniform probability distributions on γ-levels.

If we have an empirical data set and we make a frequency histogram, then it
can be viewed in at least two different ways. On one hand it can be considered as
an empirical approximation of a probability density function, on the other hand it
can be considered as empirical approximation of a membership function (possi-

bility distribution). From this dual interpretation it is obvious to define a non-
uniform probability distribution on γ-levels for which the joint density function
has the same shape as of the joint possibility distribution over [C] . If we use this
γ
definition, the joint probability density function will be the following for every γ:
 1 γ
 K C ( x, y ) if ( x, y ) ∈ [C ]
f ( x, y ) =  γ ,
γ

0 otherwise
where C ( x , y ) is the joint possibility distribution and Kγ is the integral of
[C]
γ
C ( x , y ) over (i.e. 1 is a normalizing factor to get a density function).
Kγ
A γ-level set of a possibility distribution contains every elements whose mem-
bership grade is at least γ, that is elements of [C] have membership at least γ. If
γ
we know that two elements are in [C] , then the main information about their
γ
membership grades is the difference from γ. The part of the joint possibility distri-
bution which is appropriate to this property is the 'cap' over the γ but shifted down
by γ. So in this case the probability density function over [C] has the same shape
γ
as of this 'cap', and it will be:
 1
 K ( C ( x, y ) − γ ) if ( x, y ) ∈ [C ]
γ
fγ ( x, y ) =  γ
0
 otherwise
where C ( x , y ) is the joint possibility distribution and K γ is the integral of
[C]
γ
C ( x , y ) − γ over (i.e. 1 is a normalizing factor again).
Kγ
These two approaches use only information contained in the joint possibility
distribution. Of course it is possible to define other non-uniform probability distri-
butions on [C] but if we have no more information about the probability distri-
γ
bution, we could use only the shape of C ( x , y ) .
4 Examples
In this Section we will compute correlation coefficients for A and B when their
(
joint possibility distribution, C, is defined by C ( x, y) = T A( x) , B ( y) , where T )
is a t-norm [10]. We will consider the Mamdani t-norm [11], Lukasiewicz t-norm
[12] and Larsen t-norm [13]. For simplicity we will consider fuzzy numbers of
same shape,
 x if 0 ≤ x ≤ 1  y if 0 ≤ y ≤ 1
A(x) =  , B ( y) =  .
 0 otherwise  0 otherwise
In [14] for a similar case was proved for that the correlation coefficient depends
only on the joint possibility distribution, but not directly on its marginal distribu-
tions. So it is an interesting problem to consider different joint possibility distribu-
tions with same marginal distributions.
Recall that if X and Y are random variables then their correlation coefficient is
computed from
E ( XY ) − E ( X ) E (Y )
ρ ( X ,Y ) = .
E ( X 2 ) − E 2 ( X ) E ( Y 2 ) − E 2 (Y )
In the followings we will refer to the cases as below [15]:

• Standard possibilistic correlation coefficient: Uniform distribution (see
[16]).
• The joint probability density function has the same shape as of the
joint possibility distribution over [C] : Non-uniform density function I.
γ
The joint probability density function over [C] has the same shape as
γ
•
C ( x , y ) shifted down by γ: Non-uniform density function II.
4.1 Mamdani t-Norm
In this case the joint possibility distribution C ( x , y ) is defined by the min

operator,
 x if 0 ≤ x ≤ y ≤ 1

C ( x, y ) = min { A ( x ) , B ( y )} =  y if 0 ≤ y < x ≤ 1
0 otherwise

Then a γ-level set of C is
[C ]γ = {( x, y ) ∈ 2 | γ ≤ x, y ≤ 1}
4.1.1 Uniform Distribution
The joint density function of a uniform distribution on [C] is

γ
1
T if ( x, y ) ∈[C ]
γ
f ( x, y ) =  γ
0 otherwise

where
Tγ = (1−γ )
2
denotes the area of the γ-level set. The marginal density functions are
1 1
 (1 − γ ) if γ ≤ x ≤ 1  (1 − γ ) if γ ≤ y ≤ 1
f1 ( x ) = Tγ , f 2 ( y ) = Tγ
0 otherwise 0 otherwise
 
The expected values are
1+ γ
E ( Xγ ) = E (Yγ ) =
2
1+ γ + γ 2
E ( X γ 2 ) = E (Yγ 2 ) =
3
(1 + γ )
2
E ( X γ Yγ ) =
4
From these results we get the correlation coefficient is equal to zero:
ρ ( Xγ ,Yγ ) ≡ 0
It is not surprising, since in this case A and B are non-interactive fuzzy numbers.
4.1.2 Non-uniform Density Function I
The normalizing factor (integral of C ( x , y ) over [C] ) is

γ
1
Kγ = (1 − γ ) (1 + 2γ )
2
3
The joint probability density function is
1
 K C ( x, y ) if ( x, y ) ∈ [C ]
γ
fγ ( x, y ) =  γ
0
 otherwise
The marginal density functions are

 1  x2 γ 2   1  y2 γ 2 
  x − −  if γ ≤ x ≤ 1   y − −  if γ ≤ y ≤ 1
f1 ( x ) =  Kγ  2 2 , f 2 ( y ) =  Kγ  2 2 
 
0 otherwise 0 otherwise
E ( X γ ) = E ( Yγ ) =
1 1
⋅
K γ 24
( 5 − 6γ 2 − 8γ 3 + 9γ 4 )
E ( X γ 2 ) = E (Yγ 2 ) = ⋅ ( 9 − 10γ 2 − 15γ 4 + 16γ 5 )
1 1
K γ 60
E ( X γ Yγ ) = ⋅ ( 4 − 10γ 3 + 6γ 5 )
1 1
K γ 30
From these results we find that

21γ 2 − 18γ − 3
ρ ( X γ , Yγ ) = −
107γ 2 + 114γ + 19
lim ρ ( X γ , Yγ ) = lim ρ ( X γ , Yγ ) = 0 .
3
and
γ →0 19 γ →1
4.1.3 Non-uniform Density Function II
The normalizing factor (integral of C ( x , y ) − γ over [C] ) is

γ
1
Kγ = (1− γ )
3
3
 1
 K ( C ( x, y ) − γ ) if ( x, y ) ∈ [C ]
γ
fγ ( x, y ) =  γ
0
 otherwise
1  x2 γ2 
  x − − γ +  if γ ≤ x ≤ 1
f1 ( x ) =  Kγ  2 2 ,

0 otherwise
 1  y2 γ2 
  y − − γ +  if γ ≤ y ≤ 1
f 2 ( y ) =  Kγ  2 2 

0 otherwise
E ( X γ ) = E (Yγ ) =
1 1
⋅
K γ 24
( 5 − 12γ + 6γ 2 + 4γ 3 − 3γ 4 )
E ( X γ 2 ) = E ( Yγ 2 ) = ( 9 − 20γ + 10γ 2 + 5γ 4 − 4γ 5 )
1 1
⋅
K γ 60
E ( X γ Yγ ) = ⋅ ( 8 − 15γ + 10γ 3 − 3γ 5 )
1 1
K γ 60
ρ ( X γ , Yγ ) ≡
3
19
4.2 Lukasiewicz t-Norm

In this example we define the joint possibility distribution by the Lukasiewitz t-
norm, that is
 x + y − 1 if 0 ≤ x , y ≤ 1 and x + y ≥ 0,
C ( x , y ) = max { A ( x ) + B ( y ) − 1, 0} = 
0 otherwise
The γ-level set of C is
[C]γ = {( x, y ) ∈ 2 | γ ≤ x, y ≤ 1, x + y ≥ 1+ γ } .

γ
1
 if ( x, y ) ∈[C ]
γ
f ( x, y ) = Tγ
0 otherwise

where
(1 − γ )
2
Tγ =
2
1 1
 ( x − γ ) if γ ≤ x ≤ 1  ( y − γ ) if γ ≤ y ≤ 1
f1 ( x ) = Tγ , f2 ( y ) = Tγ
0 0
 otherwise  otherwise

2+γ
E ( X γ ) = E (Yγ ) =
3
γ 2
+ 2γ + 3
E ( X γ 2 ) = E ( Yγ 2 ) =
6
( γ + )( 1)
γ +
E ( X γ Yγ ) =
5
12
From these results we can compute the correlation coefficient:
ρ ( Xγ ,Yγ ) ≡ −
1
2

γ
1
Kγ = (1 − γ ) (1 + 2γ )
2
6
1
 K ( x + y − 1) if ( x, y ) ∈ [C ]
γ
fγ ( x, y ) =  γ
0
 otherwise
1 1 2 2 1 1 2 2
 K ⋅ 2 ( x − γ ) if γ ≤ x ≤ 1  ⋅ ( y − γ ) if γ ≤ y ≤ 1
f1 ( x ) =  γ , f2 ( y ) =  Kγ 2 .
0 
 otherwise 0 otherwise
E ( X γ ) = E (Yγ ) = ⋅ (1 − γ 2 )
1 1 2
Kγ 8
E ( X γ 2 ) = E (Yγ 2 ) = ⋅ ( 3 − 5γ 2 + 2γ 5 )
1 1
K γ 30
E ( X γ Yγ ) =
1
⋅
1
K γ 120
(1 + 22γ + 23γ 2 + 4γ 3 )

13γ 2 + 6γ + 1
ρ ( X γ , Yγ ) = −
19γ 2 + 18γ + 3
lim ρ ( X γ , Yγ ) = − lim ρ ( X γ , Yγ ) = − .
1 1
and
γ →0 3 γ →1 2

γ
1
Kγ = (1 − γ )
3
6
1
 K ( x + y − 1 − γ ) if ( x, y ) ∈ [C ]
γ
fγ ( x, y ) =  γ
0
 otherwise
1 1 1 1
 K ⋅ 2 ( x − γ ) if γ ≤ x ≤ 1  K ⋅ 2 ( y − γ ) if γ ≤ y ≤ 1
2 2
f1 ( x) =  γ , f2 ( y ) =  γ .
0 0
 otherwise  otherwise
E ( X γ ) = E (Yγ ) =
1 1
(1 − γ ) ( γ + 3 )
3
⋅
K γ 24
E ( X γ 2 ) = E (Yγ 2 ) = ⋅ (1 − γ ) ( γ 2 + 3γ + 6 )
1 1 3
K γ 60
E ( X γ Yγ ) = (1 − γ ) (1 − γ ) ( γ 2 + 8γ + 11)
1 1 3 3
⋅
K γ 120
ρ ( Xγ , Yγ ) ≡ − .
1
3
4.3 Larsen t-Norm

In this case we define the joint possibility distribution C by the product t-norm,
 x ⋅ y if 0 ≤ x , y ≤ 1,
C ( x, y ) = A ( x ) B ( y ) =  .
0 otherwise
Then a γ-level set of C is
[C]γ = {( x, y ) ∈ 2 | 0 ≤ x, y ≤ 1, xy ≥ γ }
For general case of triangular fuzzy numbers with multiplicative joint distribution
see [17].

γ
1
T if ( x, y ) ∈[C ]
γ
f ( x, y ) =  γ .
0 otherwise

where
Tγ = 1 − γ + γ ln γ .
 1 x −γ  1 y −γ
T ⋅ x if γ ≤ x ≤ 1  ⋅ y if γ ≤ y ≤ 1
f1 ( x ) =  γ , f 2 ( y ) = Tγ .
0 
 otherwise 0 otherwise
1 (1 − γ )
2
E ( Xγ ) = E (Yγ ) = ⋅
Tγ 2
E ( X γ 2 ) = E (Yγ 2 ) =
1 1
⋅ (1 − γ ) ( γ + 2 )
2
Tγ 6
E ( X γ Yγ ) = ⋅ (1 − γ 2 + 2γ 2 ln γ )
1 1
Tγ 4
From these results we can determine the correlation coefficient:
Tγ (1 − γ 2 + 2γ 2 ln γ ) − (1 − γ )
4
ρ ( X γ , Yγ ) =
2
Tγ (1 − γ ) ( γ + 2 ) − (1 − γ )
2 4
3
and
lim ρ ( X γ , Yγ ) = 0 lim ρ ( X γ , Yγ ) = − .
1
and
γ →0 γ →1 2

γ
1 1 1
Kγ = − γ 2 + γ 2 ln γ .
4 4 2
1
 K xy if ( x, y ) ∈[C ]
γ
fγ ( x, y ) =  γ .
0
 otherwise
 1 1 γ2   1 1 γ2 
 ⋅  x −  if γ ≤ x ≤ 1  ⋅  y −  if γ ≤ y ≤ 1
f1 ( x ) =  Kγ 2  x  , f2 ( y ) =  Kγ 2  y .
 
0 otherwise 0 otherwise
E ( X γ ) = E (Yγ ) = ⋅ (1 − 3γ 2 + 2γ 3 )
1 1
Kγ 6
E ( X γ 2 ) = E ( Yγ 2 ) = ⋅ (1 − γ 2 )
1 1 2
Kγ 8
E ( X γ Yγ ) = ⋅ (1 − γ 3 + 3γ 3 ln γ )
1 1
Kγ 9
For the correlation coefficient we get the following:
4 K γ (1 − γ 3 + 3γ 3 ln γ ) − (1 − 3γ 2 + 2γ 3 )
2
ρ ( X γ , Yγ ) =
K γ (1 − γ 2 ) − (1 − 3γ 2 + 2γ 3 )
9 2 2
2
and
1
and
γ →0 γ →1 2

γ
1 3 1
Kγ = − γ + γ 2 − γ 2 ln γ
4 4 2
1
 K ( xy − γ ) if ( x, y ) ∈ [C ]
γ
fγ ( x, y ) =  γ
0
 otherwise
 1 1 γ2 
 ⋅  x − 2γ +  if γ ≤ x ≤ 1
f1 ( x ) =  Kγ 2  x  ,

0 otherwise
 1 1 γ2 
 ⋅  y − 2γ +  if γ ≤ y ≤ 1
f 2 ( y ) =  Kγ 2  y

0 otherwise
E ( X γ ) = E (Yγ ) =
1 1
⋅ (1 − γ )
3
Kγ 6
E ( X γ 2 ) = E (Yγ 2 ) = ⋅ ( 3 − 8γ + 6γ 2 − γ 4 )
1 1
K γ 24
E ( X γ Yγ ) = ⋅ ( 4 − 9γ + 5γ 3 − 6γ 3 ln γ )
1 1
K γ 36
For the correlation coefficient we get a long expression again:
K γ ( 4 − 9γ + 5γ 3 − 6γ 3 ln γ ) − (1 − γ )
6
ρ ( X γ , Yγ ) =
K γ ( 3 − 8γ + 6γ 2 − γ 4 ) − (1 − γ )
3 6
2
and
1
and
γ →0 γ →1 3
5 Conclusion
We have introduced alternative definitions for the measure of dependence between
fuzzy numbers. These measures are based only on the information included in the
joint possibility distribution.
Acknowledgments. This work was supported in part by the project TAMOP 421B at the
Széchenyi István University, Győr.
References
[1] Jaynes, E.T.: Probability Theory: The Logic of Science. Cambridge University Press
(2003)
[2] Dubois, D., Prade, H.: The mean value of a fuzzy number. Fuzzy Set Syst. 24,
279–300 (1987), doi:10.1016/0165-0114(87)90028-5
[3] Carlsson, C., Fullér, R., Majlender, P.: On possibilistic correlation. Fuzzy Set
Syst. 155, 425–445 (2005), doi:10.1016/j.fss.2005.04.014
[4] Fullér, R., Majlender, P.: On interactive fuzzy numbers. Fuzzy Set Syst. 143,
355–369 (2004), doi:10.1016/S0165-0114(03)00180-5
[5] Dubois, D.: Possibility theory and statistical reasoning. Comput. Stat. & Data Analy-
sis 5, 47–69 (2006), doi:10.1016/j.csda.2006.04.015
[6] Fullér, R., Mezei, J., Várlaki, P.: An improved index of interactivity for fuzzy num-
bers. Fuzzy Set Syst. 165, 56–66 (2011), doi:10.1016/j.fss.2010.06.001
[7] Fullér, R., Mezei, J., Várlaki, P.: Some Examples of Computing the Possibilistic Cor-
relation Coefficient from Joint Possibility Distributions. In: Rudas, I.J., Fodor, J.,
Kacprzyk, J. (eds.) Computational Intelligence in Engineering. SCI, vol. 313, pp.
153–169. Springer, Heidelberg (2010) ISBN 978-3-642-15219-1, doi:10.1007/978-3-
642-15220-7
[8] Liu, S.T., Kao, C.: Fuzzy measures for correlation coefficient of fuzzy numbers.
Fuzzy Set Syst. 128, 267–275 (2002)
[9] Hong, D.H.: Fuzzy measures for a correlation coefficient of fuzzy numbers under Tw
(the weakest t-norm)-based fuzzy arithmetic operations. Inform Sciences 176,
150–160 (2006)
[10] Schweizer, B., Sklar, A.: Associative functions and abstract semigroups. Publ. Math.
Debrecen. 10, 69–81 (1963)
[11] Mamdani, E.H.: Advances in the linguistic synthesis of fuzzy controllers. Int. J. Man
Mach. Stud. 8(6), 669–678 (1976), doi:10.1016/S0020-7373(76)80028-4
[12] Lukasiewicz, J.: O logice trójwartosciowej. Ruch filozoficzny 5, 170–171 (1920) (in
Polish); English translation: On three-valued logic. In: Borkowski, L.(eds.): Selected
works by Jan Lukasiewicz. Amsterdam, North-Holland, pp. 87-88 (1970)
[13] Larsen, P.M.: Industrial applications of fuzzy logic control. International Journal of
Man-Machine Studies 12, 3–10 (1980), doi:10.1016/S0020-7373(80)80050-2
[14] Harmati, I.Á.: A note on f-weighted possibilistic correlation for identical marginal
possibility distributions. Fuzzy Set Syst. 165, 106–110 (2011),
doi:10.1016/j.fss.2010.11.005
[15] Fullér, R., Harmati, I.Á., Várlaki, P.: Probabilistic Correlation Coefficients for Possi-
bility Distributions. In: Fifteenth IEEE International Conference on Intelligent Engi-
neering Systems 2011 (INES 2011), Poprad, Slovakia, June 23-25, pp. 153–158
(2011) ISBN 978-1-4244-8954-1, doi:10.1109/INES.2011.5954737
[16] Fullér, R., Harmati, I.Á., Mezei, J., Várlaki, P.: On Possibilistic Correlation Coeffi-
cient and Ratio for Fuzzy Numbers. In: Recent Researches in Artificial Intelligence,
Knowledge Engineering & Data Bases, 10th WSEAS International Conference on
Artificial Intelligence, Knowledge Engineering and Data Bases, February 20-22,
pp. 263–268. WSEAS Press, Cambridge (2011) ISBN: 978-960-474-237-8
[17] Fullér, R., Harmati, I.Á., Várlaki, P.: On Possibilistic Correlation Coefficient and Ra-
tio for Triangular Fuzzy Numbers with Multiplicative Joint Distribution. In: Proceed-
ings of the Eleventh IEEE International Symposium on Computational Intelligence
and Informatics (CINTI 2010), Budapest, Hungary, November 18-20, pp. 103–108
(2010) ISBN 978-1-4244-9278-7, doi:10.1109/CINTI.2010.5672266
Differential Evolution Algorithm in Optimal
Design of Microstrip Filter
Juraj Paleček1, Peter Vestenický2, Martin Vestenický1, and Juraj Spalek2

1
Faculty of Electrical Engineering, Department of Telecommunications and Multimedia,
University of Žilina, Univerzitná 8215/1, 010 26 Žilina, Slovakia
2
Faculty of Electrical Engineering, Department of Control and Information Systems,
University of Žilina, Univerzitná 8215/1, 010 26 Žilina, Slovakia
{juraj.palecek,peter.vestenicky,
martin.vestenicky, juraj.spalek}@fel.uniza.sk
Abstract. This chapter describes utilization of differential evolution algorithm in

process of optimal design of microstrip filters. Main part of the chapter is dedicat-
ed to the definition of cost function, where filter design criteria are inserted to the
process by tolerance scheme and mathematical model used for evaluation of mi-
crostrip filter magnitude response. The solution found is evaluated and compared
to a conventional design in the concluding part of the chapter. Discussion of dif-
ference between realized samples and mathematical model is also presented at the
end of this chapter.
1 Introduction
The continuously increasing demands on the electronic circuits being used in tele-
communications impose raising requirements on their optimal design taking tech-
nological, economical, construction and other limitations into account. Therefore
research must be focused on new methods for optimized design of electronic cir-
cuits [7], [10], [11]. One of the new approaches to this challenge are the stochastic
algorithms. These algorithms are being used for numerical solving of complex and
ambiguous design tasks. The main advantage of stochastic algorithms is the easy
applicability in many areas of research and design. Next major feature is the abil-
ity of effective solving of multi-criterion design tasks with a large number of in-
dependent variables. One of application fields for using the stochastic algorithms
is the design of high frequency microstrip structures for communication systems.
, , , 0,1
,
(1)
, ,
. , , , , . , , (2)
This chapter deals with determination of dimensions of microstrip segments for
low pass filter. For the design purposes the well-known differential evolution (DE)
algorithm has been selected because of its robustness and appropriate representa-
tion of an individual. In the DE algorithm described in [3] changes have been

266 J. Paleček et al.
made to speed up the convergence process. The changes are based on modified
process of creation of an individual according to formula (1) where the vector of
differential mutation is given by formula (2).
2 Creation of Cost Function

The cost function must respect all requirements of filter optimal design as well as
limiting factors which must be considered. The complexity of cost function influ-
ences the computational demandingness and time consumption of the design proc-
ess. One of possibilities to create the cost function is the fulfilment of tolerance
scheme requirements. If the tolerance scheme is given the design process has
higher freedom as in the case when two transfer characteristics are being com-
pared. The costing of actual magnitude response is created by adding all areas
bounded by frequency response curve and the given tolerance scheme. So the cost
function includes the areas which lie out of required tolerance scheme (see Fig. 1).
To compute the size of these areas the rectangular numerical integration method
can be used. The total cost is given by formula (3) where partial costs are given by
formulae (4), (5) and (6) and they are weighted by coefficients c1, c2 and c3. Every
partial cost represents the exceeding of corresponding boundary of tolerance
scheme. This access is advantageous because the design process is not limited by
standard approximations and the transfer function that will meet the requirements
of tolerance scheme can be found. The purpose of design loop using the DE algo-
rithm is to find the minimum of cost function.
Fig. 1. Method of cost function calculation based on filter tolerance scheme
. . .
(3)
| |. (4)
| |. (5)
Differential Evolution Algorithm in Optimal Design of Microstrip Filter 267
| |. (6)
3 Calculation of Filter Transfer Function

For calculation of cost function the transfer function of actual iteration must be
known. The transfer function of microstrip filter can be calculated according to for-
mula (7) where additional parameters are given by (8) and ZS, ZL are the impedances
of source and load. Z0 is the reference impedance for which the S parameters were
calculated. Therefore, the S parameters of actual filter solution must be known. The
overall S parameters SOxx of filter can be calculated from overall T parameters TOxx
according to formula (9) where the overall T parameters are product of particular T
parameters of each microstrip segment Tkxx with respect of segment order - formula
(10) and N is the number of filter segments. The particular T parameters can be cal-
culated from formula (11) from the particular S parameters which are given by for-
mula (12) where the ABCD parameters are given by formula (13).
(7)
(8a),(8b)
(8c)
. .
(9)
∏ (10)
. .
(11)
. . . .
. .
(12)
.
. .
 . _ .  .
(13)
 .  .
_
4 Model of the Microstrip Line Segment

The microstrip line segment is characterized as a segment of long transmission
line by the complex transfer ratio γ which is in general given by the formula (14)
and by the frequency dependent characteristic impedance Z0(f) which is given by
the formula (27). Calculations of these parameters are based on dispersion model
of microstrip line (propagation of TEM wave and influence of TM0 mode are
considered) which is in detail given in [5]. The input parameters of this model are
width W and length l of segment, attributes of used material (height h, thickness t
of conductive layer, its conductivity σ, relative permittivity εr and dissipation fac-
tor tg δ (see Fig. 2).
Fig. 2. Properties of microstrip line segment
(14)
The complex transfer ratio (14) has a real part i. e. the specific attenuation (15)
which consists from losses in conductor αc given by the formula (16) where B is
an auxiliary function (17). The contribution of losses in dielectric αd is given by
(18). In this formula the relative permittivity of used material is εr and the εre(f) is
frequency dependent effective relative permittivity given by (20). Model of the
effective relative permittivity and its frequency dependency is in detail described
in [6].
(15)
(16)
2
·
B= (17)
·
  ·
 27.3 ·

·

· (18)
Imaginary part of the complex transfer ratio γ is the specific phase shift β calcu-
lated from (19). The dispersion effect (i. e. β is nonlinear function of frequency) is
caused by the frequency dependent effective relative permittivity (20). In the for-
mula (20) the relative permittivity εre is given by (21) and it depends only on geo-
metrical dimension of microstrip segment and properties of used material. The f50
in (20) is frequency of 50 % dispersion point defined by (22) according to [6]. Its
value can be calculated for a given microstrip segment from (23) where fTM0 is
critical frequency of TM0 mode. This critical frequency depends on the geometri-
cal dimension of microstrip segment and on material properties according to for-
mula (24). The last parameter in (20) is coefficient m given by formulae (25) and
(26).
··
 (19)

 
  (20)
  
· 0.04 · 1 · , 1
.
·
 (21)
  
· · , 1
.
·
(22)
.
(23)
. . . · ·

·· ·  
tg  ·  
(24)
1 0.32 · . (25)
. ·
.
1 0.15 0.235 · , 0.7
(26)
1, 0.7
The next parameter which influences on the ABCD parameters (13) is characteris-
tic impedance Z0(f) given by (27) which depends on frequency (it is caused by the
dispersion effect).
 
· · (27)
 
In (27) the parameters εr and εre(f) are described above and Z0 is characteristic
impedance dependent only on geometrical dimension of microstrip segment and
on material properties which is described by formula (28).

· 0.25 · , 1
·· 
(28)

· 1.393 0.667 · 1.44 , 1

The symbol We in (28) is an estimation of microstrip segment effective width in

accordance with (29), η is impedance of free space. The effective width of micro-
strip segment replaces the real physical width in quasi-static model of microstrip
structure.
. · ··
· 1 ln ,
 ·
. · ·
(29)
· 1 ln ,
 ·
The used dispersion model is in detail described in [5], moreover, it is used in RF

toolbox of the Matlab software environment. Note that every calculation must be
performed for all selected points of magnitude response.
6 Example
6.1 Requirements
For verification of the described algorithm the low pass microstrip filter design on
substrate FR4 was selected. The requirements on filter and properties of FR4 sub-
strate material are described in the Tab. 1. For comparison, the design was per-
formed by application of commercial design software. In this case the Chebyshev
approximation was used.
Table 1. Filter requirements and substrate properties
Type of filter Low pass Type of material FR4

Height 1.524
Type of approximation Chebyshev
h [mm]
Filter order 5 Thickness, t [µm] 35
Cut-off frequency Permittivity 4.8
862
fz [MHz] εr
Max. ripple in pass band Dissipation factor 0.03
1
amax [dB] tgδ
Stop band frequency Conductivity 59.6 . 106
1.3
fc [GHz] σ [Sm-1]
Stop band attenuation
30
amin [dB]
Terminating impedance
50
Z0 [Ω]
In the design process the algorithm of differential evolution (DE) was used.
Into the DE algorithm the limiting factors for chromosome values of individual
according to formula (30) were applied.
(30)
∈ ,
Table 2. Control constants and weighting coefficients
Constants
Number of unknown variables D 10
Number of generations Gmax 300
Minimum segment length [mm] Xlmin 0.1
Maximum segment length [mm] Xlmax 40
Minimum segment width [mm] Xwmin 0.3
Maximum segment width [mm] Xwmax 19.h
Control Constants
Number of population members NP 200
Crossing probability CR 0.9
Weighting coefficient of differential mutation 1 F1 0.5
Weighting coefficient of differential mutation 2 F2 0.5
Weighting coefficients
Weighting coefficient of cost in pass band 1 c1 1
Weighting coefficient of cost in pass band 2 c2 1
Weighting coefficient of cost in stop band c3 0.12
The application of these limiting criterions is forced by the keeping of accept-

able mechanical dimensions of microstrip segments as well as the acceptable ac-
curacy of microstrip segment mathematical model. The minimum and maximum
chromosome values, control constants of DE algorithm and weighting coefficients
are shown in the Tab. 2. The boundaries limiting chromosome values were se-
lected on the base of technological possibilities in the manufacturing process of
filter samples and limitation factors of microstrip segment model [5]. The control
constants of DE algorithm were selected in accordance with conclusions published
in [1], [3], [4] and the weighting coefficients of the cost function were experimen-
tally selected by authors to minimize the overlapping of magnitude response of the
designed filter in the pass band as well as the stop band.
7 Structure of an Individual
At the design process the dimensions of microstrip segments are being found.
Because the filter is of 5th order it consists of five microstrip segments and each
segment is characterized by its width and length. Every dimension of specific
segment represents one chromosome of an individual. The found shape of resul-
tant microstrip structure and assignment of chromosomes are shown in the Fig. 3.
Fig. 3. Assignment of weights and lengths of microstrip segments to chromosomes of indi-

vidual
8 Obtained Results
The results of design process are the dimensions of particular microstrip segments,
which are given in Tab. 3. For comparison, the dimensions of segments designed
by commercial software are listed in the Tab. 3, too.
On the base of dimensions listed above the samples of both versions of filters
were manufactured, see Fig. 4. The sample designed by the described algorithm is
at the top of figure. The manufactured samples were measured by a vector net-
work analyser and from measured S parameters the magnitude responses were
obtained. The formulae (7) and (8) were used again. The Fig. 5 and Fig. 6 illus-
trate the reached magnitude responses for both method of design, theoretically
calculated as well as practically measured.
Table 3. Dimensions of the microstrip segments
Designed by the described al-

Designed by commercial software
gorithm
Segment l [mm] w [mm] l [mm] w [mm]
1 8 27,6 18,7 14,2
2 14,8 0,3 21 0,882
3 12,5 27,6 18,4 21
4 14,8 0,3 21 0,882
5 8 27,6 18,7 14,2
Fig. 4. Samples of manufactured filters (top – designed by described algorithm, bottom –

designed by commercial software)
Fig. 5. Comparison of theoretically calculated magnitude responses of both filters
Fig. 6. Comparison of measured magnitude responses of both filters
9 Analysis of Mathematical Model Inaccuracy

Note that the theoretical and measured magnitude responses of filter designed by
commercial software (unoptimized) are matched very well but in the case of filter
optimized by the described algorithm there is a certain difference between the
theoretical and measured cut-off frequency. This situation is shown in Fig. 7
where the dashed curve is the measured magnitude response of filter sample and
the solid curve is theoretically calculated magnitude response taking geometrical
dimension from Tab. 1 and Tab. 3 into account. During analysis of this difference
the manufactured sample was checked by a stereomicroscope whereby some de-
viations from theoretical segment width (0.26 mm instead of 0.3 mm) and defects
in conductive layer appeared, see Fig. 8.
274 J. Paleček et aal.
After applying the meeasured dimension of microstrip segment into the filter
model the dotted curve in Fig. 7 was calculated. As it can be seen the deviations oof
dimensions cause approx ximately 50 % of total difference between the real annd
theoretical magnitude resp ponses (full and dashed curves in Fig. 7).
To identify other influeences on filter characteristics the influence of connectorrs,
the coupling between co onnectors and microstrip segments and used materiial
(thickness, relative permitttivity and dissipation factor) on the mathematical moddel
must be identified.
Fig. 7. Comparison of measu

ured and theoretical magnitude responses of optimized filter
Fig. 8. Defectiveness of man

nufactured filter sample
10 Conclusion
At the conclusion it is possible to say that the design task of described type has
positive results especially the increasing of attenuation in the stop band, shift of
the real cut off frequency in the direction to the required cut off frequency, higher
slope of magnitude response in the transition band and minimization of the me-
chanical dimensions of the filter. Next works will be focused on improvement of
the filter mathematical model by adding the model of connectors and its coupling
to microstrip segment. The improvements will be practically verified by manufac-
turing and measuring of new samples of filters.
References
[1] Bäck, T., Fogel, D.B., Michalewicz, Z.: Handbook of Evolutionary Computation. Ox-
ford University Press, Bristol (1997) ISBN: 0-7503-0392-1
[2] Collin, R.E.: Foundations for Microwave Engineering, 2nd edn. John Wiley and
Sons, Inc., Hoboken (2001) ISBN: 0-7803-6031-1
[3] Corne, D., Dorigo, M., Glover, F.: New Ideas in Optimization. McGraw - Hill, Lon-
don (1999) ISBN: 0-0770-9506-5
[4] Duľová, O., Zolotová, I.: Using Greedy and Evolution Algorithms for Active Con-
tour. In: Process Control 2008: Proceedings of the 8th International Scientific - Tech-
nical Conference, June 9-12, pp. c085a-1–c085a-9. University Pardubice, Kouty nad
Desnou (2008) ISBN: 978-80-7395-077-4
[5] Gupta, K.C., Garg, R., Bahl, I., Bhartia, P.: Microstrip Lines and Slotlines, 2nd edn.
Artech House, Norwood (1996) ISBN: 0-89006-766-X
[6] Kobayashi, M.: A Dispersion Formula Satisfying Recent Requirements in Microstrip
CAD. IEEE Transactions on Microwave Theory and Techniques 36(8), 1246–1250
(1988) ISSN: 0018-9480
[7] Krischuk, V., Farafonov, A., Shilo, G., Gaponenko, N.: Optimization of Microstrip
Filters Tolerances. In: Proceedings of the 7th International Conference The Expe-
rience of Designing and Application of CAD Systems in Microelectronics, CADSM
2003, pp. 251–252. Publishing House of Lviv Polytechnic National University, Lviv
(2003) ISBN 966-553-278-2
[8] Paleček, J., Vestenický, P., Vestenický, M., Spalek, J.: Optimization of Microstrip
Filter Dimensions by Differential Evolution Algorithm. In: Proceedings of the 15th
IEEE International Conference on Intelligent Engineering Systems INES 2011, Po-
prad, Slovakia, June 23-25, pp. 173–176 (2011) ISBN: 978-1-4244-8954-1
[9] Sayre, C.W.: Complete Wireless Design, 2nd edn. McGraw - Hill, New York (2008)
ISBN: 978-0-07-154453-5
[10] Shen, Y., Sans, C.: New Advances in HTS Microstrip Filter Design. In: IEEE MTT-S
International Microwave Symposium Digest, vol. 3, pp. 1885–1888 (2003) ISSN
0149-645X
[11] Singhal, S.K., Sharma, D., Dhariwal, R.M.: Design of 1.3 GHz Microstrip Highpass
Filter Using Optimum Distributed Short Circuited Stubs. In: First International Con-
ference on Computational Intelligence, Communication Systems and Networks, In-
dore, India, pp. 264–267 (2009) ISBN: 978-0-7695-3743-6
P-Single Operators in Pipeline System
of DF-KPI Architecture
Liberios Vokorokos, Norbert Ádám, and Branislav Madoš

Department of Computers and Informatics, Letná 9, 042 00 Košice, Slovakia
{liberios.vokorokos,norbert.adam,branislav.mados}@tuke.sk
Abstract. There has been a resurgence of interest in data flow architectures, be-
cause the dataflow model of execution offers attractive properties for parallel
processing. In data flow architectures the computing process is managed by the
operands flow accessed on different levels for executing instructions of dataflow
program. Data flow architectures have been traditionally classified as either static
or dynamic. The architecture described in this article belongs to a class of dynamic
data flow architectures, in which the operand process control significantly affects
the performance parameters as well as the system characteristics of the given ar-
chitecture. From the many types of operators, this article provides microprogram
managing for P-single operators in pipeline system of DF-KPI architecture.
1 Introduction
Microprocessor industry finds itself in its quest for continuous growth of micro-
processor performance at the crossroads today. For decades we have been witness-
ing a continuous increase in performance of microprocessors in accordance to the
Moore’s law. Increasing number of integrated transistors along with increasing
clock rates of chips were drawing up performance and energy efficiency of inte-
grated circuits.
Nowadays, we are reaching physical limits of silicon technology and it is be-
lieved that we will not be able to produce reliable silicon transistors with struc-
tures of size under 10 nm. As transistors are made smaller, it gets more difficult to
control how electrons are moving through the silicon channels to turn transistors
on and off. We need to find the way how to assure continuation of the process of
exponential increase of availability of raw transistor count on integrated circuits.
One of solutions is to move processor industry from flat surface of silicon die
into the third dimension and to build 3D silicon transistors. Another promising
solution is to build three-dimensional circuits by stacking silicon dies one on top
of another. This technology resembles construction of multi-storey buildings of
cities and can dramatically increase the number of transistors integrated into the
chip, with positive effect on other aspects of architectures such as shortening of
the length of on chip wiring or decrease of power consumption. 3D technology
also allows integration of heterogeneous technologies on the same 3D chip, for

278 L. Vokorokos et al.
example core layers with memory layers or analog and digital layers. In finding
solutions beyond silicon era and 10 nm frontier, we can see effort to build inte-
grated circuits based on carbon nanotube transistors (CNT) [3].
Exponentially growing number of transistors on the integrated circuit provides
opportunity to increase the performance of the chip. Unfortunately, conventional
superscalar architecture of microprocessor cannot be simply scaled up in line with
this trend, and this situation results in challenging problem how to maintain conti-
nuity in proportion between quantity of integrated transistors and performance of
the chip.
Disproportion between communication performance and computation perfor-
mance intensifies with miniaturization of transistors, because of side effect of
relative lengthening of inside chip wiring. Centralized control of microprocessor
intensifies the problem. Memory model using cache memory causes extensive
associative searches.
Increasing circuit complexity of superscalar microprocessors with longer de-
sign times, complex tests of designs and increasing amount of design bugs are also
negative aspects of the effort to increase the performance of superscalar micropro-
cessors by integration of more transistors into single processor core.
Pushing trend that allows overcome outlined hurdles is multi-core design with
integration of tens of cores in tiled configuration. Introduced in [22], this trend of
tiled architectures can be characterized by multiple use of processing elements,
memory elements, input/output elements and various types of interconnecting
networks.
Representative of general purpose computing microprocessors is Tile64 de-
signed by Tilera Corporation, described in [22]. Tile64 integrates 64 general pur-
pose cores called tiles. Each core integrates L1 and L2 cache. Tiles are arranged in
8 × 8 bi-dimensional mesh using interconnecting network with 31Tbps data
throughput. Chip utilizes 3-way VLIW pipeline for instruction level parallelism.
Each tile is able to independently run operating system, or multiple tiles are able
to run multiprocessing operating system. Performance of the chip at 700 MHz is
443.109 Operations Per Second (BOPS).
Intel microprocessor TeraScale Processor is designed under Tera-Scale Compu-
ting Research Program and is described in [12]. Terascale Processor integrates 80
cores in bi-dimensional mesh in 8 × 10 cores organization. With 65 nm processor
technology implemented, chip integrates 100 million transistors on 275 mm2 die.
Processor performs 4 FLOP per cycle per core, and with 4.27 GHz delivers theo-
retical peak performance of 1.37 TFLOPS in single precision. Instructions set
architecture consists of 16 instructions and Terascale Processor uses 96 bit VLIW.
Each core runs own program and interconnection network transfers data and coor-
dination instructions for execution of programs between cores via message
passing.
Representative of DSP microprocessors with tile organization is VGI. Organi-
zation of 64 cores is 8 × 8 in bi-dimensional mesh and each core contains approx-
imately 30 000 transistors. VGI utilizes data flow paradigm of computation.
Whereas described architectures are integrating tens of cores, with TRIPS,
RAW, SmartMemories, nanoFabrics or WaveScalar representatives, trend is
P-Single Operators in Pipeline System of DF-KPI Architecture 279
heading towards integration of hundreds or even thousands of simple cores or

processing elements (PE), often arranged along with memory elements in the grid
[1, 11, 13, 14].
Described architectures are considering both control flow model of computa-
tion also with use of very large instruction word (VLIW) and data flow computa-
tion model.
Data flow architectures are gaining attention in connection with multi-core and
multi-CPU architectures and related requirement for new paradigms of program-
ming which are able to bring more efficient process of parallel programming and
parallel program execution on these architectures.
The proposed architecture described in this article is based on data flow compu-
tational model, in which the computation is controlled on the basis of data flow
processing.
2 Data Flow Computational Model

Recall that the von Neumann architectures are control driven, and in them, data
are passive and fetched only when an instruction needs them. That is, the opera-
tion of the machine is dedicated by the instruction sequence (i.e., the program).
Data flow architectures, on the other hand, are data driven because the readi-
ness of data dictates the operation of the machine. That is, the data flow program
instructions are waiting passively for the arrival of a certain combination of argu-
ments, which are made available as a data control stream in a data-driven sense.
The waiting interval, where a program instruction waits for the arrival of ope-
rands, represents its selection phase, during which the allocation of computing
resources occurs. The fundamental idea behind the data flow computational model
is the mapping of tasks to the computing elements, which can increase the rate of
parallelism. In general, it is necessary to decompose the computational process
into smaller communicating processes represented by the data flow program [18].
The data flow program is represented by its machine representation, called a
data flow graph. Data flow graphs are directed graphs that show the data depen-
dencies between the respective instructions (operators). Their nodes represent
instructions and the arcs connecting these nodes represent the operands of these
instructions.
The implementation of the data flow computer architecture depends on the
form of execution of the data flow program instructions, which is implemented as
a process of receiving, processing and transmission of data tokens.
There are static and dynamic models of pure data flow architectures, depending
on the concept of processing of data tokens or on the scope of architectural sup-
port for its implementation.
In the static approach [2, 17, 18] a node (instruction) can be executed only
when all of the tokens are available on its input arcs and no tokens exist on any of
its output arcs. This model is capable of using data parallelism and pipelining
techniques; therefore this model has found use in applications with regular compu-
ting structures.
In the dynamic data flow model [4, 8, 18] the operator represented by a node is
executable in the dynamic data flow model if all the input edges contain tokens,
the symbols of which are identical. Each edge may contain more than one labeled
token in this model. When executing a node, the tokens belonging together are
removed from the input edges and a token with the corresponding symbol of the
output edge is generated. The dynamic data flow model uses both loop parallelism
and recursive parallelism, which appear dynamically during program run-time.
3 Architecture of the DF-KPI System

The task the computer designer faces is a complex one: Determine what attributes
are important for a new computer, then design a computer to maximize perfor-
mance while staying within cost, power, and availability constraints. This task has
many aspects, including selection of the technology [5, 6, 15], control program
model [9, 10], instruction set design, functional organization, logic design, and
implementation [7, 8]. The implementation may encompass integrated circuit
design, packaging, power, and cooling. Optimizing the design requires familiarity
with a very wide range of technologies, from compilers and operating systems to
logic design and packaging [7]. The DF-KPI system [8, 19], being developed at
the Department of Computers and Informatics at the Faculty of Electrical Engi-
neering and Informatics of the Technical University of Košice, has been designed
as a dynamic system with direct operand matching. The combination of the local
control flow model (von Neumann’s principle) with the global data flow model
allows us to effectively organize the parallel implementation of functional pro-
gram. The architecture model of the DF-KPI computer is a part of a data flow
complex system, which includes support components for data flow computing
environment for the implementation of the defined application targets.
The structural organization (Fig. 1) of the DF-KPI computer architecture model
consists of the following components:
Coordinating Processors (CP) are intended to manage, coordinate and process
instructions of the data flow program, based on the presence of their operands,
which are enabled at the CP.DI input port of the coordinating processor - either
from its CP.DO output port or from the CP.DO output ports of other CPs through
an interconnection network, or from a Data Queue Unit and from the Frame Store.
The structure of the CP is a dynamic pipelined multiple-function system.
The Data Queue Unit (DQU) is a unit designed to store the activation symbols
(data tokens), which represent operands waiting for matching during program
execution.
The Instruction Store (IS) is a memory of instructions of the data flow program,
in the form of a proper data flow graph.
The Frame Store (FS) is a memory of matching (pairing) vectors, by means of
which the CP detects the presence of operands to perform the operation defined by
the operator (node) in the data flow graph.
Fig. 1. The DF-KPI System
The short description of the item format of MV matching vector in the FS is

FS::= AF V, where AF is a flag of the operand's presence (Affiliation Flag)
and V is the value of the given operand.
Supporting components of the data flow system are needed to create a realistic
computing environment. In the given architecture they are formed by the
following:
The Main computer (HOST) provides standard functions of the computer sys-
tem during data flow computing process.
The Information Technology unit is a unit used to create dedicated application
environments (virtual reality, diagnostics, e-learning).
The I / O processors for fast direct inputs/outputs into the data flow module
(standard I/Os are implemented by the main computer).
The structure of the CP is a dynamic system with pipeline processing, com-
posed of load, fetch, operate, matching and copy segments also indicates the states
of system.
The Load segment is used for loading the data token and its preparation for fur-
ther processing in the FETCH segment. This segment is the first segment of the
processor.
The Fetch segment reads the word DFI from the instruction memory IS. The
word DFI defines the format of the data flow instruction. This segment is reacha-
ble by the processor from the segment Load and Operate. In case that the segment
Operate and segment Load requires the access to the Fetch segment a conflict
occurs. The priority system (Table 1) decides which segment will be preferred.
The Matching segment ensures matching of operands on the basis of flag

in DFI.
Table 1. Priority for segment fetch
Segment
O→F high priority
M→F high priority
L→F low priority
The Operate segment handling the data token processing based on the opera-
tion code OC stored in word DFI. The execution units of the coordinating proces-
sor are accessed from this segment. The result of the operation is sent to the Load
segment. In case that the result of the operation is intended to matching in the FS
memory, then it is sent to the Matching segment. If the result is intended to Load
and this segment is occupied, the token will be sent to another processor by the
interconnection network, or will be saved into the DQU.
Instructions needed for copying the operation results are processed in the Copy
segment.
4 Data Flow Instructions

The program instructions are stored in the Instruction Store. Each instruction is
represented by an operator of the data flow program. Data flow operators are gen-
erally classified as single input, double input and N (N ≥ 3) input operators. In the
case of single input operators, the input is formed by one operand, which meets
the condition of the data availability firing rule, meaning operand matching is not
necessary. In the case of N input operators, the operator will be executed if all
input operands are present, which are then merged in matching segment, based on
the matching vector MV. The instruction set of the DF-KPI architecture consists
of instructions represented by the single input operators (accept, if, kill, out, ret,
sel, unOp), double input (binOp, case, def, gate, load, send, tup) and N (N ≥ 3)
input (apply, constr). The “accept” operator is an entry operator of the data flow
program and subprograms. The if operator represents a two-way branch. The case
operator represents a multi-way switch. The kill operator is used for consumption
of input operand without any response. The out operator is the last operator and
ret is a return operator of the data flow program or subprograms. The sel operator
selects data from the data structure defined by the constr operator. The unOp sin-
gle input operator and the binOp double input operator represent unary and binary
arithmetic operations. The def operator defines the program constant. While load
creates copies of the data and address part of the input operand, the tup operator
creates copies only of the data section of a data token. Running the program or
subprogram is launched by the apply operator. A more detailed description of
these operators is in [8]. The format of the data flow instructions is as follows
DFI ::= OC LI {DST , [XI ]}n (1)
where
OC – operation code;
LI – literal (e.g. number of copies of the result);
DST – target address for operation result;
IX – matching index for the operations.
The data flow program instruction represented by a data token is stored in the
Instruction Store at the address defined by DST field. The data token has the fol-
lowing format
DT ::= P T , V MVB {DST , [XI ]} (2)
where
P – priority of the data token;
T – data type of operand;
V – value of operand;
MVB – base address of matching vector in the FS;
DST – destination of the resulting data token.
The structure of the DST field is the following
DST ::= MF IP ADR (3)
where
MF – matching function, with a defined set of
labels {M, B}, M stands for matching (of two DTs), B stands for by-
pass (without DT matching);
IP – input port {L(eft), R(ight)};
ADR – address of the operator or function;
IX – matching index for an input data token.
If the operands enter the two-input or multi-input operators, operand matching
occurs. The DF-KPI architecture uses the direct operand matching control me-
chanism. It is based on the allocation of a Matching Vector in the Frame Store
according to the activation code (procedure, call). Allocated Matching Vectors are
represented as a matching record in the Frame Store. The format of the Matching
Vector in the Frame Store is as follows
FS [B ACT + H + IX + 1] ::= RC, MVS BOLD DSTRET D{[B NEW ]}{D} (4)
where
BACT – pointer to the current top record in FS;
H – size of a header of record;
MVS – size of a matching vector;
RC – reference counter;
BOLD – pointer to the previous token;
DSTRET return address;

BNEW – base address for new matching record;
D – operand value.
The RC field is set according to the size of the matching vector at compile-time.
After the function associated with the operator has fired, the value of RC is de-
cremented. If RC = 0 , the Matching Vector in the frame store is released.
5 Operand Matching
One of the most important steps based on the dynamic data flow model is direct
operand matching [8, 18]. The concept of direct operand matching represents the
elimination of the costly process (in terms of computing time) related to associa-
tive searching of the operands. In this scheme, a matching vector is dynamically
allocated in the Frame Store memory for each token generated during the execu-
tion of the data flow graph. The current location of a matching vector in the Frame
Store is determined at compile time, while the Frame Store location is determined
after the program starts. Each calculation can be described using an instruction
address (ADR) and the pointer to the matching vector MVB in the Frame Store.
The MVB, ADR value pair is part of the token. A typical action is the searching
for the operands pair in the Frame Store. The matching function provides search-
ing for the tokens marked identically.
Fig. 2. The operand matching (a – P-single input, b – P-double input, c – P-double input/C-
single input, d – P-double input/C-double input, e – P-double input/C-u-single,v-double
input)
After the operand has arrived to the Coordinating Processor, the matching function
detects if a commonly entered operand is present in the Frame Store. Detection is
performed according to matching IX index. If the operand is not there yet, it is

stored in the Frame Store, in the Matching Vector specified by base address of the
MVB operand, into the item specified by index IX.
The operand matching process control at the operator input is influenced by the
process of matching, instruction execution and generation of a result at its output.
Using a compiler producing DFG output with forward searching that allows for
the detecting and eliminating of redundant computations and change order of to-
ken processing, process control can be defined as the transition of activation signs
along the edges of the data flow graph (Fig. 2), between the “producer” (P) opera-
tor and the “consumer” (C) operator.
In this article we describe the operand matching control for configuration
shown in Fig. 2a.
6 Process Control for P-Single Operators

The binary information stored in DF KPI system can be classified as either data or
control information. The main components of this system are the CPs, wherein the
data is manipulated in a datapath by using microoperations implemented with
register transfers. These operations are implemented with adder/subtractors, shif-
ters, registers, multiplexers and buses. The control unit of the CP provides signals
that activate the various microoperations within the datapath to perform the speci-
fied processing tasks. The control unit of the CP also determines the sequence in
which the various actions are performed.
Fig. 3. State (Mealy) Machine

The control unit that generates the signals for sequencing the microoperations is
a sequential circuit with states that dictate the control signals for the system. At
any given time, the state of the sequential circuit activates a prescribed set of mi-
crooperations. Using status conditions and control inputs, the sequential control
unit determines the next state. The digital circuit that acts as the control unit pro-
vides a sequence of signals for activating the microoperations and also determines
its own next state.
The control unit of the CP allows transition between different states denoted as
load, matching, copy, fetch and operate. These states (Fig. 3) are represented by
different segments (l – load, m – matching, c – copy, f – fetch, o – operate) of the
control unit in different order. The segments work in overlapped manner. Transi-
tions between segments are controlled by the microprogram.
Formal notation of the microprogram (which results from the functional system
specification and its decomposition to operational/datapath and control unit) is
represented by the program scheme. Program model scheme of the operational
part is expressed as a sequence of marked pairs
ni : p i (5)
where
ni - are labels,
p i - are instruction(s).
Instructions p i in respect to (5) display different elementary instructions, micro-

instructions, which initiate execution of different elementary operations, micro-
operations. Further following basic command types are defined
X /n (6)
α1 : n1 , α 2 : n2 ,, α k : nk (7)
α1 : X 1 / n1 , α 2 : X 2 / n2 , , α k : X k / nk (8)
STOP (9)
where
STOP – special operation;
X , X i – operations or microoperations;
n, ni – labels;
α ,α i – predicates (conditions), with the characteristics α i ∧ α j = 0, i ≠ j.
Command (6) shows the microinstruction intended for the execution of microope-
ration, after which a transition is made to an instruction marked with the label n.
Commands (7 – 8) represent control instructions (microinstructions) intended for
execution of branch and conditional jumps in the program (microprogram). These
commands test the condition defined by the predicate. If the condition α i is valid
a transition is made to an instruction marked with the label ni (7) or an operation
X is simultaneously executed (8).
Let the execution of the operations X i be launched by the control word R i con-
sisting of the control signal sequence
Ri = Rm,1 Rm,2  Rm,k , ∀Rm, j ∈ {0,1}, j = 1,2,, k .

~ ~ ~ ~
(10)
Let the predicates α p be represented by the status information word
Lp = Ln,1Ln,2 Ln,q , ∀Ln,r ∈ {0,1}, r = 1,2,, q .

~ ~ ~ ~
(11)
Then the micro program of executed P-single input operators DFG (Fig. 2),
through microoperations for the various stages of multifunctional pipeline unit has
a form
l1 : R1 / f1 o5 : R10 / l 2
f1 : L1 : R 2 / f 2 , L1 / f1 l 2 : L6 / l3 , L6 / l 2
f2 : R3 / f3 l3 : L7 / l 6 , L7 / l 4
f3 : R 4 / f4 l 4 : L5 / l 7 , L5 / l5
f 4 : R 5 / o1 l 5 : L 4 / l8 , L4 / l 2 (12)
o1 : L2 : R 6 / o 2 , L2 / o1 l 6 : L3 : R11 / f1 , L3 / l 6
o 2 : L3 : R 7 / o5 , L3 / o3 l 7 : L3 / l1 , L3 / l 7
o3 : L4 : R 8 / o5 , L4 / o 4 l8 : L3 : R12 / f1 , L3 / l8
o 4 : L5 : R 9 / o5 , L5 / o 4
The labels l, f, o represent the segments Load, Fetch and Operate. The microopera-
tions and predicates of individual control words and status words are listed in
Table 2.
The function isFree (X) tests the busy state of segment X. Micro-operations,
which can be executed in parallel, are placed in a single command block
(processing steps) in the program flowchart. Initialization of the coordinating pro-
cessor is done by sign Init = 1. The boot command of the data flow program loads
the data token from the DQU to the LOAD segment, sets the busy flag for the
LOAD segment to 1 (i.e. the LOAD segment is occupied) and blocks the
processing of the following tokens (GetDT = 0). If the next segment, the Fetch
segment, is free, the token is loaded into the Load/Fetch register. After that, the
micro program releases the LOAD segment and activates the loading of other
tokens into the coordinating processor. The control mechanism determines the DF
address operator based on the LFR.DST.ADR address. The operator will be
loaded from the instruction store into the Fetch/Operate register. If the Operate
segment is not busy (isFree (Operate) = true), the operator is fetched from the
Fetch/Operate register and processed.
In the next step, if the CP is not busy, the result of the operator consumption
and processing is available for processing in the same CP. Otherwise; the result is
to another CP through the interconnection network. If all CPs are busy, the token
is stored in the DQU. The proposed architecture at a logical level of operand
matching control is show in Fig. 4. FIFO registers with the following specifica-
tions have been inserted to increase the throughput coefficient between the various
stages of coordinating processor: between the stages L and F → register LFR;
between the stages F and O → register FOR.
Table 2. Control words
Control words Microoperations Status words Microoperations

Load_free:=0
R1 = R1,1R1,2 R1,3 GetDT:=0 L1 isFree(Fetch) = 1
CP.DI:=DQU.OUT
Fetch_free:=0
R = R2,1R1,1 R1,2
2
Load_free:=1 L2 isFree(Operate) = 1
GetDT:=1
R = R3,1
3
LFR:=CP.DI L3 isFree(Load) = 1
R 4 = R4,1 IS.ADR:=LFR.DST.ADR L4 isEmpty(ICN) = 1
FOR.OC:=IS[ADR].OC
FOR.DST:= IS[ADR].DST
R 5 = R5,1 R5, 2 R5,3 R5,4 L5 isEmpty(DQU) = 1
FOR.LI:= IS[ADR].LI
FOR.LD:=LFR.D.V
CP.DO.D:= PEn(FOR.LD,-)
R = R6,1R6,2
6
CP.DO.DST:=FOR.DST L6 GetDT = 1
CP.DO.IX:=FOR.IX
R = R7,1
7
PutDT:=1 L7 PutDT = 1
R 8 = R8,1 ICN.IN:=CP.DO
R 9 = R9,1 DQU.IN:=CP.DO
Operate_free:=0
R10 = R10,1 R1,2
GetDT:=1
Load_free:=0
R = R1,1 R1, 2 R11,1
11
GetDT:=0
CP.DI:=CP.DO
Load_free:=0
R 12
= R1,1 R1, 2 R12,1 GetDT:=0
CP.DI:=ICN.OUT
Fig. 4. Pipeline system for processing of single input operators
7 Conclusion
With the requirements on high performance, a great focus was given to a specific
class of parallel computers in the 60's, denoted as data flow architectures. In dataf-
low architectures the computing process is managed by the operands flow ac-
cessed on different levels for executing instructions of data flow program. The
processing of the instructions is subject to the rules of firing an instruction (an
instruction is executable if all of its operands are available) and activation
(an instruction is activated when it is fired and the resources required for activa-
tion are available). Executing the program instructions can be done sequentially,
in a flow (pipelining), parallel, or in different hybrid modes, depending on the
used data flow computational model. Data flow architectures can be used advanta-
geously for computation-oriented applications that exhibit a fine grain parallelism.
Examples of such applications are image processing, scene analysis, aerodynamic
simulation, weather prediction etc. The proposed architecture described in this
article is based on data flow computational model, in which the computation is
controlled on the basis of data flow processing. Essential element of DF-KPI ar-
chitecture is coordinating processor (CP), which is responsible for managing and
organizing the execution of instructions. Structural organization of CP is designed
as a dynamic multifunctional system. By executing operations CP can pass
through different states, which results that this unit is a dynamic flow system.
Transitions and the order of transitions between different states are defined by the
type of executed operator by interpreting flow of operands. From the many types
of operators, this article provides microprogram managing for single input opera-
tors. The currently developed DF-KPI system with its principle of data processing
and its parameters is intended for solving tasks requiring brute force. The devel-
opment of the DF-KPI system is focused on the fields of virtual reality [16] and
computer security [20].
Acknowledgments. This work was supported by the Slovak Research and Development
Agency under the contract No. APVV-0008-10 (50 %), also the authors are pleased to
acknowledge to the project implementation: Development of the Center of Information and
Communication Technologies for Knowledge Systems (ITMS project code: 26220120030)
supported by the Research & Development Operational Program funded by the ERDF
(50 %).
References
[1] Budiu, M., Venkataramani, T., Chelcea, T., et al.: Spatial Computation. In: Interna-
tional Conference on Architectural Support for Programming Languages and Operat-
ing Systems, pp. 14–26 (2004)
[2] Dennis, J.B.: Data-Flow Supercomputers. Computer, 48–56 (1980)
[3] Franklin, A.D., Luisier, M., Han, S.J., et al.: Sub-10 nm Carbon Nanotube Transistor.
Nano Letters 12(2), 758–762 (2012)
[4] Gurd, J.R., Kirkham, C.C., Watson, I.: The Manchester Prototype Data-Flow Com-
puter. Commun. ACM 28, 34–52 (1985)
[5] Györök, G., Makó, M., Lakner, J.: Combinatorics at Electronic Circuit Realization in
FPAA. Acta Polytechnica Hungarica 6(1), 151–160 (2009)
[6] Györök, G.: The FPAA realization of analog predictive circuit. In: 8th IEEE Interna-
tional Symposium on Applied Machine Intelligence and Informatics: SAMI 2010,
pp. 105–108 (2010)
[7] Hennessy, J.L., Patterson, D.A.: Computer Architecture – A Quantitative Approach,
4th edn. Morgan Kaufmann, San Francisco (2006)
[8] Jelšina, M.: Design of Data Flow KPI Computer System. elfa s.r.o, Košice (2002) (in
Slovak)
[9] Kopják, J., Kovács, J.: Event-driven control program models running on embedded
systems. In: 10th Jubilee International Symposium on Applied Machine Intelligence
and Informatics (SAMI 2012), pp. 377–382 (2012)
[10] Kopják, J., Kovács, J.: Event-driven control program models running on embedded
systems. In: 6th IEEE International Symposium on Applied Computational Intelli-
gence and Informatics, Timişoara, Romania, pp. 323–326 (2011)
[11] Mai, K., Paaske, T., Nuwan, J., et al.: Smart Memories: A Modular Reconfigurable
Architecture. In: ISCA 2000, Vancouver, British Columbia, Canada. ACM (2000),
doi: 1-58113-287-5/00/06-161
[12] Mattson, T.G., Wijngaart, R., Frumkin, M.: Programming the Intel 80-core network-
on-a-chip Terascale Processor. In: SC 2008, Austin, Texas, USA (November 2008),
doi: 978-1-4244-2835-9/08
[13] Mercaldi, M., Swanson, S., Petersen, A., et al.: Modeling Instruction Placement on a
Spatial Architecture. In: SPAA 2006, July 30- August 2. ACM, Cambridge (2006),
doi: 1-59593-452-9/06/0007
[14] Sakaralingham, R., Nagarajan, R., McDonald, R., et al.: Distributed Microarchitec-
tural Protools in the STRIP Prototype Processor. In: The 39th Annual IEEE/ACM In-
ternational Symposium on Microarchitecture, MICRO 2006 (2006), doi: 0-7695-
2732-9/06
[15] Selow, R., Lopes, H.S., Lima, C.: A comparison of FPGA and FPAA technologies for
a signal processing application. In: Proceedings of the International Conference on
Field Programmable Logic and Applications, pp. 230–235 (2009)
[16] Szabó, C., Korečko, Š., Sobota, B.: Data Processing for Virtual Reality. In: Gulrez,
T., Hassanien, A.E. (eds.) Advances in Robotics and Virtual Reality. ISRL, vol. 26,
[17] Verdoscia, B., Vacarro, R.: ALFA: A Static Data Flow Architecture. In: Proceedings
of Fourth Symposium on the Frontiers of Massively Parallel Computation, McLean,
VA, USA, pp. 318–325 (1992)
[18] Vokorokos, L.: Data Flow Computer Principles. Copycenter, spol. s.r.o., Košice
(2002) (in Slovak)
[19] Vokorokos, L., Ádám, N., Madoš, B., Danková, E.: Operand processing and P-single
operators. In: INES 2011: 15th IEEE International Conference on Intelligent Engi-
neering Systems, pp. 209–214 (2011)
[20] Vokorokos, L., Ádám, N., Baláž, A., et al.: High-Performance Intrusion Detection
System for Security Threats Identification in Computer Networks. Computer Science
and Technology Research Survey 4, 54–61 (2009)
[21] Waingold, E., Taylor, M., Srikrishna, D., et al.: Baring it all to software: RAW ma-
chines, Technical Report, Massachusetts Institute of Technology Cambridge, MA,
USA, IEEE Computer, vol. 30(9), pp. 86–93 (1997)
[22] Wintzlaw, D., Griffin, P., Hoffmann, H., et al.: On-Chip Interconnection Architecture
of the Tile Processor. IEEE Micro. 27(5), 15–31 (2007), doi:10.1109/MM.2007.89
Potential Application of P-Graph-Based
Workflow in Logistics
József Tick
Óbuda University, H-1034 Budapest, Becsi út 96/b, Hungary

tick@uni-obuda.hu
Abstract. The free movement of goods and the liberalization of the markets in
Europe resulted in heavy transport and overloading of motorways and public
roads, definitely putting an obstacle to economic growth in the European Union.
The further construction and reconstruction of such roads cannot keep pace with
the increasing demand, furthermore, puts a heavy burden on the states’ budget.
Since traffic cannot be increased further more some other solutions must be found.
Such an alternative solution could be to develop optional models for logistics.
This paper examines the possible usage of the P-graph-based workflow for model-
ing logistical problems. The application of P-graph-based workflow is justified by
the fact that it has already been applied successfully in other fields of industry.
Further, the paper will examine the opportunities of P-graph modeling in a specif-
ic field of logistics Supply Chain Management, and will also introduce a new term
the P-graph based Supply Chain Model.
1 Introduction
Economic development in developed countries of Europe generated a dramatic
increase in mobility. Delivery of raw materials, intermediate materials and fi-
nished goods restricts the capacities of road-, water-, rail- and air-transportation.
The “just in time” shipping demands require calculable and reliable traffic-
conditions on all delivery routes. Air-traffic also suffers from increased demand,
hence air-corridors are ever more condensed (vertical and horizontal isolation of
planes are smaller), flights are more frequent and an even more coordinated air-
traffic control is installed. The most severe situation however occurs on the public
roads. Day by day, freight-trucks, cars, trams and other vehicles of community
transportation and transit circulate on the overwhelmed road-networks.
It is obvious that the overloaded and overcrowded paths of transport become a
major inhibiting factor of economic growth. In Europe the enlargement of the
European Union augmented the problem, since an integrated free market also
meant the free movement - and thus shipping - of goods (particularly in the west-
to-east direction). The relocation of production to sites with low-cost labor-force
enlarged the volume of transport of raw materials and finished products.
The pace of development of highways, roads and railways cannot meet the ever
growing demand. The financial resources of the EU and member states are heavily

294 J. Tick
burdened by such expenses. Also, with the progress in the field of logistics, the
older terminology of shipping was replaced by the more complex process of trans-
portation. In order to better serve this more multifaceted concept information
technology with its boundless opportunities is at hand. IT, on the one hand, contri-
butes to the improvement of the efficacy of traffic-regulation and the supply of
drivers with up-to-date and reliable information, i.e. the introduction of ITS (Intel-
ligent Traffic System). On the other, it facilitates the perfection of transportation,
the constant upgrading of logistical-information systems. The optimization of
logistic workflow-models is a key element in using IT in this field. This paper
offers a new approach in this area.
2 The Opportunities for General Workflow Models

Workflow Management Coalition (WfMC) offers the best and most compelling
definition for the term ‘workflow’. According to this: „The automation of a busi-
ness process, in whole or part, during which documents, information or tasks are
passed from one participant to another for action, according to a set of procedural
rules” [1].
The most authentic description of ‘logistic management’ was introduced by the
Council of Supply and Chain Management Professionals (CSCMP): “Logistics
management is that part of supply chain management that plans, implements, and
controls the efficient, effective forward and reverses flow and storage of goods,
services and related information between the point of origin and the point of con-
sumption in order to meet customers' requirements” [2].
The dynamic development of logistic planning and logistic management was
accelerated by the intensive use and support of IT tools. Beyond planning, model-
ing, controlling and follow-up, these systems and applications also work in tradi-
tional business and data management areas, like accounting, cash-flow manage-
ment. In the field of logistic planning several solutions have already been devel-
oped for analyzing business processes and modeling the flow of raw materials and
products. In the following sections this paper examines the adaptability of a new
tool, the P-graph based workflow model in the field of logistics.
2.1 The Description of the Traditional Workflow Model

Based on the definition above, workflow is the combination of the segmental and
parallel branches of a network of business processes broken into professional
spheres. Aalst, Hofstede, Hee és Jablonski [3], [4] [5] provided a good functional
description of such a network - its formula and the procedural rules behind it.
Figure 1 shows a general pattern of a workflow model, using the notation of
WfMC. The figure clearly demonstrates the activities (A1-A14) necessary to
complete a business process, and how the activities are linked. The terms ‘split’
and ‘join’ picture the parallel and alternative activities; their synchronization is
also explicative. This model also illustrates that this presentation of the workflow
chart is process and activity oriented and it unambiguously shows the structure
and timing of the tasks of a business process.
Potential Application of P-Graph-Based Workflow in Logistics 295
Fig. 1. Workflow model with WfMC notation
Tick [6], [7], [8] lists and describes the universal workflow-models, and elabo-
rates on the workflow-modeling with Petri-network, colored Petri-network and
UML-diagrams. The models tallied by Tick are functional in several instances and
have many advantages. Various editing and modeling tools support their applica-
tion. The modeling with Petri-network is particularly outstanding, as - beyond the
static structure of it - through the tokens it can demonstrate the dynamic features
of workflow. Thus, through the modeling, the flaws of workflow structure, the
tight cross-section and the incidental redundancies can easily be identified.
2.2 The Introduction of P-Graph Workflow-Model

P-graphs in workflow-modeling were first introduced in the modeling of adminis-
trative tasks [7], [8]. P-graphs were initially constructed for the modeling of
industrial, particularly chemical industrial processes. Here it illustrates the trans-
formation processes and the flow of raw materials, as the works of Friedler at al
[9], [10], [11], [12] show.
The original P-graph approach differs from the workflow approach, P-graph ra-
ther emphasizes the flow of materials. A tremendous advantage of P-graphs is that
based on the graph-principles an ideal structure of network can be devised by
systematic mathematical means. This is a major gain in the planning process, but
the monitoring during the simulation process is also facilitated. Adjusting different
parameters, adding or removing resources to the system, a new automatically gen-
erated ideal structure is established, that can further be examined.
The fact that the generation of a new structure happens almost real-time, also
simplifies the tracking of the process. This means that the change of load (reduc-
tion of load, overloading, reduction of the power of one element) in an ongoing
process can model the breakdown of the system, and thus initiate quick response
in planning. In a suddenly changed situation an ideal structure is still available.
This ensures the optimal and lasting performance of a running system. Figure 2
shows the original P-graph model.
296 J. Tick
Fig. 2. The original P-graph model based on [9], [10], [11]
The figure clearly shows the production of product A, the necessary raw mate-
rials (E, F), the semi-manufactured products (C, D, G), the by-product (B) and the
steps of the production process (1-4). The figure also displays the qualities of ma-
terial flow, which beneficial in the modeling of administrative and logistic tasks.
These advantages of P-graphs are persuasive to create the new model of
workflow, the P-graph based workflow modeling. In the construction of the model
the modeling of administrative tasks has priority, thus a new terminology and a
new approach was required at the transformation of the P-graph. The status of
materials and products easily correlate with the administrative tasks, but material
per se has a completely different sense in the administrative process than for ex-
ample in a chemical industrial process. These features are summarized in paper [7]
and [8]. Figure 3 shows a P-graph based workflow model of administrative
processes.
Fig. 3. P-graph-based workflow model of a simple administrative process based on [6],

[7], [8]
Figure 3 demonstrates a P-graph based workflow model with the document re-
sources (d1-d4), the intermediate documents (d5-d10), the final document (d11) and
the activities in the process (a1-a5).
A new feature of the model is the introduction of source (s1-s5), which is linked
to each activity.
298 J. Tick
In the simplest model an activity can be realized if all documents and the ne-
cessary sources are all available at the input. Paper [8] analyzes the special quali-
ties of documents and sources in detail. In cases of documents it has to be taken
into account that at each activity input document-linking or – coupling is neces-
sary. Documents, on the contrary to materials in the original P-graphs – are
distinctive and not fungible. Processing (activity) can only start in case of the
coexistence of documents with identical reference numbers. Another special fea-
ture is the availability of resources, which in some cases is not continuous, but
fragmental. For instance, the referent every now and then needs to take a break
from the work, a lunch-break is also mandatory. This relates to the fact that the
quantity resources for the activities is quantified (process time necessity) and with
respect to different documents it may differ. Activity therefore can only start if
resources are available for at least the entire period of the activity. Resources
linked to activities are shown on the figure with r1-r5.
Based on the assumption that the processing of various documents may not be
determined, [13] introduces a new extension to solve the issue. The fuzzy-
extension of P-graph based workflow model “fuzzies” the resource demand of
activities, thus renders the operation of the model more realistic. The introduction
of fuzzy-sets is a relatively new solution, its efficiency is currently under investi-
gation.
3 Opportunities of Using P-Graph Based Workflow Models

in Logistics
This paper investigates the usage of P-graph based workflow models in logistics
from two aspects. First, it will review the relevance of this model to business
processes; secondly the paper will check the model’s importance in a particular
field of logistics.
3.1 Study of Modeling Business Processes in Logistics

The implementation of logistic tasks is easily modeled by business processes, and
it can be mapped by workflow charts. In this case workflow provides a general
model, where activities are of any kind. The advantage of this is that general tools
of workflow models are freely applicable; the drawback however is that visual
symbols are basic, hence it gives only minor aid for professionals. A considerable
gain still is its flexibility with the introduction of new elements to the model.
There are already more focused solutions in devising business process models
in logistics. In contrary to general models, in these solutions the models offer a
range of elements, thus improving first sight impression.
The outstanding example in business process modeling is the Event-driven
Process Chain modeling, created within ARIS Framework [14], in use since the
1990s. In the EPC-model, events drive business functions. By the introduction of
logistic operators it can be forged to a full network. This network constitutes a
guided graph [15], with events and functions at the nodes and links are the edges.
Using the WfMC symbols, splitting and parallel implementation shown on Figure
1 is realized by the application of logistic operators (AND, OR, XOR). The basic
EPC-model can be elaborated and refined by the application of Data flow, Organi-
zations Unit and IT-system components.
Another remarkable solution is Business Process Modeling Notation (BPMN),
developed by the Business Process Management Initiative, which is a ubiquitous
graphic application. BPMN achieves three goals: to present the business process in
graphical form, to describe business activities and specify process properties in
detail, to support communication by a simple set of notations. BPMN is a common
platform for several Tools, which support the user in process analysis, planning,
improvement, simulation, and execution. BPMN uses five different types of ele-
ments to model business processes:
• swimlanes - graphical containers (pools, lanes) to store flow objects. The idea is
well known and used in object-oriented theories, methodologies.
• flow objects - primary elements (events, activities, gateways) to define the beha-
vior of process.
• connecting objects - connectors (sequence flows, message flows, associations,
data associations) that connect the flow objects to form a flow.
• Data – information (data objects, data inputs, data outputs, data stores) needed or
produced when executing a business process.
• Artifacts – artifacts (groups, text annotations) provide additional information
about a business process.
The elements of BPMN clearly show that notion is closely related to IT and busi-
ness modeling. This set of symbols has a significant benefit, since the communica-
tion gap closes between planning professionals, developers, analytical staff and
end-users, thus a more precise, demand-driven system can de developed. Business
Process Management Initiative and the Object Management Group (OMG) an-
nounced the merger of their Business Process Management (BPM) activities to
provide thought leadership and industry standards for this vital and growing indus-
try. The combined group has named itself the Business Modeling & Integration
(BMI) Domain Task Force (DTF). It means that the BPMN has a strong object-
oriented technical and technological background. Several Tools were developed to
support expertise in their business process modeling activities. These tools offer a
great variety of elements, making modeling easy, simply and quick. All these boil
down to the argument that BPMN is the leading standardized solution of business
modeling in industry.
The elements of P-graph based workflow models is less rich (though it can be
extended) and less widespread, than those of BPMN. Therefore, a general usage of
P-graph based workflow models in logistic business modeling is not anticipated.
However, there are a few particular areas, research and optimization, where the
theoretical background of graphs can benefit the modeling. This originates in the
automatic optimal generation of the structure. In case the activities are not mani-
fold and a linked cost-function is available, generation of the ideal structure is
handily realized. If the cost-parameters change, a new ideal structure is quickly
generated.
300 J. Tick
3.2 Modeling a Particular Field of Logistic Processes

Transportation is a key element in logistic business modeling. Particularly impor-
tant is the so called Supply Chain Management (SCM) in ensuring the uninter-
rupted production process.
Keith Oliver defined the concept first in 1982 as follows: “Supply chain man-
agement (SCM) is the process of planning, implementing, and controlling the
operations of the supply chain with the purpose to satisfy customer requirements
as efficiently as possible. Supply chain management spans all movement and sto-
rage of raw materials, work-in-process inventory, and finished goods from point-
of-origin to point-of-consumption”. The supply chain is often represented as a
network, or can be represented as a graph.[26] The nodes represent facilities and
the edges represent the links between them. Links are direct transportations by the
company, which managing the supply chain. The network or the graph is a useful
device to depict the structure of the model.
SCM includes the flow of materials (raw materials, semi-manufactured prod-
ucts, finished goods) and the collateral flow of information. The flow of informa-
tion is a prerequisite for uninterrupted operation, and is an integral part of the
information system of the company. During the planning of the supply chain the
flow of information can be separated from the planning of the real supply network.
As mentioned above, supply can be depicted by a network- or graph-like structure.
This graph describes the ways of raw materials to the production area, the chan-
nels of by-products from suppliers to the manufacturer, the intermediate materials
to other manufacturers, and the finished goods to the retail chain. This model is
very closely related to the P-graph based approach. That makes P-graph based
model arguably useful in the planning of the Supply Chain. The correlation is
evident. With respect to the special requirement of a given field, the elements are
shown on Figure 4.
Fig. 4. The elements of the P-graph-based Supply Chain Model
This figure shows that compared to P-graph based workflow this P-graph based
Supply Chain has materials, activities and supply route, but resources are not de-
picted. The reason for this is that resources and timing is not part of the P-graph
based Supply Chain Model (PSCM), rather it belongs to the Supply Chain Infor-
mation Model (SCIM). The two together make the Supply Chain Model.
Based on the mathematical model of SCIM [6], there is a finite set of material
M (which contains the sets of P products, I intermediate materials and R raw-
materials) and the finite set of S supply activities. Consequently, the set of P prod-
ucts, the set of I intermediate materials and the set of R raw-materials must be
subsets of M and the set of M materials and the set of S supply activities are dis-
junctive. Each supply activity “produces” (transports, stores, etc.) output materials
from input materials. Therefore two disjunctive sets can be assigned to each oper-
ating unit, i.e. the set of input and the set of output materials. In the network (in
the graph) the output materials of each supply activity are the inputs of different
supply activities. The primary aim of the p-graph based model, is that we can use
the theory of the process-network synthesis PNS [6]. As the first step of PNS, all
the plausible supply activities S and intermediate-materials must be determined.
By determining P,R and S, the set of all the material in the network (graph), M is
also defined. The optimal solution structure generated by process-network synthe-
sis must have several basic features that are taken for granted as axioms, and the
introduction of which improves the efficiency of the combinatorial search during
the process [6]. The Maximal Structure Generation (MSG) polynomial algorithm
elaborated by Fiedler et al., [10] which uses the axioms, generates that maximal
structure, all the combinatorially possible structures of which are its subsets. The
maximal structure generated by the MSG algorithm contains all such combinato-
rially possible network structures that are able to produce the end-product from the
given raw-materials. Consequently, it contains the optimal network as well. In
most cases the optimization means to find the most cost effective solution. The
application of the SSG (Solution Structure Generation) algorithm enables the pro-
duction of all the solution structures. The SSG is a new mathematical tool based
on the application of the Decision-mapping (DM) which has been developed by
Friedler et al. [10].
The advantage of PSCM is that the ideal structure of the model, i.e. the number
of supply chains and their connections, is determined by network-synthesis, a
mathematically legitimate process. As the computer based network-synthesis
based on P-graphs is quick and responsive, changes in conditions automatically re-
devise the structure of the supply chains. This ensures that materials necessary to
production are always ideally available.
4 Conclusion
Based on the sections above, it can be concluded that while P-graph based
workflow models are suitable for administrative tasks, in logistic business models
it will not appear generally. However, the advantages of the model based on the P-
graph network synthesis will prevail in certain fields where the constituting ele-
ments are scant, but the network itself is intricate.
The field discussed in Section 3 differs. It can be argued that in the field of
Supply Chain Management P-graph based models, based on its network structure,
have an advantage. The theoretical mathematical background of P-graphs, the
network-synthesis approaches already in use in other fields, the automatic and
mathematically proper generations of the ideal structure are valuable in supply
302 J. Tick
chain planning. The PSCM, introduced in this paper, still require additional stu-
dies in these fields. The method still needs verification through case-studies. In
case of positive outcome an optional direction for development is the fuzzy-
extension, already working in P-graph based workflow models. This could lead to
more authentic and reliable description, even in the case of Supply Chain Models.
References
[1] The Workflow Management Coalition Specification, Workflow Management Coali-
tion Terminology & Glossary; Document Number WFMC-TC-1011; Document Sta-
tus - Issue 3.0 (February 1999)
[2] definitions, http://cscmp.org/aboutcscmp/definitions.asp (Febru-
ary 12, 2012)
[3] van der Aalst, W.M.P., van Hee, K.M.: Workflow Management – Models, Methods,
and Systems. The MIT Press, Cambridge (2002)
[4] Jablonski, S., Bussler, S.: Workflow Management: Modeling Concepts, Architecture
and Implementation. International Thomson Computer Press (1996)
[5] van der Aalst, W.M.P., ter Hofstede, A.H.M.: Verification of Workflow Task Struc-
tures: A Petri-net-based approach
[6] Tick, J.: P-Graph-based Workflow Modelling. Acta Polytechnica Hungarica 4, 75–88
(2007)
[7] Tick, J.: Application of P-graph-based Workflow for Administrative Process Model-
ing. In: Proceedings of the 9th IEEE International Symposium on Applied Machine
Intelligence and Informatics, Slovakia, pp. 15–18 (2011)
[8] Tick, J.: Visualisation and Simulation of P-graph based Workflow Systems. In: Pro-
ceedings of the 6th IEEE International Symposium on Applied Computational Intelli-
gence and Informatics, Romania, pp. 231–234 (2011)
[9] Friedler, F., Tarjan, K., Huang, Y.W., Fan, L.T.: Graph-Theoretic Approach to
Process Synthesis: Axioms and Theorems. Chem. Engng. Sci. 47, 1973–1988 (1992)
[10] Friedler, F., Fan, L.T., Imreh, B.: Process Network Synthesis. Problem Definition
Networks 28, 119–124 (1998)
[11] Friedler, F., Varga, J.B., Fan, L.T.: Decision Mapping: A Tool for Consistent and
Complete Decisions in Process Synthesis. Chemical Eng. Sci. 50, 1755–1768 (1995)
[12] Friedler, F., Tarjan, K., Huang, Y.W., Fan, L.T.: Combinatorial Algorithms for
Process Synthesis. Computers Chem. Engng 16, 313–320 (1992)
[13] Tick, J.: P-gráf alapú workflow modellezés fuzzy kiterjesztéssel. PhD disszertáció,
Pannon Egyetem (2007)
[14] Nüttgens, M., Feld, T., Zimmermann, V.: Business Process Modeling with EPC and
UML. Transformation or Integration? In: Schader, M., Korthaus, A. (eds.) The Uni-
fied Modeling Language – Technical Aspects and Applications, Proceedings, Heidel-
berg (1998)
[15] Kovács, Z.: Logisztika és Üzleti Modellezés, Typotex, Budapest (2011) ISBN: 978-
963-279-510-2
[16] http://CSCMP.org (January 27, 2012)
[17] Pataricza, A. (ed.): Formalis modszerek az informatikaban, Typotex. Budapest (2006)
[18] Jeffry, J.M.: Using Petri Nets to Introduce Operating System Concepts. SIGCSE Bul-
letin. In: Proceedings of the SIGCSE Technical Symposium on Computer Science
Education, San Antonio, TX, USA, vol. 23(1), pp. 324–329 (1991)
[19] Kemper, P.: Logistic Process Models Go Petri Nets. In: Proceedings 7. Workshop
Algorithmen und Werkzeuge für Petrinetze, Koblenz, Germany, pp. 69–74 (2000)
[20] Jensen, K.: Coloured Petri Nets and the Invariant-Method. Theor. Comput. Sci. 14,
317–336 (1981)
[21] Merz, M., Moldt, D., Müller-Jones, K., Lamersdorf, W.: Workflow Modelling and
Execution with Coloured Petri Nets in COSM. In: Proceedings of 16th International
Conference on Application and Theory of Petri Nets (1995)
[22] Liu, D., Wang, J., et al.: Modeling workflow processes with colored Petri nets. Com-
put. in Industr. 49(3), 267–281 (2002)
[23] Qu, Y., Lin, C., Wang, J.: Linear Temporal Inference of Workflow Management Sys-
tems Based on Timed Petri Nets Models. In: Han, Y., Tai, S., Wikarski, D. (eds.)
EDCIS 2002. LNCS, vol. 2480, pp. 30–44. Springer, Heidelberg (2002)
[24] van der Aalst, V.M.P.: Interval Timed Coloured Petri Nets and their Analysis. In:
Ajmone Marsan, M. (ed.) ICATPN 1993. LNCS, vol. 691, pp. 453–472. Springer,
Heidelberg (1993)
[25] Shih-Yang, Y., Po-Zung, C., Chu-Hao, S.: Using Petri Nets to Enhance Web Usage
Mining. Acta Polytechnica Hungarica 4, 113–125 (2007)
[26] Shairo, J.F.: Modeling The Supply Chain, USA. Duxbury Applied Series (2009)
ISBN: 0-495-12609-8
Part V
Softcomputing in Biomedical
Engineering
Linear Surgical Instrument for Bipolar
Multielectrode Radiofrequency Ablation
Marek Penhaker1, Petr Vavra2, Milada Hlavackova3, Jan Kracmar3,

Tadeusz Sikora1, and Lukas Prokop1
1
Technical University of Ostrava, Faculty of Electrical Engineering and Computer Science,
17. listopadu 15, 708 33 Ostrava, Czech Republic
2
Faculty Hospital, Clinic of Surgery, 17. listopadu 1790, 708 52 Ostrava, Czech Republic
3
Technical University of Ostrava, Faculty of Mechanical Engineering, 17. listopadu 15,
708 33 Ostrava, Czech Republic
{marek.penhaker,milada.hlavackova,jan.kracmar,
tadeusz.sikora,lukas.prokop}@vsb.cz, petrvavra@yahoo.co.uk
Abstract. The aim of the work is about the design and realization of linear surgic-
al instrument for bipolar multielectrode radiofrequency ablation. The Radiofre-
quency intervention is more used in an abdomen surgical technique. The present
tools are universal for most of the operation with several disadvantages for large
ablation. Our designed and tested surgical tool consist of ten needles linear elec-
trodes bipolar driven by existing standard used generators. Proposed surgical tool,
in comparison with existing tools, used in extended ablation saves significant part
of healthy tissue and also operational time.
Keywords: radiofrequency, ablation, surgical, tool.
1 Introduction
Liver tumors present ambitious problem in clinical oncology, due to high morbidi-
ty and mortality. Liver resection is admitted as the good treatment for disposable
liver tumors and leads to a significantly ameliorated survival. However, resection
of liver malignancies is possible in only 10 – 25% of patients at the time they are
first seen [1,2], due the tumor extent or localization, inadequate hepatic reserve or
very high patient’s co-morbidity.
Patients with abstractedly disease may be qualifier for local ablative techniques,
chemo embolization, systemic or local chemotherapy [2-4].
Radiofrequency ablation (RFA) has been increasingly used for unresectable he-
patic tumors [3]. Multiple trials have evaluated RFA for the treatment of unresect-
able primary and secondary liver tumors and proved that RFA can control hepatic
malignancies with few associated complications [2-6]. However, surgical treat-
ment is the “gold standard” for patients with liver metastases or primary liver
tumors. Intraoperative blood loss and the subsequent need for blood transfusion
are considered significant risk factors for increased mortality and morbidity, poor

308 M. Penhaker et al.
postoperative outcome and a shorter duration of disease-free interval [7-8]. Over

the years different techniques have been developed to allow safer liver resection.
The role of radiofrequency energy in the liver surgery has been expanded in the
last years from simple tumor ablation to its use in the technique of radiofrequency
(RF) assisted liver resection. This new technique employs the heat produced by a
radiofrequency needle electrode to coagulate the liver tissue before cutting it, thus
permitting liver resection with minimal blood loss [7,9].
Thermal energy induced by radiofrequency generator is applied through special
RF needle (Habib Sealer 4X) repeatedly along the intended resection line. This
leads to the creation of approximately 2cm wide zone of coagulative necrosis,
where liver parenchyma can be divided by scalpel with minimal blood loss [7,9].
Hemostasis is obtained only by radiofrequency energy, no additional devices (e.g.
stitches, knots, clips or glue) are needed. The pioneer in using RF energy for liver
resection is professor Nagy Habib, that’s why it’s also called Habib’s resection.
Radiofrequency assisted liver resection is a simple method which can be used
in open or minimally invasive surgery. RF – assisted technique of resection can be
used for all types of liver resections from wedge resections to hemihepatectomies.
It has been reported to be associated with minimal blood loss, low blood transfu-
sion requirement and no need of intraoperative hepatic inflow occlusion tech-
niques such as Pringle’s maneuver, dissection and clamping of hepatic pedicle
[7,9-11]. This innovative technique allows performing more minor hepatic resec-
tions and less major hepatectomies and thus spares patient’s normal hepatic paren-
chyma for future re-resections.
In terms of disease recurrence there is very important advantage of RF assisted
resection over standard liver resection techniques. Resection line is treated by
thermal ablation leaving 1cm wide zone of coagulated tissue in situ. The antic-
ipated consequence is lower disease recurrence in resection bed and in indicated
cases it allows to drift resection line closer to the tumor [1,10].
2 Radiofrequency Principles
In 1891 in Paris, D'Arsanval physicist described the effects of alternating current
at a frequency of 250 kHz to biological tissue, where it produces heat without
stimulation of motor nerves and muscles. When applying power slowly with a low
current density on the tissue is heated to a cell, tissue is dehydrated and the colla-
gen structure degenerates. Alternating electric current at a frequency of 500 kHz is
used in surgery to cut tissue or coagulate, depending on the wavelength, which is
created in the tissue of the electric arc. When using alternating current at a fre-
quency of 460 kHz with the introduction of a special needle into the tissue, it will
be warm and destruction. At present, we can perform three types of radiofrequen-
cy needles: RITA Starburst needle SDE offers a different number of electrodes of
different length 3 to 5 cm, mostly 16G needles are used, but are made thinner. The
advantage of these needles is to monitor the temperature in the heated tissue,
which allows for ablation of high-risk areas such as near the spinal canal and
nerve roots, intestinal loops, the urinary bladder. [11]
Linear Surgical Instrument for Bipolar Multielectrode Radiofrequency Ablation 309
Radionics produces two types of needles, 18G needle ablation can be per-
formed one centimeter, the most of them performed ablation of the nerves. Inter-
nally cooled needle with a diameter of 16-17 G can be made 3 cm oval-shaped
Albac, using RITA needle ablation is performed in the shape of a sphere. The last
type needles produced Berchtold, a diameter of 16-18 G to create the ablation
diameter of 3 cm, oval-shaped, the disadvantage is non-constant application of
radiofrequency energy and the lack of temperature monitoring. The transfer of
radio frequency energy to tissue vibration of electrons occurs in the tissue and thus
its heat. At present the generators producing power 50 W in the tissue temperature
from 90 to 120 ° C with duration of 30 to 60 min. depending on the location of the
bearing ablation. If the bearing is located in the vicinity of blood vessels, the ap-
plication period is extended because the blood flowing through the heat of tissue
and extends the application. It may happen that sometimes fail to achieve the de-
sired temperature 90 ° C at the point where the electrode is close to blood vessels.
Application time can be shortened to administration of saline through the needle
into the bearing prior to ablation. At present, it is recommended to apply the F 1/1
by slow infusion through a needle introduced implementing ablation. At the end of
ablation, or sometimes served during ablation of the needle 95% alcohol to an
exaggeration of necrosis. Never, however, do not apply at the beginning of abso-
lute alcohol ablation in metastatic disease, because no tissue solidity to its diffu-
sion into the tissue deposits, whereas after ablation is a better diffusion in the
bearing. This leads to an exaggeration of blood clotting at the bearings. [11-21]
For liver tumors, especially in hepatocellular carcinoma is now emerging works
that combine radiofrequency ablation, which is used for the destruction chemoem-
bolization residual tumor cells. In most cases, this combination is used for larger
bearings than 5 cm in diameter, where TACE does not ensure tumor cells. Should
always be performed RF ablation within 1 cm of healthy tissue around the tumor
deposits to reduce the likelihood of residual tumor and the relapse process. Based
on the success achieved with radiofrequency ablation of liver tumors and liver
metastases, were carried out research in the use of this technique in the treatment
of kidney tumors, lung, bone, adrenal, pelvic, thyroid, breast and metastases in the
lymph nodes. Kidney tumors are one of the first extrahepatic tumors, where RF
ablation was performed. Most of the ablation of renal tumors carried in peripheral
tumors with the appearance of stripe peripheral infiltration of fat, this is roughly
parallel striations due to the area where ablation is performed [11,14,15,17].
Radiofrequency ablation is an electro - surgical therapeutic method for local
destruction of tissue using the thermal effect of electric current. The method of
tissue destruction is based on thermal effects of electric current at a frequency of
500-1000 kHz. The electrical circuit is closed between the two needle tool (bipolar
principle), which is not electrically stimulated tissue outside the wound. High
current density required for tissue destruction is achieved only at the tip of the
catheter or parallel conductors deposited in the electrolyte (tissue) in which high-
energy flows. High frequency of passing current therapeutic irritates nerves and
neuromuscular endings. In point of contact with the tissue catheter leads to coagu-
lation necrosis with a depth only about half a centimeter around the ring elec-
trodes. In the development of necrosis gradually increases tissue electrical
impedance (resistance) and thus prevent further spread of the tumor tissue and at
the same depth is controlled by coagulation radiofrequency generator [21] [22].
Limitation is that the tumor can not be larger than 5 cm. Because the liver is a
very fragile body vascular is one of the advantages of RF ablation that also coagu-
lates the blood vessels and, therefore, builds and bleeding. Technically not an
endovascular power to the liver up a needle electrode through the skin punctures
under ultrasound, CT or MRI [22].
For the actual effect of ablation is very important to the size and shape the re-
sulting coagulation necrosis, should cover the entire tumor volume, including the
security line, which is 10 mm. In terms of size and shape of the resulting ablated
deposit depends on
• heat generated inside the bearing,
• bio-heating properties of the target tissue, which is characterized by loss
of heat,
• the size and shape of electrodes used.
The resulting coagulation necrosis is correlated with the average course of abla-
tion electrodes (duration of a selected algorithm). Maximum thermal energy gen-
erated during the passage Tissue radiofrequency current is concentrated around the
active electrode and decreases with the square of the distance. The length is ap-
proximately ablated bearing insulated work of the active electrode.
3 Radiofrequency Instrumentation
High temperature is very effective at destroying unwanted tissue damage. Once
the tissue temperature rises above 50 ° C is permanently destroyed. The process is
very fast, usually just less than 10 to 15-minute exposure to 3 cm ablation. Size of
the ablation site, of course, depends on the size of the probe, tissue temperature
and the time required for application of RF energy. Adverse damaged tissue abla-
tion can be removed without damaging the surrounding healthy tissue.
There are several ways to make power, and each has different advantages and
limitations of use. Percutaneous access laparoscopic approach and open access:
performed under general anesthesia, the doctor sees a direct procedure. Most
treatments are also used surgical ultrasound. Recovery time is longer than pre-
vious approaches. [23]
3.1 Radiofrequency Generator RITA 1500X

Nowadays radiofrequency generator RITA 1500x RF Electrosurgical (Fig1.) unit
is included in a category II comes from the U.S. and is the second generation of
advanced technology that operates according to the selected program from 1 to
250 W at 0.5 to 0.6 MHz. It is possible to achieve a tissue temperature to 125 ° C,
using a nine-channel control resistance.
Fig. 1. RITA generator 1500x [RITA Medical Systems-http://www.cattus-ritamedical.com/

index.php?area=5&language_id=8&page_id=17]
This generator consists of three generators operating state of the art security
system controlled by software technologies, which controls the impedance, tem-
perature in the tissue of the patient and electrodes on the patient's temperature.
The automatic program of the unit automatically monitors the impedance and
temperature achieved the need to eject needle device signals. In addition, it checks
the current temperature of each needle, therefore it is not necessary to cool the
needle. If the needle temperature reaches some higher temperature than the other,
a device then automatically notify - print "HI" and turns off the current to the elec-
trodes, thereby preventing overheating of the antenna and the tissue.
The shape and size of the instrument affects the quality of performance in terms
of speed and damage to healthy tissue. Today is a very widespread system of the
company RITA on figure 1.
Fig. 2. RITA software for visualization additional and basic information about the therapy
progress. There is impedance, temperature monitored continuously
RF Generator provides additional functionality, which allows recording and ac-

curate information (figure 2) as performance, impedance, temperature tips, process
time, information about cooling electrodes, indications for the ablation.
3.2 Radiofrequency Electrodes

Radiofrequency surgery tools allow creating different shapes ablation, and thus
the resulting model of coagulation necrosis. They vary in size active part, shape
and grouping of electrodes may be mono-polar, bipolar and multi-polar. Mono-
polar radiofrequency set consists of one or more active electrodes can be devel-
oped into an umbrella shape with a small area, which is located in the target tissue
and two electrodes placed on dispersion patient's legs enclosing an electrical
circuit.
Fig. 3. Principle of operation of electrodes vary depending on types of tools - general

scheme, [RITA Medical Systems - http://www.cattus-ritamedical.com/]
Bipolar radiofrequency set includes two separate insulated active electrodes

each, placed in the same shaft, the electrical current spreads in the direction of the
applicator. Multipolar RF system includes a set of three applicators, each with
separate electrodes within the lumen, electrical current can pass between the ap-
plicator and for different lengths of time.
The resulting system allows up to 15 combinations of shapes ablation. The ad-
vantage of this arrangement of electrodes is a higher volume ablated tissue and
formation of various shapes ablated bearings. Another advantage is two-way lu-
men applicator to enable continuous application of saline during ablation and thus
enhance the effect of treatment. The electrodes used in conjunction with radio-
frequency generator for use in open surgical wound and laparoscopy applications
figure 3. [24]
Fig. 4. Electrode Habib 4X, 2X, and 4X Laparoscopic [http://www1.imperial.

ac.uk/medicine/research/researchthemes/healthtechnologies/surgicaltechnologies/devicetec
hnology/]
Electrodes used for open surgery, the electrodes Habib 4X (short, long). The
electrode is made of two structural variants differing by the length of the elec-
trode. Short length 60 mm, long - length 100 mm. Electrode diameter is 1.4 mm,
electrode distance in the longitudinal direction of 6.9 mm, electrode distance in the
transverse direction of 5.9 mm see on Figure 4.
3.4 Power Output Measurement from Radiofrequency Generator

There were measured the output energy for fine design concerning the material for
plastic packaging and metal for the electrodes. The generator was set to 1500x
Rita 100W power probe with Habib 4x when measurements were measured on an
oscilloscope curve indicative of the principle of operation of the apparatus.
Fig. 5. The time course of the open circuit voltage - short pulses for measuring the contact
resistance (left), 'full power - burning tissue in 460 to 100 kHz (right)
Load device generates short pulses of about 50V pulses with a frequency of 460
kHz active parts, which are likely to have to measure the contact resistance and
thus ending the insurance protection by closing foot pedal against sunburn. They
Fig. 6. Frequency spectrum of the generator load (left), spectrum generator switched on -
ablation (right)
are also visible harmonics at multiples of the fundamental carrier frequency. In

closing - the level of tissue ablation, 100 V at a frequency of 460 kHz figure 5.
4 Methods
In these existing solutions is the first application of rapid tissue ablation with a
relatively wide range of applications. Disadvantages arise in cases of removal of
damaged tissue. In such case, the removal of unwanted healthy tissue from the
shape tool, or a significant extension of operating times. This assumption has been
verified in the following measurements existing HABIB4x electrode and the new-
ly proposed modified linear electrode octopolar.
4.1 Electrosurgical Tool Design

The biggest challenge was the deployment of needles so that their uniform dis-
tance set on to 8mm, and there was a more even flow of electrical current between
the electrodes (needles). Any deviation from this distance would exclude one or
more needles and there would be burning the same liver tissue and the tool would
not work properly. Furthermore, it was still required to have a simple shape tool -
like Habib 4x. When designing multiple needles, it should then be possible to
shorten the operating time. When the proposal was based on the placement of
needles such that their square ground plan formed in all directions, in which
needles are placed. Originally based on the n-wall whose edges have the same
length, but there was a problem. After adding the edges of the square was not the
same distance at the rear edges. Length of square edges labeled a, distance from
the center to the edge of the n-wall (the show was elected dodecagon) is indicated
by h.
The breakthrough came after drawing the shape of the tumor, in most cases it is
a circle or ellipse, which reaches up to size 30mm and trying to copy his forms.
These dimensions are suitable and required distance between the needles. Which
succeeded, when the imaginary shape of the tumor divided into four sections of
equal size, and around its circumference three squares were added to each section,
showing the layout placement of needles see Figure 7 below.
Fig. 7. Layout design of electrodes for tumor removal within 3cm diameter
This layout it became clear that the basic condition is met, the same distance
between the needles. But a new problem appeared, in this proposal there are two
critical points because of the possible failure of the protective rim, which was set
at 1 cm. This problem does not refute or confirm, however, to directly test the
prototype in practice, when measurements on liver tissue. The practical test was
then performed at University Hospital, where he worked with the operating tool,
MD. P. Vavra Ph.D.
Fig. 8. Final layout design placement of electrodes for tumor removal within 3cm diameter
That the entire tumor is removed, it is necessary to use the 4x. This is seen in
Figure 8, where it is drawn, however, only half of the tumor. It will therefore need
ten electrodes (needles), which will be distributed as shown in Figure 8.
4.2 Material Composition

From the chemical analysis which was carried out to determine the material exist-
ing needles showed that this material CrNi 18/81. The marking according to IEC
17 241.4 corresponded to steel. The structure of stainless steel containing 18% Cr
and 8% Ni austenitic is and remains at low temperatures austenitic. There is no
potential for improving the strength of the heat treatment, such as hardening. Nev-
ertheless, the impact values achieved good work, ie toughness at low tempera-
tures. The material reacts strongly to cold forming. Reinforcement is noticeable
when bending, shearing, but also for straightening after welding. The most impor-
tant property of this steel is corrosion resistance. Molecular adsorption force on
the material surface creates a thin layer of oxygen, which is not in the form of
oxides like aluminum, but in the form of gas. This steel is used mainly in food and
pharmaceutical industries. Measurements were performed manually Niton X-ray
analyzer XL3t900 GOLD. The surface layer is composed of 3 elements - Ni
(17.68%), Fe (8.081%), Cu (0.814%). The composition of elements in the core is
already identical to those of steels of this class.
Ti layer
Fig. 9. Microscopic detail of gold surface from existing Habib 4x electrosurgical tool
Analyses were performed on the Oxford EDS spectrometer INCA Energy 350
and are indicative only. Layer on the tip of surgical instruments is titanium with a
thickness of an average of around 2.6 micron.
Table 1. Representation of the materials for the golden spike-probe Habib 4X
(hm. %) Si Cr Mn Ni Cu Mo
Gold
0,6 ≈17,0 1,3 9,7 0,6 2,4
needle
Table 2. Representation of the materials to paint golden-tip probe Habib 4X
(hm. %) C O Na Al Si K
Paint on
55,5 27,5 2,5 4,2 8,8 1,5
needle
Table 3. Representation of the materials for the metal probe tip-6X Habib
(hm. %) Al Si Cr Mn Ni Mo
Paint on
0,1 0,7 ≈19,0 1,5 8,0 0,3
needle
Paint layer
Fig. 10. Detail of the painted surface of the electrode tip
Painted part has a thickness of about 55 micrometers. Analysis of small areas of

this layer is shown below. In addition to organic compounds some contains grains
of minerals.
4.3 Primary Model Proposal

In order to model tools made, had to be used in 3D graphics software. In this case,
was used Autodesk Inventor 2010. With this physical model is based directly from
the distribution of needles, so its shape near the most plan placement of needles.
Dimensions were chosen as small as possible, due to its improved operability.
Length was 39 mm and width 16 mm. This way downsizing was not suitable and
other models from her abandoned. Although small size is required in the operation
must take into account the surgeon's ability to manipulate this device in a small
space, and it was worse than the tools that are ergonomically designed where the
height was shaped. It involved of individual needles and reached 16 mm. For easi-
er manipulation an ergonomic shape was designed on top of the bracket, which of
course could not be produced by 3D printing. Therefore, since his production
dropped and the model accounted for only 2 parts that have the same dimensions.
The only difference between them was that the bottom of the model had holes for
needle guidance, which reached a height of 5 mm and had a constant position to
ensure mutual needles. In the upper part of the model were these holes, but was
added gain the upper walls, needed for resting the ends of needles. This amplifica-
tion reached a thickness of 2 mm and was designed power in order to reflect
events that draw the end of the needles when the force that is necessary to pene-
trate tips of needles into the liver tissue. The thickness of the top surface of the
reinforcement was not sufficient and the design model further increased. Like the
walls, which in the original proposal had a thickness of 1 mm, it was necessary to
intensify. On the back of the model was in the hole for the cable that powers the
instrument. At the end was the power cable connector, which is needed to connect
to the RF generator from the company RITA.
4.4 Testing Model

Although the model took a shape from the original design, there had to be made
corrections and adjustments. The new model was anyway made up of 2 parts - the
upper and lower part. A fundamental change came with an enlarged scale. It was
increased in length and width dimensions, the height is increased to 20mm. Wall
thickness was increased to 3 mm, which were reinforced with external structure
tools. The progress to enlarge the inner space that is visible around the outer
needles see. figure 11. This simplifies handling of needles in their wiring. At the
bottom of the model there was seen the increased wall thickness, which serves as a
guide needles. The new model achieves a thickness of 5 mm. At the top of the
wall thickness was increased (resting) and reached the final thickness of 3.5 mm.
The biggest change occurred over the original model on its outer shape, though its
length was increased from the original 39 mm to 45 mm. And its width was in-
creased to 22 mm.
Fig. 11. Final testing prototype of linear octopolar radiofrequency surgical tool
An adjustment forms for easier and more secure grip in the fingers surgeon.
Especially when picking thumb. On the back of the withdrawal was made that the
shape of the thumb to hold the tool better. Sharp edges, which were located at the
top and bottom of the model, were rounded to avoid injury. The upper surface
contains the hole, which is used for the power cord. At the end of the power cord
there is connector that plugs into the RF generator RITA.
4.5 Prototype
For creating the prototype there had to be prepared a 3D model of the instrument.
The prototype was produced by 3D printing, which is exact, but in its use there are
limitations shapes. An equipment needed for its production it was given the Facul-
ty of Metallurgy and Materials Engineering.
The prototype was made using the 3D printing. The exact method name is Rap-
id Prototyping - Rapid prototyping using FDM technology. Was employed Di-
mension SST 768 (Soluble Support Technology). Using the Catalyst software,
which is part of the device will load a 3D model in STL format supported, the
software automatically generates a structural layer and all the necessary support
needed for the production model and data transfer for the production of 3D model
of computer network directly to a 3D printer figure 12 . This is the method of
production when using heated plastic (heated to 260 °C) and by its applying in
layers to form the desired shape. Even the temperature in the inner workspace
apparatus must be accurately set. When printing its value is 76 ° C. The internal
temperature depends on the quality of the connection layer and at the same time I
have to avoid rapid cooling. At the top of the unit is the print head that moves,
adding plastic to the destination. Construction ABS plastic in white (standard),
blue, yellow, black, red, green and metallic gray and is applied in thin layers of
size 0.24 or 0.33 mm according to the type of device. Besides, we also used the
plastic support, which is brown in color and is printed at the beginning to correct
horizontal inequalities working boards. Creates a first thin base and building up
then you can use plastic construction. Support plastic is also used in the manufac-
ture of parts which have the shape of its cavity, hole, recess. Simply fill up and
then no longer possible to support the layer of plastic applied layers of plastic
construction, it is not possible to apply a layer of free space.
Fig. 12. 3D model of housing the electrode needles
Support plastic is then simply wash out of these cavities, because the prototype
to put lye bath and dissolve the plastic support. Working pad, on which printing is
done with dimensions 203x203x20 mm and is stored on a pedestal in the working
chamber. The maximum size of the modeling work space is 203x203x305 mm. At
the bottom of the device is a magazine with two cartridges - ABS plastic for con-
struction (volume 922 cm3) and the plastic support (volume 922 cm3). The 3D
printer automatically sets the desired temperature and plastic. To print, its length,
etc. is displayed on the LCD screen on the device. Dimension SST prototypes can
produce much more cheaply, without the need for operator expertise and specia-
lized equipment working environment. The only disadvantage of this method of
production is the need to adjust the prototype shapes, especially the edges in the
model. As already mentioned, can not be printed for free space, and therefore
often have to be canceled rounded edges, which was the original proposal. Specif-
ically, the test model had to be canceled rounding the top and bottom surfaces.
4.6 Assembly and Wiring the Electrosurgical Tool

The composition of the prototype had to be made both of which are managed us-
ing the 3D printing - see above. To complete the test model was chosen because
the original model had a significant disadvantage to completion - a small wall
thickness, internal spaces and a very small size. It was also necessary to have a
sufficient number of needles. For this prototype it is needed 8 needles with a di-
ameter of 1.6 mm and a length of 60 mm. In order to engage the needle into a
circuit, having at their ends to be connected terminals. Because the needle itself
can not solder, it would be their overall baking and then would only have been
connected with the supply wire. Due to the small interior space in the tool had to
be free of plastic blocks and left only the copper core. These nuclei were then
separated from each other, and each needle has been assigned a core that is anc-
hored with a small screw on the end of the needle. For better tighten the screw on
the ends of needles were made longitudinal cuts to screw shank needle firmly
pressed to the core.
The hole in the heart was filled with tin, in order to connect the needle with the
supply wire. Subsequently, the screw heads grounded to hamper the wiring. Then
the needle in their division pole colored wires connected in order to visually dis-
tinguish them and avoid possible error involved. Blue wire is used for connecting
the outer rows of needles, yellow for medium range engagement needles.
Fig. 13. Wire interconnection between electrode needles present interconnection (left) and
alternative involvement (right)
Involvement needles are done with the scheme figure 13. Each has the same
number of needles oriented electrical pole. Excitation takes place between the two
rows of needles. The positive and negative electric pole led to wire the wires
together and joined to the power wire to the end connector for connection to the
RF generator from the company RITA.
Then poised involvement needles inserted into the bottom of the prototype. It
had to be used more force, because the holes in the model are a 0.1 mm smaller
diameter are therefore 1.5 mm (due to overlap solid connection). The connection
is between the electrical poles and pulling cable. The upper part of the prototype
was drilled hole for the cable. Then the whole inner space filled with liquid plastic
weld gun, it will prevent axial displacement of needles and also serves as an insu-
lating filler element to prevent the wires connect with the opposite electrical pole.
After assembly, it was evident on tenterhooks minor deviation from their original
directions. Total unloading needles reached 50mm. To establish the correct
needles in their working distances used the lower part of the original model (as
lead). The lower part of the original model is used only for handling and transport
tools.
With the knowledge from surgeon and measurement we did with existing tool
HABIB 4X we designed linear width octopolar tool for rapid ablation in time and
mass of tissue. Distance between axial needles is 8mm and lateral needle distance
is 10mm. Length of the needles is 80 mm. The mater was use different from the
commercial one. We use chrom molybden coated steel staple. There were also
used RITA Generator 1500x and beef liver in 520g.
Fig. 14. Prototype of radiofrequency linear octopolar tool for testing
4.7 Infrared Imaging

Thermal images and video in the infrared spectrum have been used as a comple-
mentary method for monitoring the functionality of each ablation tools. The rate
was monitored by both warming tissue (liver) and by the temperature difference
and live in the coagulated liver tissue.
4.7.1 Emissivity
For accurate temperature measurement with thermal imagers is usually the biggest
problem in determining the correct emissivity. This parameter basically
determines the proportion of infrared radiation intensity (in a particular spectrum

of radiation, and if we want to be completely accurate, and in a particular direc-
tion) emitted by a material to the intensity of radiation of absolutely black body at
the same temperature.
In medicine (particularly human) imager applications are limited to measuring
the surface temperature of the human body, the issue of measuring the internal
organs thermal imager, we can find many documents. Emissivity of the skin is
determined from several sources, but the measured emissivity of liver tissue, we
find professional articles. It is also possible that the emissivity will change during
the ablation. For this reason, thermal measurements made only a complementary
method of verifying that new types of liver ablation tools.
When measuring, we assumed that the value of emissivity of the internal organs
will have nonmetallic character (it will move in values higher than 0.9) and should
be close to the value of the emissivity of the skin.
Infrared imaging temperature measurements with an infrared camera require
the assignment of a targeted surface emissivity value. Emissivity strongly influ-
ences the characterization of surfaces like the human skin with optical devices in
the thermal infrared range because emissivity is a measure of surface radiation and
absorption efficiency [25, 26].
Skin emissivity is an Important Factor in determining the true skin temperature.
Steketee [25, 27, 28] reported the emissivity of normal human skin to be about
0.98 ± 0.01 in the range of 3 m-19 m and concluded That pigmentation of the skin
does not Affect ITS value. The emissivity pericard 0.83 (fresh heart) and 0.90
(after 7 h and after 9 d) was evaluated within this paper. Togawa [30] reported the
hand skin emissivity and 0.972 ± 0.004 in the range of 8 m to 14 m while Villa-
señor-Mora et al. [29] eported the value of 0.976 ± 0.006 using an INDIRECT
measurement in the thermal infrared range through reflection of a CO laser beam.
An interesting work is [27], which deals with the methodology of identification
of skin tumors comparing two areas that are placed symmetrically on the body.
The temperature of the lesion shows a temperature higher by 0.08 to 0.2 ° C than
symmetrical, healthy area. For authors consider normal temperature difference to
about 0.8 ° C (depending on the specific areas of the body).
With such sensitive measurements accurate emissivity setting plays an impor-
tant role. In our case, when the temperature of liver tissue ranged from 15 ° C (pig
liver imported chilled) to about 90 ° C (ablated tissue), we need only relatively
coarse setting emissivity to 0.98.
4.7.2 Equipment Used
The ablation process was monitored using infrared camera Flir E45.
4.7.3 Measurement on HABIB 4x
Measurements were performed using thermal cameras, stop watches, and non-
contact thermometer with an emissivity of 0.96. There were the time and volume
of abladed tissue evaluation, which is sourced from the sample.
Table 4. Infrared camera parameters
Thermal sensitivity 0.10 °C at 30 °C

Image frequency 50/60 Hz non-interlaced
Detector type Focal Plane Array (FPA)
Resolution 160 x 120
Spectral range 7.5 to 13 μm
Temperature range -20 °C to 120 °C
Accuracy ± 2 °C, ± 2 %
Repeatibility ± 1 °C, ± 1 %
Beef liver was used at a temperature of 13.5 °C, 230g. Gradually electrodes
were penetrating to the liver and measured ablation and run time. This report
shows only the beginning and end of measurement.
49,1°C 49,1°C
40 40
SP01 SP01
30 30
20 20
19,5°C 19,5°C
Fig. 15. Measurement the Habib 4x 100W parallel with the surface of the liver - 3 s after
the start of the ablation (left), end of ablation of the ablation 15s (right)
Measurements were carried out in parallel with the injection of the liver and the
measured ablation time. The device controls the change of impedance between
the electrodes during ablation in tissue. This time was measured for various per-
formances, always an intact sample of the new liver.
38,3°C 38,3°C
30 30
SP01 SP01
20 20
10 10
8,7°C 8,7°C
Fig. 16. Measurement the Habib 4x 70W parallel with the surface of the liver – 3s after the
start of the ablation (left), end of ablation of the ablation 20s (right)
Measurement he Habib 4x 70W parallel with the surface of the liver –Last in a
series of measurements was measured perpendicular in the liver tissue stuck in
length and 30 mm length measurement ablation. All stitch tool Habib 4X was
burned tissue volumes in a rectangle at the base 10mm x 10mm.
Table 5. Measured values set ablation habib 4x
Power [W] Time [s] Direction of application

125 8 ┴
125 5 ═ parallel
125 8 ┴ deep
60 14 ┴
The difference between orthogonal and paralell electrode application were frac-
tional, because both values are given, but only values of vertical ablation. Tem-
peratures tools and liver ablation in the vicinity of the maximum of 107.6 ° C at
the Habib 4X in liver the power owas 150 W.
35,8°C 35,8°C
35 35
30 30
SP01 SP01
25 25
20 20
16,5°C 16,5°C
Fig. 17. Measurement he Habib 4x 100W upright with the surface of the liver – 3 s after
the start of the ablation (left), end of ablation of the ablation 22s (right)
Records of the thermal image shows the principle of burning and a picture of
the shape of the electric power field both instruments, which will be used in the
design of the prototype have also been detected value generating tools for applica-
tion Habib 4X and 100V at a frequency of 460 kHz. [34-36]
4.8 Modeling of Electrical Field

Concerning the right radiofrequency function of the surgical tool is primary paral-
lel lead of electrodes in tissue. There were done simulation of near and far elec-
tromagnetic field at the electrodes figure 18.
Fig. 18. Simulation of near (left) and far (right) electromagnetic field at the electrodes
The complete electromagnetic direction (figure 19) and current density of de-
signed radiofrequency tool in tissue environment is shown in figure 20. This simu-
lation allows to use the prototype of the surgery tool for relevant testing.
Fig. 19 Simulation electromagnetic field direction during the radiofrequency
Fig. 20. Simulation of current density of designed radiofrequency tool in tissue

environment
5 Measurement and Tests

There were realized several measurement and test on designed and realized radio-
frequency surgical tool. First test of penetration the needles present the correlation
between the numbers of needles and force necessary for puncture the tissue.
5.1 Experimental Findings Loading Force

Experimental findings loading forces carried out on Zwick Roell ZHU unit 2.5,
which was awarded the Center for Advanced Innovation Technology (hereinafter
CAIT). To measure was necessary to adjust the end of the needle, so it can be
attached to the clamping device of the measuring device. To simulate the human
liver is again used pork liver, which are structurally related to liver tissue. After-
wards, panned force required to puncture the liver. The device stores data, which
contain the necessary strength, depending on the displacement of these data was
then plotted a graph and then evaluated by loading force. Three measurements was
performed to avoid possible measurement errors. Force curve is shown in chart.
Maximum force, which was subtracted from the graph has a value NF3, non-
linearity is caused by inhomogeneous composition of the liver. Photo documenta-
tion of the measurements were added to Annex D. [31]
Habib 4X Electrodes
The aim was to obtain measurements of the force values to inject into the liver -
set of 4 needles. Measuring the initial sample was prepared Habib 4X device. The
device has been modified so that it can be attached to the chuck own measuring
machine Zwick Roell.
Pork liver was in a water bath preheated to a temperature of human body. The
liver was placed in a box that was to simulate a hand surgeon during surgery. [32],
[33]
A stroke length chuck, into which the needle has been clamped, was set to 50
mm, Chuck speed (speed of injecting) was set to 50 mm / min. Three measure-
ments were performed, and acquired three power curves to move the injection site
(nesting needle into the liver). Values are shown in figure below. When you start
measuring the needle does not touch the liver (approximately 10 mm).
The values obtained from the measurement can be stated:
• When all 3 measurements were obtained similar values and similar

curves without extreme deviations. Deflection was observed only in Me
• From the curves can determine the point of penetration needle into the
liver tissue (after compression)
• Comparing the values of the forces for one injection needle can be stated
that the injection needles with more force required to penetrate the needle
into the liver tissue increases linearly (force x number of needles needed
for one injection needle). This means that there is no mutual action of
forcible needles.
Single Needle Penetration

Measurement, whose aim was to determine the power needed from one needle
penetration into the liver. The needle was mounted in a socket, and this was
mounted in the chuck measuring equipment. Measurements were performed on
the tensile machine - Zwick Roell. [38, 45]
Pork liver was in a water bath preheated to a temperature of human body. The
liver was placed in a box that was to simulate a hand surgeon during surgery.
Stroke length chucks, into which the needle has been clamped, was set to about 40
mm, which corresponds to the maximum length of the needle into the liver, feed
rate was 50 mm / min.
Fig. 21. Measurement branch of penetration force to liver tissue - Zwick Roell
Three measurements were made. Results are dependent graphs, "Strength to

move." All 3 measurements obtained similar values and similar curves without
extreme deviations. Since the liver is homogeneous mass, the measured power
curve-shifting is not "smooth".
Fig. 22. Experimental test of one needle penetration to liver tissue
When you start measuring the needle touches the surface of the liver. The graph
shows that before the neck-down the needle into the liver occurs in the compres-
sion range of 13 - 20.5 mm. Once the needle penetrates into the liver (puncture
strength on its own is in the range 0.5 to 0.7 N) - the step change. The next shift
the force required to extrude needle liver tissue and increases nonlinearly. It can
be assumed that the surrounding tissue not only acts against its own motion the
needle (needle in the axial direction), but also perpendicular to the axis and formu-
late frictional forces that cause the growth of the forces needed to move the needle
further tissues. [43,44]
The obtained results are consistent with experience in working with the operat-
ing tool Habib. The result shows that the correlation between the number of
needles and force is linear. That fact helps to adjust the difficulty in using the sur-
gical tool in practice. [39]
5.2 Thermal Measurement on Linear Octopolar Tool

There were realized the tests on pork liver, which are most similar to human liver,
although the condition of human liver depends on the lifestyleof each person it-
self. . Together with colleagues from the FEI we had to ensure the thermal imager,
which can be monitored with increasing temperature in both the instrument and in
pig liver. For testing, it was most desirable to monitor the temperature of needles
in a place where there is contact needle and plastic prototype parts of the body.
This is a small thermal resistance of the plastic prototype. However, when measur-
ing the temperature at the bottom part of the prototype come close limits for the
plastic and it was possible to continue the tests.
For proper function of the tool it had to be plugged into the RF connector wave
generator made by RITA. From the surgeon's fingers spread was evident that it
will be necessary to adjust the external shape of the prototype to fit the given
layout. The first puncture in liver tissue was carried out without turning on the RF
generator because of the estimated power that is needed to develop a prototype
that went through the needle into the liver tissue. Magnitude of the force was sub-
sequently detected experimentally using instruments with higher accuracy. The
first engagement occurred in the RF generator when the needles were stuck in the
liver tissue. Burning process lasted less than 35 seconds, but the area between the
needles was not completely burned. The largest burn tissue was intermediate be-
tween the needles. Imperfection was caused by needles, which passes through the
liver tissue came out originally intended location.
Before the second attempt was necessary to return the needle to its original po-
sition by mechanical means, some needles managed to cope. When stuck needles
into the liver tissue and their passage through the needle is again thrown some of
the desired position. In order to define the relative positions are therefore used the
lower part of the original model and was used as a guide for the needle. When
running RF generator already burned area increased and reached well beyond
the outer needle. Time to burn was smaller and its value was 20 seconds. From the
foregoing it is an important tool in this uniform and constant distance between the
needles. After ejecting a needle and then examining the place burn surgeon, has
been recognized as a successful burn and it would be sufficient even if the actual
surgery. However, there still remains a problem with the establishment of the rela-
tive positions of needles. The surgeon will be able to place in operation or deploy
to another part of the needle tips of instruments that would compose the
establishment of needles. For this testing new ideas came to modify the shape of
the prototype, or the potential to add other parts of the prototype. [40-42]
Measurements confirm the theoretical assumptions about the nature and effec-
tiveness of ablation in the modified instrument Habib 8X with eight electrodes
equals the wheel spokes.
Table 6 Measured values set ablation on linear octopolar Tool
Power [W] Time [s] Direction of application

100 73 ┴
60 81 ┴
This tool also allows for ablation and excision of tissue volume with a greater
range than current electrode. It is at the base of rectangle 10mm x 40mm maxi-
mum height of 80 mm.
The measurements used the same correlation parameters and entered into
Table 2 Temperature measurement instruments do not exceed the temperature of
92.6 °C. The critical moment was the introduction of measurement instruments to
penetrate the tissue, because the electrodes with different consistency of liver
forked thereby changing the distance between the needles. This will affect the
distance for the propagation of electromagnetic fields and radiofrequency then
proceeded quickly to the electrodes closer together than those more distant.
Fig. 23. Measurement the linear octopolar tool 100W upright with the surface of the liver -
3 s after the start of the ablation (left), end of the ablation (right)
Carried burning time is lengthened in proportion to the size of the new instru-
ment with eight straight needle electrodes. However, the burning time was longer
than the standard electrode Habib 4 X ablated tissue volumes was the wind and the
possibility of separating the liver was faster. This is good operating performance,
which require fast drastic action in the shortest time for example in a car accident
and torn kidney, possibly massive bleeding when surgery.
Fig. 24. Measurement the linear octopolar tool 70W upright with the surface of the liver - 3
s after the start of the ablation (left) end of the ablation (right)
Simultaneously with the introduction of electro-surgical instruments was a

problem in the injection, when the introduction of power tools has been around
two times more than Habib 4x, which corresponds to the strength during the appli-
cation of about 26.4 N. This fact in applying the physician was identified as a
negative. With this and related position of needle electrode during ablation, since
their imposition in the tissue was strictly parallel.
Fig.25. Linear radiofrequency octopolar tool for the introduction into the tissue (left), refer-
ence cut for quality ablation verification. (right)
5.3 Discussion
As the biggest problem with this new design appeared deflection electrode
needles, which occurred after the passage of liver tissue. Despite this, the instru-
ment has demonstrated the need for lower operating time and the ability to burn a
larger section than the existing tools. Another advantage was to ensure compatibil-
ity with RF generator RITA waves.
6 Conclusion
Application of radiofrequency energy to the liver tissue is at present a unique op-
portunity to solutions to issues of trauma and cancer of the tissues. The current
surgical tools are allowing nearly universal application of radiofrequency in the
tissue. Weaknesses applicability as a resection and ablation of large pieces of tis-
sue removed our proposed linear radiofrequency octopolar tool. This tool allows
one site with a comparable time of application to remove three times the volume
of the liver. The advantage of the application is only one inject in the tissue and
reduce operating time than in any re-application of Habib 4X. The disadvantage of
the system is only comparable to radiofrequency coagulation time as the Habib 4X
and great force of the site with the introduction of instruments into the liver until
20.4 N. The disadvantages can also sort the uneven distribution of needles to in-
ject a result of tissue resistance and swerving to the side electrodes. This causes a
burning sensation after inhomogeneous tissue and surgery to cause real massive
bleeding. A partial solution to the problem is to use a sliding template tense.
Acknowledgments. The work and the contribution were supported by the project: Ministry
of Education of the Czech Republic No. MSM6198910027 Consuming Computer Simula-
tion and Optimization. Student grant agency SV 4502241, SP2012/114 “Biomedical engi-
neering systems VIII” and TACR TA01010632 “SCADA system for control and measure-
ment of process in real time”. The paper has been elaborated in the framework of the
IT4Innovations Centre of Excellence project, reg. no. CZ.1.05/1.1.00/02.0070 supported by
Operational Programme 'Research and Development for Innovations' funded by Structural
Funds of the European Union and state budget of the Czech Republic.
References
[1] Skalicky, T., Treska, V., Snajdauf, J.: Chirurgie jater, Praha. Maxdorf, pp. 25–138
(2004)
[2] Blumgart, L.H., et al.: Surgery of the liver, biliary tract, and pancreas, 4th edn., pp.
1269–1416. Saunders, Philadelphia (2007)
[3] Lencioni, R., Crocetti, L., Cioni, D., Della Pina, C., Bartolozzi, C.: Percutaneous ra-
diofrequency ablation of hepatic colorectal metastases: technique, indications, results,
and new promises. Invest Radiol. 39, 689–697 (2004)
[4] Strasberg, S.M., Linehan, D.: Radiofrequency ablation of liver tumors. Curr. Prob.
Surg. 40, 451–498 (2003)
[5] Bowles, B.J., Machi, J., Limm, W.M.L., et al.: Safety and efficacy of radiofrequency
thermal ablation in advanced liver tumors. Arch. Surg. 136, 864–869 (2001)
[6] Abdalla, E.K., Vauthey, J.N., Ellis, L.M., et al.: Recurrence and outcomes following
hepatic resection, radiofrequency ablation, and combined resection/ablation for colo-
rectal liver metastases. Ann. Surg. 239, 818–827 (2004)
[7] Ayav, B.P., Habib, N.A., Pellicci, R., Tierris, J., Milicevic, M., Jiao, L.: Impact of ra-
diofrequency assisted hepatectomy for reduction of transfusion requirements. AM J.
Surg. 193, 143–148 (2007)
[8] Jarnagin, W.R., Gonen, M., Fong, Y., et al.: Improvement in perioperative outcome
after hepatic resections: Analysis of 1803 consecutive cases over the past decade.
Ann. Surg. 236, 397–406 (2002)
[9] Weber, J.C., Navarra, G., Jiao, L.R., et al.: New technique for liver resection using
heat coagulative necrosis. Ann. Surg. 236, 560–563 (2002)
[10] Milicevic, M., Bulajic, P., Zuvela, M., Dervenic, C., Basaric, D., Galun, D.A.: Radio-
frequency-assisted minimal blood loss liver parenchyma dissection technique. Dig.
Surg. 24, 306–313 (2007)
[11] Kaspar, M.: Radiofrekvenční ablace metastáz jater, Přehledný článek referátového
výběru z radiodiagnostiky svazek 50, č. 1/2004 (2004)
[12] Allgaier, P., Deibert, P., Zuber, I., Olschewski, M., Blum, H.E.: Percutaneous radio-
frequency interstitial thermal ablation of small hepatocellular carcinoma. Lancet 353,
1676–1677 (1999)
[13] Bilchik, A.J., Rose, D.M., Allegra, D.P., Bostick, P.J., Hsueh, E., Morton, D.L.: Ra-
diofrequency ablation: a minimally invasive technique with multiple applications.
Cancer Journal 5(6), 356–361 (1999)
[14] De Baere, T.: RF ablation targets extrahepatic tumors. Diagnostic Imaging Europe,
30–47 (2003)
[15] Dowlatshahi, K., Bhattacharya, A.K., Silver, B., et al.: Percutaneous interstitial laser
therapy of patient with recurrent hepatoma in transplanted liver. Surgery 112,
603–606 (1992)
[16] Lencioni, R., Golleti, O., Armilotta, N., Paolicchi, A., Moretti, M., Cioni, D., Dona-
tin, F., Cicorelli, A., et al.: Radiofrequency thermal ablation of liver metastases with
cooled tip electrode needle: results of a pilot clinical trial. Europ. Radiol. 8,
1205–1211 (1998)
[17] Livraghi, T., Goldberg, S.N., Lazzaroni, S., et al.: Hepatocellular carcinoma: radio-
frequency ablation of medium and larger lesions. Radiology 214, 761–768 (2000)
[18] Rossi, S., Di Stasi, M., Buscarinin, E., et al.: Percutaneous RF interstitial ablation in
the treatment of hepatic cancer. Amer. J. Roentgenol. 167, 759–768 (1996)
[19] Solbiati, L., Goldberg, S.N., Ierace, T., et al.: Hepatic metastases: percutaneous radio-
frequency ablation with cooled-tip electrodes. Radiology 205, 367–373 (1997)
[20] Vogl, T., Müller, P.K., Hammerstingl, R., Weinhold, N., Felix, R.: Malignant liver
tumors treated with imaging guided laser induced thermoterapy, technique and pros-
pective results. Radiology 196, 257–265 (1995)
[21] Andrasina, T., et al.: Intervenční radiologie v léčbě maligních procesů jater. Postgra-
duální medicína, čís 2, 203–207 (2008) Dostupné online. ISSN: 1212-4184
[22] Hora, M.: Nádory ledvin. Urologie pro praxi 6(1), 28–30 (2005) Dostupné online.
ISSN: 1213-1768
[23] S.A.B. Impex, http://www.sab-medical.com/radiofrekvencni-
lecba.php
[24] Novakova, M., Derner, M.: Využití metody radiofrekvenční ablace v léčbě tumorů.
Journal Lékar a Technika, ROČNÍK 38(4) (Deember 2008)
[25] Zivcak, J., Hudak, R., Tkacova, M.: Termovízna diagnostika / - Košice, p. 200 (2010)
ISBN: 978-80-553-05333-2
[26] Flores-Sahagun, J.H., Vargas, J.V.C., Mulinari-Brenner, F.A.: Analysis and diagnosis
of basal cell carcinoma (bcc) via infrared imaging. Infrared Physics and Technolo-
gy 54(5), 367–378 (2011)
[27] Zivcak, J., Hudak, R.: Termovízna diagnostika / - Košice, p. 200 (2010) ISBN: 978-
80-553-05333-2
[28] Zivcak, J., Hudak, R.: Human limbs diagnostics by thermovision systems / - 2008. In:
AHFE International Conference, p. 7. USA Publishing, s.l (2008) ISBN:
9781606437124
[29] Steketee, J.: Spectral emissivity of skin and pericardium. Physics in Medicine and Bi-
ology 18(5), 686 (1973)
[30] Togawa, T.: Non-contact skin emissivity: measurement from reflectance using step
change in ambient radiation temperature. Clinical Physics and Physiological Mea-
surement 10(1), 39 (1989)
[31] Zivcak, J., Hudak, R., Kloc, J., Michalikova, M., Toth, T.: Základy bioniky a biome-
chaniky, vol. 1, p. 256. ManaCom, vyd - Prešov (2004) ISBN: 80-89040-25-X
[32] Hudak, R., Gal, P., Toporcer, T., Kilik, R., Sabo, J.: Mechanical tensile strength tests
of rat skin wounds during the first seven days of the healing process. Journal of Met-
als, Materials and Minerals 14(1), 11–19 (2004) ISSN: 0857-6149
[33] Hudak, R., et al.: Mechanical tensile strength tests of rat skin wounds during the first
seven days of the Healing processes. Journal of Metals, Materials and Minerals 14(1),
11–19 (2004) ISSN: 0857-6149
[34] Klener, P.: Klinická onkologie, Galén, p. 237 (2002)
[35] Válek, V., Kala, Z.: Maligní ložiskové procesy jater. Avicenum (2006)
[36] Clasen, S., et al.: Multipolar Radiofrequency Ablation with Internally Cooled Elec-
trodes. Radiology 238, 881–890 (2006)
[37] Simsik, D., Porada, V., Majernik, J.: Marker Free Analysis of Gait Pattern Applied in
Identification of Individuals. In: 5th European Academy of Forensic Science Confe-
rence, EAFS 2009, Glasgow, Scotland, UK, p. 57 (2009)
[38] Berjano, E.: Theoretical modeling for radiofrequency ablation: state-of-the-art and
challenges for the future (2006), http://http.www.biomedical-
engineering-online.com/content/5/1/24
[39] Chány, I.A., Neuzen, U.D.: Thermal modeling of lesion growth with radiofrequency
ablation devices (2004), http://www.biome-dical-engineering-
online.com/content/3/1/27
[40] Liu, Z., et al.: Radiofrequency Tumor Ablation: Insight into Improved Efficacy Using
Computer Modeling. American Journal of Rentgenology 184, 1347–1352 (2005)
[41] Dziakova, M., Zelinsky, L., Simsik, D., Majernik, J., Dolna, Z.: Objectification of re-
habilitation treatment effects of static disorders and stereotype of gait using SMART
system. In: XIV International Scientific Congress of the Polish Society of Physiothe-
rapy, Wroclaw, October 19-21, pp. 248–251 (2007)
[42] Majernik, J., Simsik, D., Galajdova, A.: Gait Analysis in Clinical Praxis. Acta Me-
chanica Slovaca In: YBERC 2005, vol. 9, 303–306 (2005) ISSN: 1335-2393; Kra-
tochvíl, B., et al.: Úvod do studia materiálů. Vysoká škola chemicko-technologická v
Praze, Praha (2005) ISBN: 80–7080– 568–4
[43] Majernik, J., Galajdova, A., Simsik, D., Zelinsky, L., Zemberova, I., Vojtkova, M.,
Dziakova, M., Katuscakova, M., Knazovicky, P.: Some Clinical Studies of Human
Movement at the Rehabilitation Clinic. In: 12th Annual Meeting, ESMAC, Marseille,
France (2003)
[44] Elgendi, A., Vavra, P., ElKayal, E., Habib, N., Baki, N.A.: Radiofrequency Assisted
Hepatic Resection: a New Technique Using a Bipolar Device. Journal of Hepatolo-
gy 52(suppl. 1), S189–S189 (2010)
[45] Curro, G., Habib, N., Jiao, L., Baccarani, U., Scisca, C., Navarra, G.: Radiofrequen-
cy-Assisted Liver Resection in Patients With Hepatocellular Carcinoma and Cirrho-
sis: Preliminary Results. Transplantation Proceedings 40(10), 3523–3525 (2008)
Biomedical Applications of Diagnostics and
Measurements by Industrial Computer
Tomography
Radovan Hudák1, Jozef Živčák1, Peter Kaťuch1, and Bruno Goban2

1
Technical University of Kosice, Faculty of Mechanical Engineering, Department of
Biomedical Engineering and Measurement, Letná 9, 042 00, Kosice, Slovakia
{radovan.hudak,jozef.zivcak,peter.katuch}@tuke.sk
2
CEIT-KE, s.r.o., Tolstého 3/6, 040 01, Kosice, Slovakia
bruno.goban@ceit-ke.sk
Abstract. The aim of presented study is a review on biomedical applications of

industrial computed tomography (CT). Industrial CT enables in contrast to medi-
cal CT scanning of technical materials (plastic materials, low density metals,
wood, etc.) but also living tissues in vitro. Important parameters for scanning are
maximum size of the part 300 x 300 x 300 mm; and the best obtainable resolution
of the digitalized scan is 9 µm. The termination for the scanning is also density of
material, where low density medical materials like titanium or medical polymers
are applicable. Presented review shows potentials of industrial CT for biomedical
applications; an assembly inspection, damage analysis, inspection of materials,
porosity analyses, conventional defect, and the reverse engineering.
Paper presents several biomedical studies realized by industrial CT, including
epithetic foot and obtained 3D model for inner structures and materials analysis,
epithetic hand glove for material homogeneity analysis, where recent thickness is
required, reverse engineering of human teeth to get the STL file for further appli-
cations (fabrication by additive manufacturing), animal skulls scanning and digita-
lization for preservation of museum exhibits, scanning of orthosis and total hip
replacements. The end of the study shows the full process of porous implant de-
sign, using metrotomographic bone-like porous structure.
1 Introduction
More than one hundred years ago X-ray technology started its triumphal
procession when Wilhelm Conrad Roentgen discovered a new kind of radiation in
his laboratory in Wuerzburg, Germany in the year 1895. Up to this moment most
of the developments on X-ray technologies and computed tomography have been
focused on special medical applications. Another twenty years later computer
tomography (CT) has become a powerful, well accepted tool in industrial
applications as well. Today industrial CT is on its way to become a major tool of

336 R. Hudák et al.
industrial and medical tools, devices and aids quality control in high-tech
branches, not only for material testing but for geometry analysis as well [13].
Besides the challenge to further optimize the CT systems hardware, its perfor-
mance and capabilities, the probably biggest challenge at this moment is to
process the huge amounts of data resulting from today’s CT scanners in reasona-
ble amounts of time.
The result of a CT scan displays the part as a three-dimensional image composed
of so called voxels [1]. Each voxel has a gray value that represents the local X-ray
absorption density. The resolution of this 3 D imaging system is given in part by the
number of voxels in the 3 D image, which in turn is given by the number of pixels in
the detector. By using sub-pixel respectively sub voxel interpolation the effective
resolution can be increased. Using techniques derived from 2 D image processing, 3
D voxel images can be evaluated to arrive at dimensional measurements for the
scanned parts. This approach is especially advantageous because the 3 D image
reproduces the complete part including both outer and inner measurement features.
Ultimately the user needs to know whether the part under inspection meets his toler-
ance requirements and whether his measurement device can answer this question or
not. In addition, reproducibility and operator independence are important criteria for
the shop-floor deployment of a measurement tool.
Industrial CT uses a series of 2-dimensional (2 D) images taken at specific in-
tervals around the entire sample. Basically any type of industrial CT system uses
three principal components: an X-ray tube, an X-ray detector, and a rotational
stage. Everything is enclosed within a radiation shielding steel/lead/steel cabinet
that typically ranges between four and 10 feet cubed. This allows use of the sys-
tem in a public environment without any additional safety concerns. Micro com-
puted tomography (micro-CT) is primarily the same as standard CT except it uses
a micro focus tube instead of a traditional tube. A micro-CT scan yields resolu-
tions in microns because the focal spot of a micro focus tube is only a few microns
in size. For comparison, micro-CT resolution is about 100 times better than the
best CAT scan in the medical field. Today also nano focus tubes with focal spot
less than one micrometer are presented for very precision measurement and most-
ly for material inspection.
In biomedical applications, micro-computed tomography scanners can function
as scaled-down (i.e., mini) clinical CT scanners that provide a 3 D image of most,
if not the entire, torso of a mouse at image resolution (50–100 μm) scaled propor-
tional to that of a human CT image. Micro-CT scanners, on the other hand, image
specimens the size of intact rodent organs at spatial resolutions from cellular (20
μm) down to subcellular dimensions (e.g. 1 μm) and fill the resolution-hiatus be-
tween microscope imaging, which resolves individual cells in thin sections of
tissue, and mini-CT imaging of intact volumes.
High quality industrial X-ray detectors used for CT are typically a new generation
amorphous silicon flat panel area detector. They offer a very high sensitivity,
resolution and bit depth. The resulting 2D X-ray images are very clear and the con-
trast is unparalleled. A modern high-end CT scan consists of taking several 2D X
ray images around the object, preferably covering 360 degrees (complete rotation).
CT systems typically acquire between 360 images (one image every degree) and
Biomedical Applications of Diagnostics and Measurements by Industrial CT 337
3600 images (one image every 0.1 degree) depending on the final desired resolution.
Each image is between three to 10 megapixels and is also averaged and filtered to
reduce noise. The 2D digital images taken during this step are saved directly into a
single folder, which will be used in the next step of the CT process [4].
Once the acquisition process of the CT scan is completed, CT calibration and
CT reconstruction algorithms are used to reconstruct the 3D CT volume. These 3D
images are made of voxels (three-dimensional pixels), and with the use of visuali-
zation software the 3D volume can be manipulated in real time. Because of this it
is possible to slice through anywhere inside the object, inspect and look for de-
fects, take accurate measurements, reconstruct a surface model and so forth. In-
dustrial CT technology is improving very quickly. While a few single CT slices
would take hours to generate years ago, it is now possible to reconstruct complete
3D models with billions of voxels in just seconds. This opens the door for numer-
ous new applications like 3D reverse engineering, rapid prototyping, 3D metrolo-
gy and more. In that regard, industrial CT has become a very competitive
technology for 3D scanning. The principal benefit of using 3D CT for scanning or
digitization is that we obtain a complete model with both external and internal
surfaces of an object without destroying it.
One of the new systems available today is Metrotom (Carl Zeiss, Germany).
This system opens beside industrial applications also new applications in the area
of biomedical engineering.
2 Materials and Methods
Metrotomography presented by Metrotom (Fig. 1) device uses X-ray technology

based on a simple principle: an x-ray source illuminates an object with an electro-
magnetic beam – the x-ray beams. The beams meet on a detector surface and are
recorded in varying degrees of intensity depending on the thickness of the material
and its absorption characteristics. The result is a two-dimensional gray-scale im-
age. However, this image is only meaningful for visual inspection when shown as
a cross section. Metrotom rotates the component 360° around its own axis, thus
producing a 3D image of the interior of the part.
Metrotom has a measuring volume of 300 mm x 300 mm. The ability to per-
form a metrotomograph of a material depends on the principle of cumulated ma-
terial thicknesses. It is possible to metrotomograph plastics, for example, up to a
thickness of 250 mm; light metal alloys such as aluminum or magnesium up to
120 mm. It is even possible to perform a tomograph on steel up to a thickness of
10 mm; the pure defect check also works up to a thickness of 50 mm.
The metrotomography procedure provides testing technology with a complete
range of evaluation possibilities: from assembly inspection to damage analysis,
inspection of materials, porosity analyses and conventional defect checks. At a
glance, metrotomography now clearly shows the defects in areas where previously
an inspection was not possible at all, or only using very timeconsuming and costly
cross section.
338 R. Hudák et aal.
Workpiece data record ded using the metrotomography procedure can be applieed
to all areas of quality asssurance and evaluated to traditional metrology, reversse
engineering applications and
a comparison of geometries.
Fig. 1. Metrotom (Carl Zeisss, Germany) and X-ray sensor
Metrotomography usess software environment called Calypso. Every metrotoo-

mographed component (w whether a standard geometry or freeform surfaces) can bbe
extensively evaluated. Byy Calypso, the following steps can be performed for mee-
trological purpose: readin
ng the features and orientation from the drawing and spee-
cify a measurement strateegy, generating a CNC measurement plan based on thhe
computed STL or CAD data,d entering and aligning the CAD model with the C CT
data, starting new CNC measurement
m plan and evaluate the protocol.
Fig. 2. Technology Center off Computed Tomography, Department of Biomedical Engineeer-

ing, and Measurement, Techhnical University in Kosice
Industrial CT hardwaree and software is placed at the Department of Biomediccal

Engineering and Measureement at the Technical University of Košice (Fig. 2.).
The industrial CT iss the ideal and, primarily, fast solution, particularlly
when there is no CAD data
d on a component (reverse engineering). When evaa-
luated workpiece (e.g. cu
ube) is tomographed, a point cloud is computed from thhe
Biomedical Applications of Diagnostics
D and Measurements by Industrial CT 3339
resulting 3 D model. Splines and knots are generated via reverse engineering. A
complete CAD model, wh hich can be displayed in all standard formats, is available
at the end of the process. Duration
D of this process is approximately 1 hour.
For comparison of geeometries, standard data formats such as IGES, STL L,
VIDA and STEP can be easily compared using a 3D model generated with thhe
Metrotomography proced dure. The plan/actual comparison work (e.g. a dampeer
bearing evaluation) as it is
i described in following steps: scanning the CAD desiggn
data of the bearing, enterring the point cloud of the bearing from the metrotomoo-
graph, aligning the data sets of each other, displaying of the deviations as mea-
surement flags or extensiv ve, colorful displays of the deviations. The utilization oof
Metrotom is presented by b measurement of the multidirectional valve (Fig. 33).
Following pictures show description of reverse engineering function of Metrotom m,
material inspection and metrological
m possibilities of the scanner and software.
Fig. 3. CAD models of the multidirectional

m valve – presentation of the reverse engineeringg
In particular steps an X-rays

X scan rotated measured object. From X-rays the poo-
sition of each voxel is callculated. According to density of material, each voxel haas
a value from the gray scaale in the range from 0 to 65535. This allows to separaate
low density materials as an air and other support materials. The 3 D point clouud
gives a possibility to see the
t component from indifferent direction. Therefore, it is
possible to evaluate not only
o the surface of the model but also the inner environn-
ment or cross-section.
Fig. 4. An analysis of the con

nstructional junctions in multidirectional valve
Figure 4 presents insp pection function of metrotomography. An advantage oof

this method is possibilityy of inspections of constructional junctions in assembleed
work pieces (contact surffaces analysis, quality control, and detection of foreiggn
materials or air spaces).
Fig. 5. Geometry deviations analysis of the multidirectional valve
Metrotom measures wiith relatively high accuracy. Calypso software allows eva-
luating of each obtained component.
c By comparison of the CAD model (Fig. 5, 66),
which is created with nomminal geometry and dimensions together with metrotomoo-
graphs is possible to generrate a view of geometry deviations. Calypso also offers aan
evaluation of length param
meters, angles and geometry or position deviations.
Fig. 6. CAD model (left) wiith CT data and evaluation of geometry deviations with defauult
toleration field (middle and right)
r
Metrotomography by CarlC Zeiss was used for scanning epithetic hand, foot annd
plastic skull (Figure 7). Above
A mentioned products were selected to check plasttic
materials, which are frequuently used in medical practice and biomedical engineeer-
ing research.
Fig. 7. Samples from the fiield of biomedical engineering evaluated by metrotomographhy

(epithesis and plastic human skull)
Also wooden materialss and metallic parts of skull were investigated in order tto
check the capability of ind
dustrial CT.
3 Results of Measu
urement
The products of biomediical engineering research or research in prosthetics annd
orthotics (Fig. 7) were scanned by industrial CT machine and investigated by diif-
ferent software applicatioons. Presented initial study shows functions and possibilli-
ties of industrial CT and software
s applications (Calypso and VG Studio Max).
Software VG Studio Max M (Volume Graphics) is a tool, which except visualiza-
tion of obtained data also o allows to apply following analysis: geometric analyssis
(2D & 3D dimensioning g, actual/nominal comparison, actual/actual comparison,
wall thickness analysis an nd extraction of STL surface from point cloud) and ma-
terial analysis (detection of
o defects, inclusions detection and possibility of materiial
distribution and position in
i composite material).
Fig. 8. Epithetic foot and obttained 3D model (visible transparency of lower density materiial
– soft surface plastic and wo
ood)
VG Studio Max or oth her related applications offers many useful tools and mee-
thods for data visualizatioon. The points cloud is possible to separate by grey coloor
intensity which representts different density of materials and resultant inner ma-
terial detection (Fig. 8). Within
W the artificial foot are visible reinforcing materiaals
and inner structures and holes
h for foot assembly.
By the translucency of cover material (skin color plastic) with lower densitty
and air spaces (originatin ng at the plastic casting) are getting more apparent. V VG
Studio Max enables to deetect material inhomogenity often presented by air cavvi-
ties. By use of histogram function in the relation with an air cavity geometry therre
xact position of the cavity.
is possible to assess an ex
Fig. 9. Epithetic hand glovee (left), its X-ray scan (middle) and an environment of the VG
G-
Studio Max software with thhree cross-sections and 3D view of rendered model of the glovve
An approximately 900 0 X-ray scans were captured by single steps during onne
revolution of the measureed object around the vertical axis. Subsequently, the 3D
model was processed (F Fig. 9, middle), which allows non-destructive cross-
sectional analysis. Figuree 9 right shows the window of VG Studio Max 2.0. IIn
four windows is possible to visualize the 3D view of rendered model and epithettic
glove cross-sections in peerpendicular views.
The software enables easily
e to use the measure features, e.g. a glove wall thickk-
ness. The histogram in th he right lower corner of the window gives informatioon
about number and types ofo used materials (different density) in observed object.
VG Studio Max allows an animation consisting of multiple overflights arounnd
the object or crossing thee object in optional direction. Data can be transported tto
generally supported STL format and further processed for reverse engineering oor
rapid prototyping/manufaacturing.
Fig. 10. Rendered 3D mod del created by point cloud grid (left), cross-section of the 33D
model in indifferent direction
n and cross-section of the glove model in one of the three view
ws
Fig. 11. Two views of skull X-rays

X
Different grey scales in

n presented scans (Fig. 11) are caused by reduction of XX-
ray energy detected by X-ray detector. This is presented by higher material thickk-
ness in the place where X-ray
X is passing through (skull borders) or higher materiial
density (springs and clasp
ps).
The skull model as it is shown on Fig. 12 consists of two types of materialls,
metallic springs and claspps and plastic skull. By separating of different materiaals
density, there is possiblee to hide selected material (e.g. only metallic parts arre
visible).
Fig. 12. By skull translucenccy is possible to see metallic components of the assembly (left
ft);
window with control tools off the VG-Studio Max software (right)
Technology center of computed

c tomography at Technical university of Kosicce
serves as support for man ny kinds of researches. Biomedical research or research iin
prosthetics and orthotics are important supported areas. Tomography can be useed
for various evaluations. One
O of the most used is digitalization of very complicateed
shaped objects. In biomed dical engineering there are many objects with so compleex
surfaces, that they are notn measurable by any other methods of digitalization.
Touching probes, lasers or optic scanners cannot reach areas which are mostlly
hidden or unobtainable. WithW CT scanners are obtained 100% information abouut
object shape. Because of that for example human teeth (Fig. 13) or animal skullls
(Figure14) can be digitizeed. VGStudio Max is software which allow extract ST TL
model from surface points. STL is universal triangular model which can be im m-
ported to every CAD/CAE E/FEM/RP software for further processing.
With CAD software itt is possible to do parameterization or design a supple-
ment for scanned objeect. In FEM (Finite element method) software thhe
stress/strain analysis can be
b performed. STL model is also ideal type of model foor
Rapid Prototyping. With 3D printer the plastic or even metal copy of real objeect
can be done. On the figuree 14 can be seen results from project of digitalization annd
printing of animal skullss. On the bottom picture there are the original monkeey
skull and printed copy off monkey skull. More about express creation of compleex
shaped object copy can bee found in authors other publications [10].
In area of orthosis dessign there are also very complicated plastic parts, whicch
are difficultly digitalized with any other scanning technology. Examples of digitaa-
lization of orthosis for haands made from low temperature thermoplastics (LTTP P)
are pictured on figure 15. With obtained data it is possible to do further analyssis
for example analysis of cu urvature, stress/strain analysis etc.
Fig. 13. STL format of humaan teeth (middle and right)
Fig. 14. Skulls of animals (to

op) and of monkey’s skull printed with 3D printer (bottom leftt)
Metrotom 1500 with maximum

m power of x-ray source 250 W is ideal for plasttic
ys. Also metals with higher density can be scanned witth
parts and light metal alloy
Metrotom 1500, but wall thickness of the part is limited. Nice example of digitaa-
lization for dental clinic is a STL model of metal dental bracket on Fig. 16.
The maximum dimension of this bracket is 3 mm. With so small parts it is possi-
ble to achieve resolution of pointcloud around 9µm. With sub-voxeling process
the resolution can be increased. Small objects can be magnified more then bigger
parts, because they are placed closer to the x-ray source. The size of the projection
is zoomed many times and because of that size of voxel (volume pixel) of final
point-cloud is just a few micrometers.
Fig. 15. Plastic orthesis (top) and their pointclouds (bottom)
X-ray is kind of radiation which can penetrate through objects. Because of that
we have information not only about surface of object, but also about inner volume.
After reconstruction of point cloud we can look inside the object by using virtual
cross-sections without destroying of real object or separate materials with different
density.
Fig. 16. Pointcloud of mall metal

m dental bracket
Inspection of inner stru

ucture of pork femur is pictured on figure 17. The different
gray-values depend on den nsity of material. The air around the bone is black, becausse
have a lower density. Thee outer structure of the bone has a higher density and thhe
color of the pixels in this area is light gray. The density of bode is changing from m
outer surface to inner volu
ume what is considerable from cross- section picture.
Fig. 17. Pork femur (left) and its virtual cross-section (right)
Another research supp ported by CT technology at our laboratory was porositty

analysis at two samples ofo human bones (Fig. 18). The research was focused tto
compare amount of solid phase of two human demineralised bone matrices, wherre
one of them was cultivateed with mesenchymal stem cells (MSC) for two weeks.
After cultivation the am
mount of solid phase increased rapidly and the results caan
be seen also on figure 18. Porosity of selected regions of interests have been calcuu-
lated and compared. Resullts of research can be found in authors’ publication [6].
Industrial tomography is i a branch of technology which gives not only advantagees

of nondestructive testing and
a viewing of inner structure, but the reconstructed daata
are so accurate, that this technology
t can be used for very precise geometrical mea-
surements. Also in biomed dical applications it is necessary to measure characteristiccs,
which are difficultly meeasurable by any other technology without destructioon.
Spherical form of used aceetabular component was measured with Calypso softwarre.
With Calypso we can evaaluate characteristics of objects and make a protocols oor
graphical presentations of results. Graphical interpretation of form deviations at
acetabular component in magnification
m are presented on figure 19. Deviations havve
been filtered to eliminate scratches
s on the spherical surface.
Quality of reconstructeed point cloud and accuracy of measured data markedlly
depends on density and cumulative
c wall thickness of scanned object. Because oof
that material with lower density
d are more suitable for scanning than denser matee-
rials. Titanium, in compaarison with other metals (steel, copper, etc.), is a mettal
with relatively low densitty and because of that small components made from titaa-
nium are measurable with hout restrictions.
Fig. 18. Demineralised human bone matrix without hMSCs (top) and demineralised humaan
bone matrix with hMSCs aftter two weeks of cultivation (bottom)
One of the latest researrches supported by our laboratory is digitalization of tita-

nium dental implants. Analyses of material homogeneity, porosity, assembly quaal-
ity and geometry have been b performed on four different implants (Figure 200).
With digitalized data we are able to do some modifications and print new modd-
ified dental implants withh appropriate rapid prototyping technology.
Fig. 19. Used acetabular com

mponent (left) and form deviations on spherical surface (right)
Fig. 20. Four dental implan nts (top left), rendered point-cloud (top right), cross-section of
point-cloud (bottom left) and
d detail to STL file imported to metrology software Calypso
4 Example of the Use of Tomography in Biomedicine
CT scan of the patient head was obtained by medical CT in clinical environment.

For further reconstruction and implant design the optimal density of the imaging is
less than 1mm. Medical screenings are stored in normalized format DICOM (Digi-
tal Imaging and Communication in Medicine). Nowadays exists several generation
of DICOM format, so software application are adapted (MIMICS, 3MATIC -
Materialise, Belgium).
Transformation of DICOM format to 3D model – for that purpose is used
MIMICS software. DICOM format is in general the series of the slices – screens
located one above each other, so finally they represents complete model. Single
shade of grayscale represents tissues with different density (Fig. 21).
Fig. 21. Mimics software environment
It is necessary to separate soft and hard tissues, therefore to separate bone tissue
and other human tissues. By tool of the mask, the requested effect is achieved.
For next steps it is necessary to modify colors by contrast and brightness level,
so final model will represent only requested areas. In some cases is more effective
to convert to 3D also soft tissues. For instance in cranial implants the design has to
follow brain tissue in order not to create a pressure. After mask release, software
creates the 3D model.
Fig. 22. Steps of creation of 3D model form DICOM format (CT)
3D model is created ass a cloud point that has its own surface structure and it is
often needed to clean the surface of the noise and other extra tissue burrs by sofft-
ware tools. For that appliccation the anatomy experiences level is important, as duue
to low level of the knowlledge some important areas can be deleted and it can aaf-
fect final implant design.. By this modification, model is prepared for further inn-
duction. It is saved in MCS format. Software can not only read DICOM formaat,
but also bitmap pictures or
o jpegs. Due to older formats of DICOM it is necessarry
to perform manual setup ofo the resolution (Fig. 22).
4.1 Adaptation of 3D
D Model and Implant Design
After creation of the 3D model,
m the next step was made by 3MATIC (Materialisse,
Belgium) software appliccation. This software allows transforming uncounted 3D
model to series of polygon
ns.
Fig. 23. Model represented by

b mesh
The size of the trianglees represents the precision and the quality of the model –
more triangles, smoother surface of the model (Fig. 23). On the start of the process
steps of the optimization of the triangles were performed. Like that, the work witth
the model is smooth. Nex xt step is the modification necessary to link chosen typpe
of implant and its localizzation. In case of damaged jawbone, over-material thhat
was in potential interferennce with implant was removed. In this step the consulta-
tion with the surgery is reequired for preparation of the general requests for contaact
points/areas and final application of the implant.
Fig. 24. Damaged area and implant replacement
Missing part of the moodel is reconstructed based on present quality of the bonne
tissue. The bone-like geo ometry was obtained by industrial CT (Metrotom 1500,
Carl Zeiss, Germany). Im mportant is the complex geometry of the surface that cann’t
affect physiological function of the jaw. Also the replacement must by morphologg-
ically similar with missinng part. In case of differences of the size of the implannt,
the complete change of raatios of the jawbone can happen and that can lead to abb-
normal changes in the joiint. This difference can affect other parameters in maxiil-
lofacial area with potential pathologic effect to complete system. After modelinng
of the replacement it is necessary
n to perform analysis of the bone wall structurre
thickness. This step is immportant to create holes and interfaces for fixation of thhe
implant to the bone (Fig. 24). Openings are designed individually and it is necees-
sary to involve also posittioning that cannot affect important anatomic points, likke
vessels, nerve etc. Numb ber of the openings supposed to be limited, but enouggh
robust to fix the implant to
t the bone. For better interface contact between implannt
and bone, structure of the implant is made with predefined porosity (Fig. 25). Thhat
allows healthy bone to grow
g through the implant and stabilize its position. Thhis
stabilization is expected due
d to fact that jawbone is the part of articular system oof
human and it is exposed to o all possible mechanical stresses.
Afterwards it is suitablle to repair texture errors like overhanging triangles, unn-
wanted holes etc. These errors can affect the relief of the final design and afteer-
wards the implant is ready y for production.
Fig. 25. Creation of porous implant

i
5 Conclusion
Study shows possibilities and limits of industrial CT used for different biomedical
applications. Where betterr resolution of output data is required, there is a space foor
micro CT technology, wh here on the other hand often the size of analyzed object is
limited. Data obtained by
y metrotomography are usable in many biomedical applli-
cations including materiaal structure and homogeneity analysis, assembly inspecc-
tion of medical aids or devices, evaluation of geometry deviations – qualitty
control in medical producction or reverse engineering. After acquisition of digittal
point cloud, the software tools enable different metrological or material analysiis;
an added value in biomeddical research and development.
Acknowledgments. This contribution is the result of the project implementation: Researcch

of New Diagnostic Method ds in Invasive Implantology, MŠSR-3625/2010-11 Stimuls ffor
Reasearch and developmentt of Ministry of Education, Science, Research and Sport of thhe
Slovak Republic) and the prroject: Creation and promotion of technologies in diagnostics of
components and junctions with
w computed tomography (lTMS: 26220220038) supported bby
the Research & Developmen nt Operational Programme funded by the ERDF.
References
[1] Ritman, E.L.: Micro-Computed Tomography—Current Status and Developmentts.
Annual Review of Biomedical Engineering 6, 185–208 (2004)
gi, B., Weiß, D.: Means to Verify the Accuracy of CT System
[2] Lettenbauer, H., Georg ms
for Metrology Applicaations (In the Absence of Established International Standardss).
In: DIR 2007 - Intern
national Symposium on Digital industrial Radiology and Com m-
puted Tomography, Ly yon, France, June 25-27 (2007)
[3] Lee, Y.S., Seon, J.K., Shin, V.I., Kim, G., Jeon, M.: Anatomical evaluation of CT-
MRI combined femoral model. BioMedical Engineering Online 7, 6 (2008),
doi:10.1186/1475-925X-7-6
[4] Noel, J.: Advantages of CT in 3D Scanning of Industrial Parts. 3D Scanning Tech-
nologies Magazine 1(3), 18 (2008)
[5] Andreu, J.-P., Rinnhofer, A.: Modeling Knot Geometry in Norway Spruce from In-
dustrial CT Images. In: Bigun, J., Gustavsson, T. (eds.) SCIA 2003. LNCS, vol. 2749,
[6] Gromošová, S., Rosocha, J., Živčák, J., Hudák, R., Kneppo, P.: New modular semiau-
tomatic system for preparation of the demineralized bone matrix for clinical trans-
plantation. In: World Congress on Medical Physics and Biomedical Engineering –
WC 2006: Imaging the Future Medicine, August 27 - September 1, p. 4. Springer,
Heidelberg (2006) ISBN 978-3-540-36841-0
[7] Toporcer, T., Grendel, T., Vidinský, B., Gál, P., Sabo, J., Hudák, R.: Mechanical
properties of skin wounds after atropa belladonna application in rats. Journal of Met-
als, Materials and Minerals 16(1), 25–29 (2006) ISSN: 0857-6149
[8] Hutníková, L., Hudák, R., Živčák, J.: Scoliosis severity diagnostics: the new applica-
tion for automatic X-ray image analysis. In: Infusing Research and Knowledge in
South East Europe: 3rd Annual South-East European Doctoral Student Conference:
Thessaloniky, June 26-27, pp. 1791–3578 (2008) ISBN: 978-960-89629-7-2, ISSN:
1791-3578
[9] Gal, P., Kilik, R., Spakova, T., Pataky, S., Sabo, J., Pomfy, M., Longauer, F., Hudak,
R.: He-Ne laser irradiation accelerates inflammatory phase and epithelization of skin
wound healing in rats. Biologia. 60(6), 691–696 (2005) ISSN: 0006-3088
[10] Cierniak, R.: X-Ray Computed Tomography in Biomedical Engineering. Springer,
London (2011) ISBN: 978-0-85729-026-7
[11] Gajdos, I., Katuch, P.: Complex approach to the reverse engineering techniques.
Scientific Papers of University of Rzeszow: Zeszyty Naukowe Politechniki Rzes-
zowskiej: Mechanika 80(273), 81–86 (2010) ISSN: 0209-2689
[12] Wagner, J.: Contribution to the design optimization of electromechanical actuators of
mechatronics systems. Metalurgija/Metallurgy 49(2), 600–603 (2010) ISSN: 0543-
5846
[13] Penhaker, M., Darebnikova, M., Cerny, M.: Sensor Network for Measurement and
Analysis on Medical Devices Quality Control. In: Yonazi, J.J., Sedoyeka, E., Ariwa,
E., El-Qawasmeh, E. (eds.) ICeND 2011. CCIS, vol. 171, pp. 182–196. Springer,
Heidelberg (2011)
Infrared Thermography in the Evaluation
of Carpal Tunnel Syndrome
Mária Tkáčová1, Radovan Hudák2, Jozef Živčák2, and Ivana Živčáková3

1
CEIT-KE, Košice, Slovakia
maria.tkacova@ceit-ke.sk
2
Technical University of Košice, Slovakia
{radovan.hudak,jozef.živčák}@tuke.sk
3
University of Prešov, Slovakia
izivcak@unipo.sk
Abstract. The Aim of the study was to identify the temperature difference be-
tween physiologically healthy and pathologically hand affected by carpal tunnel
syndrome (CTS). For the purposes of this study were created two databases of 190
thermograms of dorsal side of hands. The first database consists of normal ther-
mograms of physiologically healthy hands called reference group (RGn= 174). The
second is the database with 16 pathological hands (CTSn=16) with clinically diag-
nosed CTS in 9 patients. Pre-surgical thermograms of the volunteers’ hands af-
fected by CTS were obtained using infrared thermography (Thermocamera Fluke
Ti55/20). We have observed the temperature distribution in the entire hand and
partial temperatures for two years. The main significant areas for this study were
the center point of carpals (T1) and the area of distal phalanges on middle fingers
(T5). The Index of Median Nerve (DMI=(T1-T5) were determined based on result-
ing data. Test results obtained from measurements of the two points on the dorsal
side of hands have showed that the temperature of CTS hands is typically higher in
the phalanges as opposed to the wrist (DMI≤0; 75,0%), while the temperature is
the lowest in the distal phalanges (T5) of healthy hands (DMI>0, 68.4%). The
calculated sensitivity of diagnostic test with the confidence interval 0,750±0.202 and
the specificity of diagnostic test with the confidence interval 0,684±0.069 show that
thermography could be a promising method in the diagnostic process of CTS.
1 Introduction
The goal of the experimental study is to discuss the role of medical thermography
in diagnostics of carpal tunnel syndrome (CTS). It is the most common form of
entrapment neuropathy with significant lifetime risk in the general population. It is
getting more common especially in repetitive task workers such as computer users
and has become an important challenge to clinicians [4, 7, 13].

356 M. Tkáčová et al.
Examination of this syndrome is a global object of the current investigations by

experts. Numbers of publications in databases such as PubMed (7563 papers),
Medline (7449), Springer (2,732) or Elsevier (36) are the evidence. Database of
MEDLINE in search ‘carpal tunnel syndrome thermography’ consists of 16 papers
in publication date from 1985 to 2009. Database of Springer or PubMed in search
‘carpal tunnel syndrome thermography’ consists of 28 papers in publication date
from 1969 to 2009.
This syndrome is often difficult to diagnose, and a late diagnosis may result in
permanent nerve damage. Electromyography (EMG1), ultrasonography (US2),
magnetic resonance imaging (MRI3), and computed tomography (CT4) may be
performed for the diagnosis. The diagnostic accuracy of these tests is well docu-
mented, but most of these studies accept EMG as the gold standard [4].
Although the primacy of the diagnosis, disadvantage of EMG is that it uses
electrodes either in contact with the body or invasive electrodes. Modern medicine
seeks to be contactless and noninvasive, which also makes thermography a suita-
ble diagnostic method. The exploration of the skin surface temperature of the
affected areas may provide additional information as regards to the possible ab-
normalities in the circulation and accordingly in the function of the sympathetic
nerves.[4] A nerve conduction study, inflammation and temperature distribution
may be of benefit to clarifying the diagnosis and this is the goal of this study.
2 Medical Thermography in the Diagnosis of CTS
Medical thermography could be a great tool for diagnostic of surface temperature

of the human body. It is because it represents non-invasive, non-contact, safely,
no-radiation and painless image technique, which could be used for diagnostic in
oncology (breast cancer, skin cancer, etc.), pain (management / localization / con-
trol), dermatological diseases, etc. Symptoms of Carpal Tunnel Syndrome are
often associated with temperature variations on skin surface such as inflammation,
paresis or plegia, pain or athrophy, which could be visible by the use of medical
thermography [2, 5, 6, 7, 10, 11, 12, 13].
Different clinical tests are helpful in diagnosis of the syndrome for example Tinel
sign; Phalen test and carpal tunnel compression are known provocative tests. To
confirm the diagnosis, nerve conduction study is used which is the most sensitive
method of diagnosis and grading of the severity of the syndrome and evaluation of
other problems. Nerve conduction studies are normal in 13–27% of the patients that
can make some problem in diagnosis and follow up of the patients [18].
1
Carpal tunnel syndrome EMG’ - 901 registered articles in PubMed database.
2
Carpal tunnel syndrome US’ – 389 registered articles in PubMed database.
3
‘Carpal tunnel syndrome MRI’- 324 registered articles in PubMed database.
4
‘Carpal tunnel syndrome CT’- 93 registered articles in PubMed database.
Infrared Thermography in the Evaluation of Carpal Tunnel Syndrome 357
The documentation of the functional abnormalities in the median nerve is impor-

tant for the diagnosis of CTS as well as for postoperative follow up. The nerve con-
duction studies (NCS) and the electromyography (EMG) are the fundamental
methods and considered as golden standard in the diagnosis of CTS. However, with
NCS, only the function of the myelinated fibers (fast fibers) can be examined, the
responses of unmyelinated fibers, e.g. Cafferents and post ganglionic sympathetic
fibers cannot be detected [4].
Fig. 1. Schematic of EMG electrodes in the examination of A) n.medianus, B)n.ulnaris
Other methods, such as magnetic resonance imaging (MRI) and ultrasonic exami-
nation, can only reveal the structural changes of the carpal tunnel and its contents [4].
For diagnostic is very important to compare the diagnostic utility of Electromyo-
graphy (EMG), Ultrasonography (US), Computed Tomography (CT), and Magnetic
Resonance Imaging (MRI) in idiopathic Carpal Tunnel Syndrome determined by
clinical findings. From one of these studies were used 69 patients with clinically di-
agnosted bilateral CTS. It was evaluated that EMG or US could be used as the first-
step test in most cases. If they are both available, EMG should be the first choice.
They may be performed together when diagnosis is challenging. CT may especially
be preferred for bone-related pathological conditions, while MRI may be preferred
for soft tissue-related pathological conditions. Even though imaging studies have
been proven to be powerful diagnostic tools for CTS, no conclusive information cur-
rently exists to support replacing EMG with imaging studies [16].
In patients with a clinical diagnosis of CTS, the accuracy of sonography is
similar to that for EMG. Sonography is probably preferable because it is painless,
easily accessible and preferred by the patients [17].
Nerve conduction study is the most sensitive test for diagnosis of carpal tunnel
syndrome (CTS). This test is normal in some patients with mild CTS. Median
nerve conduction study evaluation after a provocative test (e.g. wrist flexion) may
be helpful for diagnosis of mild CTS [18].
Jesensek Papez B. et al. (in 2008 and 2009) using a large database of 502 ther-
mal images of the dorsal and palmar side of 132 healthy and 119 pathological
hands. It confirmed the hypothesis that the dorsal side of the hand is of greater
importance than the palmar side when diagnosing CTS thermographically. Using
this method it was possible to classify correctly 72.2% of all hands (healthy and
pathological) based on dorsal images and more than 80% of hands when only
severely affected and healthy hands were considered [1, 3].
Ming at al. in study: “Postoperative relieve of abnormal vasoregulation in car-
pal tunnel syndrome“ (Clin Neurol Neurosurg, 2007) used forty-one healthy hands
(n=41) of 22 volunteers and 22 hands (n=22) with clinically diagnosed CTS of 16
patients were examined. A series of infrared photos of the hands of each subject
were taken and stored by using digital infrared thermography (DIRT) before and 6
months after the CTR. The temperatures of the finger tips from digit 1 (D1) to
digit 5 (D5), the center point of thenar (Th) and hypothenar (Ht) eminences were
measured, the median nerve index (MI=(D1-D2)+(D1-D3)+(D2-D3)) and the
temperature differences between the median and the ulnar nerve distribution area
(MED.ULN=(D1-D5)+(D2-D5)+(D3-D5)+(Th-Ht)) were calculated. The results
of DIRT measurements suggest that the blood flow regulation in CTS is abnormal
possibly because of disturbed sympathetic vasomotor regulation and that the circu-
lation gets normal together with the alleviation of the other symptoms of CTS as
recorded 6 months after CTR operation [4].
On the basis of this knowledge, also based on our measurement, we have set up
our own method of processing thermograms, which is described in this study.
3 Carpal Tunnel Syndrome
Carpal tunnel syndrome (CTS), or median neuropathy in the wrist, is a medical

condition in which the median nerve is compressed at the wrist, which leads to
paresthesia, numbness and muscle weakness in the hand. As a result of these pre-
disposing factors, the finger flexor tendons cause friction as they move over one
another and the generated heat causes local inflammation. This inflammation irri-
tates the nerve and the resulting pain is what patients experience when they have
CTS. The reason there is so much pain associated with CTS is because there is a
lot of pressure on the Median Nerve [8].
CTS is the most common form of entrapment neuropathy with significant life-
time risk in the general population. It is getting more common especially in repeti-
tive task workers such as computer users and has become an important challenge
to clinicians [4].
4 Methodology of Measurement
The Methodology of this study has consisted of conditions for Object, Infrared
Thermal Imaging Camera, Room and Processing for all two years.
5 Object of Measurement
For this study was created a database of reference thermograms (RG; RGn=174)
and database of pathological (CTS) thermograms (CTS; CTSn=16). Altogether 7
people with bilateral positive carpal tunnel syndrome (CTS) and 2 persons with
unilateral disability have undergone thermographic diagnostics. The entire patho-
logical group was confirmed by test of EMG. Majority of patients represents
women in middle age (CTSnF=14). Current (thermographic) research has focused
on the dorsal side of hands [1, 3, 7]. Therefore our calculations have been focused
on thermograms of dorsal surface. The study has based on comparison of tempera-
tures in healthy hands, which represents reference thermograms and pathological
thermograms.
All volunteers were asked to complete a questionnaire regarding his/her general
information, and to have a physical examination positive EMG. The volunteers
who had a history or physical examination suggestive of diabetes mellitus, exces-
sive alcohol consumption, or exposure to toxins were excluded. The hands with
wounds or acute infection were also excluded from the measurements.
5.1 Instructions before Infrared Imaging
• no application / removal of skin creams and cosmetics on the study area;

• avoid eating and excessive intake of tea or coffee immediately before the ex-
amination (cca2 hours);
• ban the use of drugs (completely) and smoking (about 2 hours before the ex-
amination) [19];
• avoid higher physical stress (eg, rehabilitation, training, etc.) or mental stress;
• prudent to avoid medicaments affecting the cardiovascular, musculoskeletal or
neurovascular system (unless contraindicated by physician).
5.2 The Reference Database
We have been creating the database of 174 thermograms of healthy hands

(RGn=174) for two years. The name of this database is Reference Group (RG). The
database consists of 87 right (PnRG=87) and 87 left (LnRG=87) healthy hands with-
out any symptoms in the upper limbs. The database is composed of healthy young
people in the average age of 33.831 years. They have been young people in daily
use of computer mouse and keyboard [20]. This utilisation of computer has been
representing a long-term unilateral load for hand.
5.3 The Pathological Database
The pathological group consisted of hands affected by carpal tunnel syndrome

(CTSn=16). For study were used 16 pathological hands of 9 patients. Seven of them
were diagnosed the bilateral disability and two of them were clinically detected
the unilateral disability. Each of them had been diagnosed CTS with electromyo-
graphy (EMG) 1month before infrared examination. The average age of patients
was 41.235.
6 Conditions of Measurement for Environment
Measurements have been carried out under the same conditions, still in the same
room with the ambient temperature about 21.5°C (±2°C) for two years. In this
room were retracted blinds to avoid the impact of solar radiation and the room
was equipped with air conditioning, because of the condition to maintain the same
temperature at each measurement [5, 14].
7 Methodology for Obtaining Thermograms
Thermography is a temperature measurement technique used to produce a colored

visualization of thermal energy emitted by the measured site at a temperature
above absolute zero. Each pixel in the image depicts the radiance falling on the
focal plane array/ microbolometer– type detector used in an IR camera [9].
Skin temperature on dorsal side of hands from our database (n= 190) was meas-
ured with the Infrared Thermal Imaging Camera (ThermaCam Fluke Ti55/20, Fluke,
USA). This thermographic camera produces a matrix (representing image points) of
temperature values. The thermal sensitivity of the thermograph is 0.05°C at 30°C.
Camera works in the spectral range from 8µm to 14µm (human body infrared radia-
tion is highest in the spectral range around 9.66µm) and the calibrated temperature
range from -20°C to 100°C. Camera resolution is 320×240 pixels (total 76800 pix-
els). Data were obtained through a high-speed (60Hz) analysis [9].
Emissivity of the skin was set in the camera software to 0.98 [9], the ambient
temperature was measured by infrared (laser) thermometer (Pyrometer Testo 810)
and for the control we used contact multimeter (Almemo Multimeter 2690, Ahl-
born). Before each recording the camera has calibrated using the system's internal
calibration process.
All thermograms (n=190) were processed with special software (SmartView
2.1, FLUKE, USA).
All thermograms from our both databases have been focused in grayscale. It is
the palette for obtaining thermograms. The grayscale is the best palette for distinc-
tion of human eye [9]. After in SmartView software was set Palette High Contrast,
there it is possible to narrow the temperature interval. For example the thermo-
gram from healthy right hand from volunteer Num.87 (RGx87=Px87; Fig.2) was
obtained in temperature interval <20.3; 38.7>. For analyze temperature for medi-
cal application is usually used Palette of High Contrast (Fig. 3). Processing ther-
mogram begins by deduction of Background temperatures and identifying key
areas of interest. Fig. 4 shows thermogram of volunteer Num.87 (RGx87=Px87) in
High contrast with narrow temperature interval <26.6; 32.3>.
Fig. 2. Focus in Grayscale Fig. 3. Palette of High Fig. 4. High contrast with
Contrast; working palette narrow temperature interval.
8 Results
At the beginning of the study were established two hypotheses. The first hypothe-
sis: Distribution of temperatures on hands affected by CTS is significantly differ-
ent from normal temperature of hands. The second hypothesis: The highest tem-
perature difference is expected in the line of median nerve, focusing on the region
of III. finger.
9 Results Obtained from First Hypothesis

The average temperatures in both databases were obtained from tracing hand. It
was the way for creating polygons from interesting areas of temperature mea-
surement (Fig.5).
Fig. 5. Tracing hand and polygon of temperature measurement
By comparison of average temperatures in RG database between dominant and

non-dominant hands (RGn=174) in one person (RGX1…87) was found only small
temperature difference (Fig.6; Table 1). People in database were right-handed, so
dominant was right hand (PnRG=87) in all cases.
34.5
34
33.5
33
32.5
32
31.5
Temperature [°C]
31
30.5
30
29.5
29
28.5
28
27.5
27
26.5
26
25.5
25
24.5
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85
TØ_PnRG X (1...87)
Fig. 6. Comparison of temperature between left (LnRG=87) and right (PnRG=87) hands in
reference database
Table 1. Comparison of both databases between right (PnRG=87; PnCTS=8) and left
(LnRG=87; LnCTS=8) hands
TØ_PnRG TØ_LnRG TØ_PnCTS TØ_LnCTS
Avg± SD 29.2±2.3 29.1±2.3 34.2±1.5 34.0±1.5
Temp. interval <24.1; 33.4> <24.2; 34.0> <31.5; 36.7> <31.8; 36.5>
In database RG (RGn=174) the biggest difference in average temperature was in

healthy volunteer Num.4 (RGX4).The average temperature of right hand was war-
mer than left hand about 1.9°C (Fig.7). Right hands were warmer than left hands
in 43 cases (49.4%; Fig.7). Right hands were colder in 41 people, which is 47.1%
(Fig.9). In 3 cases it was the same temperature, which means zero difference in
hands from one person (3.5%; Fig.8).
Fig. 7. Healthy volunteer Num.4 (RGX4); Fig. 8. Healthy volunteer Num.11 (RGX11); zero
right hand is warmer than left hand. difference in hands from one person
Fig. 9. Healthy volunteer Num.87 (RGX87); left hand is warmer than right hand
We have not found a significant temperature difference between the left and
right hand from one person in reference group (RGn=174) and CTSn too since be-
ginning of the study to date.
10 Results Obtained from Second Hypothesis

For the purpose of this study was established the line of the Index of Median
Nerve. We have always identified five points in the line of the median nerve (on
dorsal side of hand), on which we have measured temperatures since beginning of
the study. For the dorsal side of hands, there have been the center points of carpals
(T1), the center points of metacarpals (T2) and the fingertips of the middle finger
from proximal phalanges (T3), the intermediate phalanges (T4) to the distal
phalanges (T5), (see Fig.10).
Fig. 10. Location of points Ti
Figure 11 shows the course of temperature on the healthy hands. At no point

along RGTi we found no significant difference in temperature between the left and
right hands. The highest temperature was measured in area T2, which is the center
point of metacarpals. The average temperature was 31.4°C for right and 31.2°C
for left hands. The lowest temperature was measured in points T5, which is the
area of the distal phalanges.
Infrared Thermography in th
he Evaluation of Carpal Tunnel Syndrome 3665
31.5
31.0
30.5
30.0
29.5
29.0
28.5
28.0
T1agv T2agv T3agv T4agv T5agv
P_RG 31.2 31.4 29.8 28.8 28.4
L_RG 31.0 31.2 29.8 28.7 28.4
Fig. 11. Comparison in Ti points between left and right hands in RG database (PnRG and LnRGG)
32.4
32.3
32.2
32.1
32.0
31.9
31.8
31.7
31.6
T1agv T2agv T3agv T4agv T5agv
P_CTS 31.6 32.3 32.0 32.2 32.1
L_CTS 31.7 32.3 32.0 32.2 32.1
Fig. 12. Comparison in Ti points

p between left and right hands in CTS database (PnCTS annd
LnCTS)
Figure 12 shows the course

c of temperature on the pathological hands. At nno
point along CTSTi we foun nd no significant difference in temperature between thhe
left and right hands. The course of temperature on the hands affected by CTS haas
been dissimilar like in heaalthy hands. The biggest difference has been in the distrri-
bution of temperature in areas
a of T1 and T5.
366 M. Tkáčová et aal.
32.5
32
31.5
31
30.5
30
29.5
29
28.5
28
T1avg T2avg T3avg T4avg T5avg
CTS 31.6 32.3 32.0 32.2 32.1
RG 31.1 31.3 29.8 28.8 28.4
Fig. 13. Comparison of Tavg

g in five defined points Ti on the line of median nerve betweeen
RG and CTS databases
The biggest differencee between RG (RGn=174) and CTS (CTSn=16) databasees

was taken inareaT5. The graph
g (Fig. 13) illustrates the relationship between poinnts
of line of the median neerve from RGT (RGn=174) and CTST (CTSn=16).There waas
found 3.7°C difference beetween database CTST5 and RGT5.
Table 2. Comparison of av verage temperatures obtained from tracing hands and in fivve
defined points in line of med
dian nerve between RG and CTS databases
RG RGTØ RGT1 RGT2 RGT3 RGT4 RGT5
Tavg 29.1 31.1 31.3 29.8 28.8 28.55

±SD ±2.3 ±1,6 ±1.9 ±2.4 ±3.347 ±4.22
MODE 27.6 32.3 32.0 30.1 32.6 33.11
CTS CTSTØ CTST1 CTST2 CTST3 CTST4 CTST5
Tavg 31.4 31.6 32.3 32.0 32.2 32.11

±1.6 ±1.3 ±1.5 ±1.8 ±2.1 ±1.88
MODE 31.4 32.7 33.4 30.8 34.8 33.00
Difference 2.7 0.6 1.0 2.2 3.4 3.77
Where RGTØ/ CTSTØ is average

a temperature from polygons of tracing hand;RGT1/
CTST1 is average temperatture from the center points of carpals (T1); RGT2/ CTST2is
average temperature fromm the center points of metacarpals (T2); RGT3/ CTST3is aveer-
age temperature from the fingertips of the middle finger from proximal phalangees
(T3); RGT4/ CTST4 is averaage temperature from the intermediate phalanges (T44);
RGT5/ CTST5is average temp
perature from the distal phalanges (T5).
36
35
34
33
32
31
30
29
28
27
26
25
24
23
22
21
20
19
1
8
15
22
29
36
43
50
57
64
71
78
85
92
99
106
113
120
127
134
141
148
155
162
169
T5_RG T1_RG
Fig. 14. Comparison of temperatures from T1 and T5 in RGn (RGn=174)

The cumulative averag ges of temperature distribution have been obtained fro m
reference and pathologicaal thermograms on the dorsal side of the line of the mee-
dian nerve for two years. The cumulative averages show a big difference betweeen
healthy and CTS hands. To T be precise, there was a 4°C difference from the aveer-
age temperature which are a obtained from the RGT5 and CTST5. This difference haas
been the main reason for establishing the equation for the Index of Median Nervve
(MI; Fig.14, 15, 16, 17):
MI = T − T (1)
D 1 5
where DMI is the Index of o Median Nerve, which is calculated on dorsal side oof
hands. T1 is the area on the center point of carpals on dorsal side of hand. T5 is
area of the distal phalangees on dorsal side of hand.
8.5
MI
6.5
4.5
[°C]
RG(T1-T5)
2.5
0.5
1
7
13
19
25
31
37
43
3
49
55
61
67
73
79
85
91
97
103
109
115
121
127
133
139
145
151
157
163
169
175
-1.5
-3.5
DRG; nRG=174
Fig. 15. Index of the Median

n Nerve (DMIRG) from RGn (RGn=174)
From graph (Figure 14 4) DRG means dorsal side of healthy hands. A negativve
difference RG(T1-T5) (DMII≤0) from RGn (RGn=174) has been recently detected in 555
healthy hands (which acccount for 31,6% of total normal thermograms from thhe
dorsal side of RGn), (see Fig.
F 14,15).
A negative difference DMICTS (DMI≤0) from CTSn (DnCTS=16, dorsal side of handds
from CTS patients) was detected
d in 12 cases (that is 75,0% from CTSn). (See Fig.
16, Fig.17). In the patho ology database was carpus (T1) warmer than the disttal
portion of fingers only upp to 2°C and only in 4 cases.
Infrared Thermography in th
he Evaluation of Carpal Tunnel Syndrome 3669
35
34
33
32
31
30
29
28
27
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
T1_CTS T5_CTS
Fig. 16. Comparison of temp

peratures fromT1 and T5 in CTSn (CTSn=16)
MI
2
1
T1-T5 [°C]
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
-1
-2
-3
nCTS; nCTS=16
Fig. 17. Comparison in temp

peratures from T1 and T5 in CTSn (CTSn=16)
Table 3. Index of the Median

n Nerve (DMI)
T Index of Median Nerve (MI)

The
[°C] ± SD DMI≤0 [%]
D MI = T −T
D1 D5
MICTS -0,474 ± 1,365 75%
MIRG 2,644 ± 3,229 31,6%
The data have recently showed that the skin temperatures of median nerve dis-
tribution area on dorsal hands varied significantly between CTS and the control
group.
11 Estimation of Index Test Results

The diagnostic test of obtained temperatures from line of median nerve perfor-
mance includes consideration of validity and reliability of the test (infrared ther-
mography), (See Table 4).
Table 4. Assessment of validity of a diagnostic test
Reference test [EMG]

Positive Negative
Diagnostic test Positive a = 12 b = 55
[Thermography] Negative c=4 d = 119
Total sample size n1 = 16 n2=174
Where:
a = True positive: Sick volunteers correctly diagnosed as sick
b = False positive: Healthy people incorrectly identified as sick
d = True negative: Healthy people correctly identified as healthy
c = False negative: Sick people incorrectly identified as healthy.
Sensitivity relates to the test's ability to identify positive results. Sensitivity (SE) is
defined as the probability that the test will be positive in patients. Calculation of
the sensitivity (SE)[15]:
a a 12
SE = = = = 0,750 (2)
a + c n1 16
D
Confidence interval for sensitivity: p ± 1,96 (pq/n1)1/2, where:

a a 12
p= = = = 0,750 (3)
a + c n1 16
c c 4
q= = = = 0,250 (4)
a + c n1 16
Therefore the confidence interval for sensitivity: 0,750±0,202.
The specificity is defined as the probability that the test will be negative in
healthy subjects. The calculation to obtain the values of specificity (SP)[15]:
d d 119
SP = = = = 0,684 (5)
b + d n2 174
D
The confidence interval for specificity: p ± 1,96 (pq/n2)1/2, where:

d d 119
p= = = = 0,684 (6)
b + d n2 174
b b 55
q= = = = 0,316 (7)
b + d n2 174
Therefore confidence interval for specificity: 0,684±0,069
Based on Tab.4 above, measures of validity and 95% confidence intervals were
calculated confidence interval for sensitivity 0.625±0.237 and confidence interval for
specificity 0.815±0.060.
12 Conclusion
Thus, the temperature measurements could bring with additional knowledge on
the pathophysiology of CTS and give an opportunity for the follow up of the
course of the condition before and after CTR operation [4]. The average tempera-
tures, which were obtained from tracing hands and polygons of temperature mea-
surement, were compared between them RGn and CTSn; PnRG and LnRG; PnCTS and
LnCTS). We haven't been found a significant temperature difference between the left
and right hand from one person in reference database (RGn and CTSn too since be-
ginning of the study to date.
Sensitivity and specificity of thermographic diagnostic tests were realized on
five anatomically defined points (Ti) on dorsal side of 174 healthy hands
(RGn=174) and of 16 pathological hands (CTSn=16). An EMG diagnostic method
has been applied as a referential test since 1 month before infrared examination.
Sensitivity and specificity were calculated using standard formulas and argument
DMI = (T1-T5) from the database of dorsal thermograms.
The calculated sensitivity of diagnostic test (0.750) with the confidence interval
0.750±0.202 and specificity of diagnostic test (0.684) with the confidence interval
0.684±0.069 show that thermography could be a promising method in the diagnostic
process of CTS. The advantages are in the access to investigation of temperature
and diagnostic symptoms. More statistically significant measurements are required
to confirm obtained results. Therefore we will continue in this study. Results will
be confirmed by further thermal screening of a statistically significant group of
patients.
Acknowledgments. This contribution is the result of the project implementation: Creation

and promotion of Technologies in diagnostics of components and junctions with computed
tomography (lTMS:26220220038) supported by the Research & Development Operational
Programme funded by the ERDF and project Research of New Diagnostic Methods in
Invasive Implantology, MŠSR-3625/2010-11, Stimuls for Reasearch and development of
Ministry of Education, Science, Research and Sport of the Slovak Republic).
References
[1] Jesensek Papez, B., Palfy, M., Mertik, M., Turk, Z.: Infrared thermography based on
artificial intelligence as a screening method for carpal tunnel syndrome diagnosis. J.
Int. Med. Res. 37(3), 779–790 (2009) PMID: 19589261
[2] Niehof, S.P., et al.: Thermography imaging during static and controlled thermoregula-
tion in complex regional pain syndrome type 1: diagnostic value and involvement of
the central sympathetic system. BioMedical Engineering Online 5, 30 (2006),
doi:10.1186/1475-925X-5-30
[3] Jesensek Papez, B., Palfy, M., Turk, Z.: Infrared thermography based on artificial in-
telligence for carpal tunnel syndrome diagnosis. The Journal of International Medical
Research J. Int. Med. Res. 36(6), 1363–1370 (2008)
[4] Ming, Z., Siivola, J., Pietikainen, S., Närhi, M., Hänninen, O.: Postoperative relieve
of abnormal vasoregulation in carpal tunnel syndrome. Clin. Neurol. Neuro-
surg. 109(5), 413–417 (2007) PMID: 17400369
[5] Ring, E., Ammer, K.: The technique of infrared imaging in medicine. Thermology In-
ternational 10, 7–14 (2000)
[6] Jones, C., Ring, E., Plassmann, P., Ammer, K., Wiecek, B.: Standardisation of infra-
red imaging: a reference atlas for clinical thermography—initial results. Thermology
International 15, 157–158 (2005)
[7] Gold, J.E., Cherniack, M., Hanlon, A., Dennerlein, J.T., Dropkin, J.: Skin temperature
in the dorsal hand of office workers and severity of upper extremity musculoskeletal
disorders. European Journal of Applied Physiology 93(1-2), 245–251 (2004)
[8] http://arthritis.about.com/od/carpal/ss/
carpaltunnelqa_6.htm
[9] http://www.infraredinstitute.com/basic.thermography.html
[10] Jakubowska, T., Wiecek, B., Wysocki, M., Drews- Peszynski, C.: Thermal signatures
for breast cancer screening comparative study. In: Proc. IEEE EMBEC, Cancun,
pp. 1117–1120 (2003)
[11] Szu, H., Kopriva, I., Hoekstra, P., Diakides, N., Diakides, M., Buss, J., Lupo, J.: Ear-
ly Tumor Detection by Multiple Infrared Unsupervised Neural Nets fusion. In: Proc.
IEEE EMBC, Cancun, pp. 1133–1136 (2003)
[12] Qi, H., Kuruganti, P.T., Liu, Z.: Early detection of breast cancer using thermal texture
maps. In: IEEE Int. Symp. on Biomedical Imaging, Washington DC, pp. 309–312
(2002)
[13] Stevens, J.C., Witt, J.C., Smith, B.E., Weaver, A.L.: The frequency of carpal tunnel
syndrome in computer users at a medical facility. Neurology 56, 1568–1570 (2001)
[14] Ammer, K.: Thermology 2009- a computer- assisted literature survey. Thermology
International 20(1), 5–10 (2010)
[15] Zvarová, J.: Základy statistiky pro biomedicínské obory I.; Biomedicínska statistika.
Univerzita Karlova v Prahe, Nakladateľstvo Karolinum (2007) ISBN: 978-80-7184-
786-1
[16] Deniz, F.E., Oksüz, E., Sarikaya, B., Kurt, S., Erkorkmaz, U., Ulusoy, H., Arslan, S.:
Comparison of the diagnostic utility of Electromyography, Ultrasonography, Com-
puted Tomography, Magnetic Resonance Imaging in idiopathic Carpal Tunnel Syn-
drome determined by clinical findings. Neurosurgery (September 12, 2011) [Epub.
ahead of print] Pub. Med. PMID: 21869718
[17] Visser, L.H., Smidt, M.H., Lee, M.L.: High-resolution sonography versus EMG in the
diagnosis of carpal tunnel syndrome. Journal of Neurology, Neurosurgery and Psy-
chiatry [J. Neurol. Neurosurg. Psychiatry] 79(1), 63–67 (2008); Date of Electronic
Publication (April 30, 2007) England NLM ID: 2985191R Publication Model: Print-
Electronic Cited Medium: Internet ISSN: 1468-330X (Electronic) Linking ISSN:
00223050 NLM ISO Abbreviation: J. Neurol. Neurosurg. Psychiatr. Subsets:
MEDLINE
[18] Emad, M.R., Najafi, S.H., Sepehrian, M.H.: The effect of provocative tests on elec-
trodiagnosis criteria in clinical carpal tunnel syndrome. Electromyogr. Clin. Neuro-
physiol. 50(6), 265–268 (2010) PMID: 21061772 [Pub. Med - indexed for
MEDLINE]
[19] Miland, A.O., Mercer, J.B.: Effect of a short period of abstinence from smoking on
rewarming patterns of the hands following local cooling. Eur. J. Appl. Physiol. 98,
161–168 (2006), doi:10.1007/s00421-006-0261-2; (accepted: June 9, 2006) Published
online: July 28, 2006, © Springer-Verlag (2006)
[20] Thomsen, J.F., Gerr, F., Atroshi, I.: Carpal tunnel syndrome and the use of computer
mouse and keyboard: A systematic review. BMC Musculoskelet Disord 6(9), 134
(2008) Review. PMID:18838001 [Pub.Med - indexed for MEDLINE]
Determination of Conditions and Positions
for Obtaining Thermographic Atlas
Mária Tkáčová1, Radovan Hudák2, Jozef Živčák2, Patrícia Čekanová2,

and Štefan Medvecký3
1
CEIT-KE, s.r.o., Košice, Slovakia
maria.tkacova@ceit-ke.sk
2
Technical University of Košice, Slovakia
{radovan.hudak,jozef.živčák}@tuke.sk
3
University of Žilina, Slovakia
Stefan.medvecky@fstroj.uniza.sk
Abstract. The study is concerned with the basic conditions to create a thermo-
graphic atlas of the human body. The main goal of research had been measure-
ment and comparison in temperature distribution of healthy human bodies which
were obtained by infrared thermography for one year. It was used Infrared Imager
with detector type 320x240 Focal Plane Array. For the purposes of research, we
created a database of 570 healthy thermograms, which were taken from 30 volun-
teers (ni=30) in nineteen positions (j).The database of atlas consisted of 15 healthy
men (15 male, Mn=15) and 15 healthy women (15 female, Fn=15). The average age
of volunteers was 22,8 years. The average BMI was 22,7 (normal weight).By
thermography, we have obtained temperature maps on bodies. In the process of
analyzing the thermograms we obtained qualitative and quantitative data on areas
of interest. The database can be helpful for comparison in the diagnostics of many
neurovascular and musculoskeletal diseases or injuries.
1 Introduction
The goal of presented paper is to create a database of normal thermograms of
young healthy people. The database could be helpful in the diagnosis of various
diseases such example inoncology (skin, breast cancer, etc.), neurovascular dis-
eases, dermatologic diseases, studies of inflammatory responses, pain (manage-
ment / localization / control), Raynaud phenomenon, sleep research, painrelated
thermal dysfunctions, Anesthesiology, Acupuncture and Complementary Medi-
cine, Physical Medicine and Rehabilitation, cardiovascular diseases, atherosclerot-
ic plague, dentristy, etc. [1, 2, 3, 4].
Thermography appears to be a useful examination method and not only in cases
for which other laboratory or imaging methods fail to provide an outcome. Ther-
mography is helpful in differentiating the inflammatory etiology of low back pain
from painful syndromes of a reflex origin induced either through the mechanism
of nociceptive sympathetic efferent or autonomic reaction. This method is charac-
terized by high sensitivity for both vertebrogenic and viscero-vertebrogenic pain

syndromes. Due to its low specificity, especially in cases of structural lower back
damage, it is not possible to make a final diagnosis solely based on this method.
Its significance lies in moving in the direction of accurate diagnostics and in as-
sisting in the choice of proper treatment [12].
Temperature is the basic measurement of a physical property for determination
of diagnosis in the medicine. From electromagnetic theory it is a form of infrared
energy being emitted from the first molecular surface of a body (skin). Infrared
imaging is the detection and conversion of energy from a section of the infrared
spectrum, into the visible spectrum. Surface energy levels are affected by the envi-
ronment, operational conditions, heat transfer processes of a human body and skin
characteristics. Because of thermograms interpretation is really important to know
physiological characteristics and high skills with thermography measurement and
picture analysis [5, 6, 14].
Medical thermography could be a great tool for diagnostics of surface tempera-
ture of the human body. Advantagesof these methods are non-invasive, non-
contact, safely, no-radiation and painless image technique. This assumption is
based on the fact that many diseases are accompanied by the change in skin tem-
perature, pain, or in asymmetry of the emissivitywhich could be visible by the use
of infrared thermography [2, 5, 6, 7, 10, 11, 12, 13].
For each of diagnostics it is important to have reference thermograms of the area
of interest. On the basis of this knowledge, also based on measurements, was set up
a method of processing thermograms, which was described in this study. For the
purpose of our study was created a database from 19 views of human body.
It was captured: TBA (Total Body – Anterior view), TBA-A (Total Body Ante-
rior - Abduction view), TBD (Total Body Dorsal view), TBD-A (Total Body
Dorsal - Abduction view), TBS-L/R (Total Body Sagittal – left/ right view), ToA
(Torso Anterior view), ToD (Torso Dorsal view), ULA-R/L (Upper Limb Anterior
– Right/Left view), ULP-R/L (Upper Limb Posterior – Right/Left View) PH-R/L
(Palmar side of Hand – Right/Left), DH-R/L (Dorsal side of Hand – Right/Left),
FC (Face) and F-R/L (Foot- Right/Left view).
2 Basic Concepts for Atlas of Thermography
2.1 Volunteer Recruitment

To conduct representative study with regard to anthropological features, data of
the local population are needed to select appropriate subjects for a representative
sample. Subjects between 18 and 70 years should be recruited, and grouped into
age groups, which are 18-30 years, 31-40 years, 41-50 years, 51-60 years and 61-
70 years. Gender distribution should be equal (half male and half female). Each
age group is subdivided into 3 sub-groups according to the body mass index of
volunteers: "underweight", "normal", "overweight" & "obese") according to the
guidelines issued by US National Heart, Lung, and Blood Institute (NHLBI), in
cooperation with the National Institute of Diabetes and Digestive and Kidney
Diseases (NIDDK). BMI calculator [10].
Determination of Conditions and Positions for Obtaining Thermographic Atlas 377
3 Physiological Temperature Distribution

Physiological temperature distribution means temperature distribution which is
measured from some position and locality of healthy human body [5, 6].
Heat transfer problems are important in various disciplines including biomedi-
cal sciences. Heat transfer in biological involves metabolic heat generation, con-
duction, convection, radiation and blood perfusion in human tissues [8].
All physiological functions generate heat by means of metabolic reactions in
biological systems. The balance between the heat generation and loss from the
body to the environment is very important to maintain body core temperature i.e.
37°C. Any physiological abnormality will disturb the homeostatic conditions for
the temperature. Therefore the study of heat transfer changes under normal and
abnormal conditions will be useful for various clinical conditions [8].
Wound healing process will be better understood not only by detecting the pa-
thology but also by considering the correlation between tissue thermal information
and human physiological needs [8].
Biothermology or Bio- fluid flow and heat transfer is an important and develop-
ing subdivision of bioengineering [9].
Bioheat transfer and biomass transfer can be seen as a subdivision of bioengi-
neering. The three-fold objectives of heat transfer, viz. insulation, enhancement
and control of temperature, when applied in the context of biological systems,
offer insights into many biological processes. Examples of bioheat transfer
processes specific to humans include:
• Thermoregulation; metabolic heat generation, evaporation, convection and

radiation to achieve steady state;
• Effect of increased Metabolic Heat Generation; temperature rise during exer-
cise;
• Bioheat transfer in muscles and tissues accompanied with effects due to
blood flow (perfusion);
• Burning; skin burning as transient heat transfer process;
• Fever and Hypothermia;
• Thermal Comfort; Convection, conduction heat transfer through clothing,
optimum temperature, humidity, energy transfer in artificial fittings like
contact lens [9].
4 Atlas of Normal Thermograms

Atlas of normal thermograms is a database of the thermograms, which were taken
from physiological healthy human volunteers in 19 views by the use of infrared
thermography [4, 6].
The eight views of human body were captured in: TBA (Total Body – Anterior
view), TBD (Total Body – Dorsal view), TBS (Total Body – Sagittal view), ToA
(Torso- Anterior view), ToD (Torso- Dorsal view), DH (Dorsal side of Hand), PH
(Palmar side of Hand), F (Foot) and FT (Facial Title).
For each view was used Standard Anatomical Position. The international stan-
dard anatomical position is the position that provides a reference point for describ-
ing the structures of a human body. In this position, the body is standing erect with
good posture and the face is looking directly forward. The feet are about six inch-
es apart, flat on the floor and the toes pointing forward. The arms are down at the
sides with the palms turned forward and thumbs pointing away from the body.
Anterior and ventral both mean toward the front of the body, while posterior
and dorsal mean the back of the body.
This position is the standard reference point in which all positions, movements,
and planes are described. The anatomical position allows us to describe the posi-
tion of structures in relation to their surroundings.
Each of normal thermograms was obtained in detail view of interesting area
and from the same angle.
5 Medical Thermography
Institute of Infrared Thermography defines thermography as the technique that
uses an infrared imaging and measurement camera to “see“ and “measure” invisi-
ble infrared energy being emitted from an object [9].
The technique that uses an infrared imaging and measurement camera to “see“
and “measure” invisible infrared energy being emitted from an object. It is a tool
for the study of surface temperature distribution of living organisms [4].
Infrared radiation is a form of energy and is a part of the electromagnetic spec-
trum. Infrared is similar in many ways to the visible spectrum, yet it is not usually
visible to human eyes.
Infrared camera main components are optics, electronics, detector, display, sto-
rage device, and batteries. The optics of an IR camera are similar to optics of a
visual camera. The function is to gather and focus incoming energy onto the detec-
tor. The system acts as an amplifier of infrared energy onto detector. This enables
the detector to “record” low levels of infrared radiation. The optical system is an
extremely important component of the infrared camera.
Electronics provides the circuitry that will digitize, process, and format the sig-
nal to a visual image. The microprocessor also performs ongoing calibration as the
internal camera temperatures change.
Advantages of infrared thermography are that it is a contactless, noninvasive,
painless and no-radiation image technique. Disadvantage of medical thermogra-
phy is processing of thermograms with software (for the thermocamera Fluke
Ti55/20, is it the software SmartView), because is missing the ability to save sud-
denly more polygons to compare, or return back in job, or copy ongoing job [8].
6 Methodology of Measurement
Methodology of this study consisted of conditions for Object, Infrared Thermal
Imaging Camera, Room and Processing for all year. Skin temperature of the human
body from our database (n=570) was measured with an infrared camera (Infrared
Thermal Imaging Camera Imager, Fluke Ti55/20, Fluke, USA). This thermographic
camera generates a matrix (representing image points) of temperature values. They

feature 320 x 240 (76 800 pixels) detectors with industry leading thermal sensitivity
(≤0.05°C; 50mK NETD) for high resolution. The camera works in the spectral range
from 8 to 14 µm (human body infrared radiation is the highest in the spectral range
around9.66µm) and the calibrated temperature range from -20°C to 100°C. Data
were obtained through high-speed (60Hz) analysis [8].
Emissivity of the skin was set up to 0.98 in the camera, the ambient tempera-
ture was measured with an infrared (laser) thermometer (Pyrometer Testo 810)
and a contact multimeter (Almemo Multimeter 2690, Ahlborn, Germany) was
used for verification. The camera was calibrated using the system's internal cali-
bration process before each recording. All thermograms (n=240) were processed
using special software (SmartView 2.1, FLUKE, USA).
For our study we used some accessories for control measurement of tempera-
ture, for example for measurement axillar temperature (Thermoval Classic, Hart-
mann, Germany), and for ambient temperature (PyrometerTesto 810).
7 Conditions of Measurement
Measurements were carried out under the same conditions, always in the same
room with the ambient temperature 22.4 °C (±1.7°C). The blinds in the room were
drawn in order to eliminate the impact of solar radiation and the room was air-
conditioned to keep the same temperature during each measurement [1, 7].
All volunteers had to be acclimated in this room for 20 minutes before
measurement and naked during thermographic diagnosis. Volunteers prior to the
examination should restrict food, liquids, cigarettes, and cosmetics, physical or
psychological stress.
Conditions for Infrared Imager position, it should to be used stable tripod for
constant height, distance and angle between thermocamera– object [6].
All thermograms were processed in the same way in software SmartView 2.1
(software for Infrared Thermal Imaging Camera Imager, Fluke) [7].
8 Subjects of Measurement
Database of atlas consists of 30 physiological healthy volunteers (ni=30; 15 male,
Mni=15; 15 female, Fni=15). Average BMI from the male database was 22.8 (nor-
mal weight), in the female database was 22.2 (normal weight). Average age of the
volunteers was 22,8 years. Nineteen views (j) of human body were realized on 570
thermograms (ΣjA = Σ1…19A=570).The main features of the data collection were
quantitatively describing temperature distribution of healthy human body in age
from 16 to 28 years.
8.1 Body Mass Index

Body Mass Index (BMI) is a measure of body fat based on height and weight that
applies to adult men and women [6].
mass[kg]
B =
[ ]
BMI
(height)2 m2
(1)
Table 1. Categorization of BMI

B
BMI
0 – 19.9 Underweight
20 – 24.9
9 Normal weight
25 – 29.9
9 Overweight
30 – 39.9
9 Obesity
> 40 Extreme obesity
Table 2. Average date of Vo

olunteers (ni=30)
Sex AgeAvvg[year] WeightAvg [kg] HeightAvg [cm] BMIAvg[kg/m2]
Mele (Mn=15) 22,5 75,2±15.7 179,0±7.8 22,8±2,6
Females (Fn=15) 22,3 69,1±7.4 173,4±5.7 22,2±2.1
Σ(M+W) 22,8 74,1±18.1 178,4±11.1 22,7±2.8
29.0 Male_BMI Female_BMI
27.0
25.0
BMI
23.0
21.0
19.0
17.0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
V
Volunteers (15 male and 15 female)
Fig. 1. Comparison of male (Mn=15) and female (Fn=15) BMI (ni=30)

Determination of Conditionss and Positions for Obtaining Thermographic Atlas 3881
9 Positions of Meassurement
1. Position - TBA (Total Body

B – Anterior view)
Description
The international standard d anatomical position defined as standing erect, with thhe
palms and feet facing forwward (Figure 2). Legs together and arms by their sides, wiith
their head, eyes, toes and palms
p of the hands facing forward. This position is the stann-
dard reference point in whiich all positions, movements, and planes are described.
2. Position - TBD (Total Body

B Posterior view)
The international standard
d anatomical position from posterior view (Fig. 3).
Fig. 2. Volunteer Num.25 in

n the first posi- Fig. 3. Volunteer Num.25 in the second
tion of view (25X1) position of view (25X2)
Both of thermograms (Figure 2, 3) are infrared images of the highest contraast

range in the narrow tempperature range <28,1; 34,9> [°C]. Thermogram (Fig. 22)
shows anatomical position
n in anterior view and Fig.3 in dorsal view.
3. Position– TBA-A (Tota al Body Anterior - Abduction view)

Upper limb in abductio on realized from standard anatomical position from m
anterior view (Figure 4). Abduction, in functional anatomy, is a movement whicch
draws a limb away from the median sagittal plane of the body. It is thus opposeed
to adduction.
4. Position– TBP-A (Totaal Body Posterior - Abduction view)
Upper limb in abduction realized
r from standard anatomical position from posterioor
view (Figure 5).
Fig. 4. Volunteer Num.255 in the third Fig. 5. Volunteer Num.25 in the fourth possi-
position of view (25X3) tion of view (25X4)
5. Position – TBS-R (Tota al Body Sagittal – Right view)

Upper limb in flexion reealized from standard anatomical position from sagitttal
plane, right view (Figure 6). Flexion occurs in the sagittal plane, when the angle oof
a joint decreased.
6. Position - TBSA (Totall Body Sagittal – Left view)
Upper limb in flexion reealized from standard anatomical position from sagitttal
plane, left view (Figure 7)).
Fig. 6. The volunteer Num

m.25standing in Fig. 7. The volunteer Num.25 standing in thhe
fifth position (25X5) sixth position (25X6)
7. Position - ToA (Torso Anterior

A view), (Figure 8)
8. Position - ToA (Torso Posterior
P view), (Figure 9)
Fig. 8. The volunteer Num.2 25 standing in Fig. 9. The volunteer Num.25 standing in thhe
the seventh position (25X7) eighth position (25X8)
9. Position–ULA-R (Uppeer Limb Anterior – RightView), (Figure 10)

10. Position–ULA-L (Upp
per Limb Anterior – LeftView), (Figure 11)
Fig. 10 The volunteer Num m.25 standing Fig. 11 The volunteer Num.25 standing in thhe
in the ninth position (25X9) tenth position (25X10)
11. Position–ULP-L (Upp

per Limb Posterior – RightView), (Figure 12)
12. Position–ULP-L (Upp
per Limb Posterior – LeftView), (Figure 13)
Fig. 12. The volunteer Num m.25 standing Fig. 13. The volunteer Num.25 standing in
in the eleventh position (25X11
1 ) thetwelfth position (25X12)
13. Position - DH-R (Dorrsal surface of Right Hand), (Figure 14)
14. Position - DH-L (Dorrsal surface of LeftHand), (Figure 15)
Fig. 14. The volunteer Num m.25 standing in the Fig. 15. The volunteer Num.25 standinng
thirteenth position (25X13) in the fourteenth position (25X14)
Upper limb in pronatio on (Fig. 14, Fig. 15). Area of interest was dorsal side oof
hand. Pronation is a rotaational movement of the forearm at the radioulnar joinnt.
The average temperaturess in thermograms were obtained from tracing hand. It waas
the way for creating polyggons from interesting areas of temperature measurementt.
Position of dorsal sidee of hands (Fig. 14, Fig. 15) is important position for ddi-
agnostic of neuropathic diseases
d such as carpal tunnel syndrome (CTS). It conn-
firmed the hypothesis thatt the dorsal side of the hand is of greater importance thaan
the palmar side when diag gnosing CTS thermographically [14].
15. Position - PH-R (Palm

mar surface of Right Hand), (Figure 16)
16. Position - PH-L (Palm

mar surface of LeftHand), (Figure 17)
Upper limb in supination
n (Fig. 15, Fig. 16). Area of interest was palmar side oof
hand.
Fig. 16. The volunteer Num m.25 standing in Fig. 17. The volunteer Num.25 standing in
the fifteenth position (25X15) the sixteenth position (25X16)
17. Position - Face (Faciial Zone), (Figure 18)
Fig. 18. The volunteer Num..25 standing in the seventeenth position (25X17)
Thermogram of face (F Fig. 18). The average temperatures in thermograms werre

obtained from tracing facee.
18. Position - F-R (Rightt Foot), (Figure 19)

19. Position - F-L (LeftF
Foot), (Figure 20)
Fig. 19. The volunteer Num..25 standing in Fig. 20. The volunteer Num.25 standing in thhe
the eighteenth position (25X18
1 ) nineteenth position (25X19)
10 Results
The result of study is dattabase of normal thermograms and review of physiologgi-
cal temperature distributiion. All values were obtained and recalculated to onne
decimal place.
Numerical order was attributed
a to Volunteers (x1,x2,...,xn) in the order of mea-
surement. In the initial taable were recorded figures of temperatures from thermoo-
grams in order in which they
t were obtained (raw data). Total number of values iin
the primary table is desiggnated as n. The figures in the primary table were labeleed
with the symbol xj.
Figures were arranged by size from smallest to largest, with the same values arre
entered as many times ho ow many times they occurred (ordered data). So variatioon
range has been:
x(1)≤x(2)≤…≤x(n) (22)
Figures in the variation range were labeled with the symbol x(j). Many figurees
were repeated, so it was compiled
c the variation table.
Where:
j(k) –class number,
n(M/F) – frequency of xj,
N(M/F) – cumulative numbeers of frequencies,
f(M/F)–relative frequency
F(M/F)- cumulative relativee frequency
Table 3. The variation table of volunteers
class
j(k) n(M) n(F) N(M) N(F) f(M) f(F) F(M) F(F)
interval
1,0 27-27,9 0 1 0 1 0% 7% 0,0 0,1

2,0 28-28,9 0 2 0 3 0% 13% 0,0 0,2
3,0 29-29,9 2 1 2 4 13% 7% 0,1 0,3
4,0 30-30,9 7 3 9 7 47% 20% 0,6 0,5
5,0 31-31,9 6 5 15 12 40% 33% 1,0 0,8
6,0 32-32,9 0 3 15 15 0% 20% 1,0 1,0
50% f(M) f(F) 47%

45% 40%
40% 33%
35%
30%
25% 20% 20%
20% 13% 13%
15%
10% 7% 7%
5% 0% 0% 0%
0%
27-27,9 28-28,9 29-29,9 30-30,9 31-31,9 32-32,9
Class Interval of Temperatures
Fig. 21. Comparison of relative frequency obtained from average temperature (Tavg) be-
tween male and female volunteers; views in frontal plane from ventral side
From the graph (Figure 21) the range for the first step (Rv=5.3°C) is the sim-
plest measure of variation. It is simply the highest value (32.8°C) minus the lowest
value (27,5°C). Male (Mn1=15) and female (Mn2=15) databases were divided into
class of intervals (k, Mk=Fk=6). Temperature distribution in male database was in
interval from 29°C to 31.7°C and the highest multiplicity (Mk4=47%) in class in-
terval <30; 30.9>. Temperature distribution in female database was in interval
from 27°C to 32.8°C and the highest multiplicity (Fk4=33%) in class interval <31;
31.9>. Accuracy of the used thermal imager (Fluke Ti55/20) is ±2°C.
The average temperatures in databases were obtained from tracing bodies and
partial segments (positions 1-19). It was the way for creating polygons from inter-
esting areas of temperature measurement.
1_T
Tmax 1_Tavg Ambient Temp.
36.0
34.0
32.0
Temperature [°C]
30.0
28.0
26.0
24.0
22.0
20.0
18.0
M M M M M M M M M M M M M M M F F F F F F F F F F F F F F F
1 2 3 4 5 6 7 8 9Serial
10 11 12 13of14
Number 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Volunteer
Fig. 22. Comparison Tempeerature (Tmax, Tavg, Tamb) between male and female volun-
teers; view in frontal plane from
fr ventral side
a) b) c)
Fig. 23. Temperature distrib bution of human body contour; view in frontal plane from a)
anterior side; b) and c) posteerior side
Pictures (figure 23) show distribution of temperature from first and second poo-
sition. First thermogram male
m volunteer (27X1) standing erect, with the palms annd
feet facing forward (Figurre 23a). For analyzing the temperature for medical applli-
cation was usually used Palette
P of High Contrast (Fig. 23). Processing of thermoo-
gram begins by deduction n of Background temperatures and identifying key areaas
of interest. Figures 23b) and
a c) show thermograms of volunteer Num.25 (25x2) iin
high contrast with narrow w temperature interval <28.1; 34.9>.
Table 4 Summary table of maximal and average temperatures from all 19 views
positions Tmax.±SD dispersionTmax modeTmax Tavg±SD dispersionTavg modeTavg

1 35,2±0,8 0,6 35,3 30,7±1,2 1,4 30,4
2 34,5±0,9 0,9 34,6 30,0±1,2 1,4 30,3
3 35,2±0,8 0,7 35,1 30,7±1,1 1,1 30,7
4 35,1±0,8 0,7 30,2 30,2±1,2 1,5 24,2
5 35,1±0,8 0,6 35,1 30,4±1,0 1,0 30,9
6 35,1±0,7 1,2 35,3 30,6±1,1 1,2 30,8
7 35,3±0,9 0,8 35,8 32,7±1,1 1,3 33,4
8 34,7±0,9 0,8 34,9 32,6±1,0 1,1 33,5
9 34,7±0,9 0,8 35,1 31,3±1,2 1,5 31,1
10 34,0±0,9 0,9 34,1 30,7±1,3 1,8 30,2
11 34,5±0,9 0,8 35,3 30,9±1,4 1,9 30,9
12 34,0±0,9 0,8 34,3 30,9±1,3 1,7 31,5
13 33,2±1,6 2,7 33,9 30,3±2,1 4,4 33,1
14 33,9±1,3 1,8 34,3 30,4±2,2 4,7 33,9
15 33,1±1,9 3,4 33,6 30,1±2,2 4,7 30,7
16 33,6±1,7 2,9 33,4 30,3±2,4 5,8 31,5
17 36,0±0,4 0,2 36,1 33,7±0,8 0,7 34,1
18 30,7±1,6 2,6 31,4 27,0±1,7 2,9 27,4
19 30,0±2,0 4,1 30,3 26,7±1,7 3,0 27,2
Tmax [°C] Tavg [°C]

35
32
Temperature [°C]
29
26
23
20
TBA TBD ToA ToD DH PH F FT
Area of Measurements
Fig. 24. The physiological temperature distribution of the human body

Graph (Figure 24, Table 4) shows the physiological temperature distribution of

the human body. The lowest temperatures were measured on acral parts of the
body (DH, F). On dorsal side of hands was calculated arithmetical average tem-
perature 23,3±1,8°C, and on feet 26,9±1,7°C. The highest value of average tempera-
ture was measured on a face of the head 33.7±0.8°C (accuracy of the used thermal
imager is ±2°C).
Areas of measurements: TBA (Total Body – Anterior view), TBD (Total Body –
Dorsal view), ToA (Torso- Anterior view), TBS (Total Body – Sagittal view), ToD
(Torso- Dorsal view), DH (Dorsal side of Hand), PH (Palmar side of Hand), F
(Foot) and FT (Facial Title).
Each of the normal thermograms was obtained in detail view of interesting area
and from same angle.
11 Conclusion
Obtained temperature values can be used as a reference for medical thermography,
where positive/negative findings of the disease or injury can be quantitatively
assessed.
By infrared thermography, we have obtained temperature maps on bodies
(Mn=15; Fn=15) in 19 views. In process analyzing of thermograms we obtained
qualitative and quantitative data on areas of interest. The results of the average
temperatures obtained from males in the first position vary in the interval <29.0;
31.9> which means the range of variation about (M)Rv=2.9°C±2°C. The female av-
erage temperatures obtained from the first position were found in the wider range
of temperatures <27.0; 32.9> with variation range (F)Rv=5.9°C. Accuracy of the
used thermal imager is ±2°C.
The database can be helpful for comparison in the diagnostics of many neuro-
vascular and musculoskeletal diseases or injuries.
Acknowledgment. This contribution is the result of the project implementation: Research

of New Diagnostic Methods in Invasive Implantology, MŠSR-3625/2010-11 Stimuls for
Reasearch and development of Ministry of Education, Science, Research and Sport of the
Slovak Republic) and the project: Creation and promotion of Technologies in diagnostics of
components and junctions with computed tomography (lTMS:26220220038) supported by
the Research & Development Operational Programme funded by the ERDF.
References
[1] Ammer, K.: Thermology 2009- a computer- assisted literature survey. Thermology
International; Published by the Austrian Society of Thermology and European Asso-
ciation of Thermology 20(1), 5–10 (2010) ISSN: 1560-604X
[2] Jain, R.K., Gullino, P.M.: Thermal Characteristics of Tumors: Applications in Detec-
tion and Treatment. Annals of the New York Academy of Science 335, ANYAA
9 335 1-542 (1980)
[3] Kerr, J.: Review of the effectiveness of infrared thermal imaging (thermography) for
population screening and diagnostic testing of breast cancer. NZHTA Tech. Brief Se-
ries 3(3) (July 2004)
[4] Ring, E.F.J.: The historical development of temperature measurement in medicine.
Infrared Physics &Technology 49(3), 297–301 (2007)
[5] Ring, E.F.: The historical development of thermometry and thermal imaging in medi-
cine. Journal of Medical Engineering & Technology 30(4), 192–198 (2006)
[6] http://www.comp.glam.ac.uk/pages/staff/pplassma/MedImaging/
Projects/IR/Atlas/index.html
[7] Sidun, J., Wejrzanowski, T., Dąbrowski, J.R., Kurzydłowski, K.J.: Quantitative de-
scription of cancellous bone by image analysis techniques. Journal of Vibroengineer-
ing 11(3), 547–550 (2009)
[8] Jain, M., Shakya, M.: A two dimensional mathematical model to study temperature
distribution in human peripheral region during wound healing process due to plastic
surgery. Global Journal of Pure and Applied Mathematics (April 1, 2009) ISSN:0973-
1768, http://www.thefreelibrary.com/_/print/
PrintArticle.aspx?id=215842184
[9] http://journal.library.iisc.ernet.in/vol201103/JIISc-
91%283%29-Web%20PDFs/JIISc-9103-ARUNN.pdf
[10] http://www.comp.glam.ac.uk/pages/staff/pplassma/MedImagin
g/Projects/IR/Atlas/index.html
[11] http://www.fluketi55.com/assets/images/pdfs/ti55/2674273_
0000_ENG_C_Wti55datasheet.pdf
[12] Gabrhel, J., Popracová, Z., Tauchmannová, H., Chvojka, Z.: Thermographic Findings
In The lower Back: Can They Be Explained By A Reflex Mechanism? Thermology
International 20(1) (January 2010) ISSN-1560-604X
[13] Zvarová, J.: Základy statistiky pro biomedicínské obory I. Biomedicínska statistika;
Univerzita Karlova v Prahe; Nakladateľstvo Karolinum (2007) ISBN: 978-80-7184-
786-1
[14] Jesensek Papez, B., Palfy, M., Mertik, M., Turk, Z.: Infrared thermography based on
artificial intelligence as a screening method for carpal tunnel syndrome diagnosis. J.
Int. Med. Res. 37(3), 779–790 (2009) PMID: 19589261
Determining the Calibration Element Location
in Planning of Surgical Interventions
Monika Michalíková1, Teodor Tóth2, Lucia Bednarčíková1, and Jozef Živčák1

1
Technical university of Kosice, Faculty of mechanical engineering,
Department of biomedical engineering and measurement, Letna 9, 042 00, Kosice, Slovakia
{monika.michalikova,lucia.hutnikova,jozef.zivcak}@tuke.sk
2
Department of automation, control and human machine interaction, Letna 9,
042 00, Kosice, Slovakia
teodor.toth@tuke.sk
Abstract. At the present time in many countries of the world preoperative plan-
ning of interventions for lumbar joint is realized with caliper, protractor, plastic
templates and x-ray images. From these reasons the measurement is time consum-
ing with many errors. Over the past few years, an increasing appreciation of the
usefulness of digital technology has emerged among various field of medicine.
CoXaM offers a simple solution to the problem of using digital X-ray images and
handmade plastic templates. The developed software combines digital X-ray im-
ages with digital templates for planning the implantation or re-implantation of hip
joints. The technique is reliable, cost-effective and acceptable to patients and radi-
ographers. It can easily be used in any radiography department after a few simple
calculations and preparation of appropriately sized discs. The CoXaM software
provides several advantages for orthopedic surgery. X-ray film is no longer neces-
sary, and there are no radiographs to store, lose or misplace. Significant clinical
studies are planned to confirm statistically the qualitative value of the software
and the quantitative precision of the output parameters.
1 Introduction
Computer technology has many applications in different fields of industry, health

care and medicine. This encompasses paper-based information processing as well
as data processing machines (Hospital information systems or Clinical information
systems) and image digitization of a large variety of medical diagnostic equip-
ment (e.g. computer images of X-ray, MR, CT). Many of these applications allow
the visualization and classification, respectively, the identification and the assess-
ment of the objects being diagnosed. The aim of computer technology in medicine
is to achieve the best possible support for patient care, preoperative surgical plan-
ning and administration by electronic data processing.

394 M. Michalíková et al.
Imaging techniques are an important element of contemporary medicine. An es-

timated seventy percent of all diagnoses are based on results obtained from ex-
aminations made using imaging methods. Imaging methods also play an important
role in total hip joint replacement surgery, as it is important to select the correct
size of the endoprosthesis in these operations, which can be determined very pre-
cisely by an X-ray image. When selecting the size, physicians use certain software
programs that help them measure the diameter of the femoral head and suggest the
optimal size of the endoprosthesis based on the obtained value. Not all software
makes it possible to correct errors (technology, methodology and personnel errors)
and distortions caused by scanning the patient using X-ray, however, and this can
then lead to an improper choice for the hip replacement.
The technology features that affect the resulting X-ray images include:
• X-ray image enlargement or reduction,

• X-ray image distortion caused by the projection type.
Historically, radiographs have not been standardized according to magnification.

Depending on the size of a patient, the film will either magnify the bone and joint
(of large patients with more soft tissue) or minimize them (in the case of thin pa-
tients). An orthopedic surgeon must estimate at what degree, plus or minus magni-
fication, to select an implant that is of the correct size. The surgeon may be helped
by incorporating a marker of known size. Then, by calculating the difference be-
tween the size of the marker displayed on the film and the actual size of the mark-
er, the orthopedic surgeon can identify the degree of magnification/minimization
and compensate accordingly when selecting a prosthetic template [6].
Kulkarni et al. (2008), for instance, devised a method whereby a planar disc
placed on the radiographic cassette accounts for the expected magnification.
Accurate preoperative planning improves the procedure’s precision [3,14], short-
ens its duration [3,23] reduces the incidence of prosthesis loosening [11,19] and loss
of bone stock [11,12,19], lowers the risk of periprosthetic fracture, helps restore
femoral offset and leg length in hip arthroplasty, facilitates optimization of align-
ment and ensures that the required implants are available while minimizing the costs
[15] and other complications (e.g., instability) [7,8,9,10,11,13,17,22,24,26].
Digital radiography is now becoming widespread. Accurate preoperative
templating of digital images of the hip traditionally involves positioning a
calibration object onto its centre. Not only can this be difficult, it can also cause
embarrassment [16].
Digital preoperative planning enables the surgeon to select from a library of
templates and electronically overlay them on the image. Therefore, the surgeon
can perform the necessary measurements critical to templating and the preopera-
tive planning process in a digital environment. The preoperative planning process
is fast, precise and cost-efficient, and it provides a permanent, archived record of
the templating process [5].
Determining the Calibration Element Location in Planning of Surgical Interventions 395
William Murzic et al. (2005) presented a study with the aim of evaluating the ac-
curacy of specific templating software (with an emphasis on femoral component fit)
and comparing it to the traditional technique of using standard radiographs [18].
The Department of biomedical engineering, automation and measurement has
developed software for preoperative planning of interventions in the hip joint area
(implantations and re-implantations of total endoprosthesis in the hip joint), which
uses calibration elements to eliminate errors arising during the scanning process.
One clear advantage is the possibility of dimensional calibration against the X-
ray image. In most software the dimensional calibration is fixed in terms of the
scanning-cartridge width, and the scanning distance is not taken into account,
which ultimately has a significant impact on error rate.
2 X-Ray Image Formation and Development

An X-ray image is a two-dimensional image of a three-dimensional object, and the
following three components are necessary to produce one:
• X-ray source – X-ray focus (O),

• Object (examined patient) (Ob),
• Film, shield, detectors and computer (F).
Hendrikus J. A. Crooijmans MSc et al. (2009) found a linear relationship between

magnification of a reference object at the pubis symphysis and the hip [13].
2.1 Central Projection
In the central projection the rays emerge from the real center, or focus, from which
they spread in a divergent way as the primary beam, whose surface is determined
by the size and slope of the focus and the primary shield setting. The beam of
radiation has the shape of a normal cone or pyramid, whose apex is at the focus.
The range of the primary beam is determined by the top angle and depends on the
slope and X-ray focus size, or the size of the outlet window on the X-ray tube
housing. All of the rays emerging from the focus and the X-ray tube create an X-
ray image, if they have proper direction and energy.
One particular representative ray from the primary beam is usually used in
practice; this is the most important ray for the projection. It is called the central
ray (CR), and it forms the primary radiation beam axis. Rays closer to the central
beam are less divergent than more distant rays. Using the central radiation beam,
we can aim the primary radiation beam appropriately. The central projection can
be vertical and oblique [1,2,4,20,27] (Fig. 1).
2.2 Parallel Projection

This is the ideal in parallel course rays, as the resulting image is without magnifi-
cation. Such a projection is close to telerengenography, where the scanning is at
the distance - the film S - F 200 cm. Increasing the distance between the X-ray
focus and film (O-F) is prevented by the extreme heat load of the X-ray tube focus
(Fig. 1).
2.3 Perpendicular Central Projection

This is a projection in which the central beam falls perpendicular to the level of
the film and eventually goes straight to the main object. It offers optimal projec-
tion and little distortion of the actual shape and size of the object (Fig. 1).
Fig. 1. Central, parallel and perpendicular and oblique projection
2.4 Oblique Central Projection
This is a projection in which the central beam falls obliquely on the plane of the
film. The object in the oblique central projection is enlarged in shape and distorted
in dimensions. The smaller the angle at which the central beam strikes the film or
shield, the greater the distortion. This projection is used to eliminate overlapping
(summing) formations (Fig. 1).
In virtually all examinations there is either greater or lesser magnification of the
displayed object. Magnification depends directly on the focus-object (O-Ob) dis-
tance and the object-film (Ob-F) distance and depends indirectly on the focus-film
(O-F) distance. The closer the object is to the X-ray tube, or the greater the dis-
tance between the object and the film, the greater the magnification.
In order for the object’s X-ray image to be as true as possible, i.e. for the en-
largement and distortion to be kept at a minimum, the studied part of the object
must be displayed in the central vertical projection and positioned as close to the
film as possible [1,2,4,20,27].
Fig. 2. Dependency of X-ray image size on the distance of focal
3 Anatomy and Morphology of the Hip Joint
The hip joint, scientifically referred to as the acetabulofemoral joint (art. coxae), is
the joint between the femur and acetabulum of the pelvis, and its primary function
is to support the weight of the body in both static (e.g. standing) and dynamic (e.g.
walking or running) postures.
The hip joint (Fig. 3) is a synovial joint formed by the articulation of the
rounded head of the femur and the cup-like acetabulum of the pelvis. It is a special
type of spheroid or ball-and-socket joint, where the roughly spherical femoral
head is largely contained within the acetabulum and has an average radius of cur-
vature of 2.5 cm [25].
Fig. 3. Right hip joint – cross-section view
The hip muscles act on three mutually perpendicular main axes, all of which
pass through the center of the femoral head, resulting in three degrees of freedom
(Fig. 4) and three pair of principal directions:
Fig. 4. Three degrees of hip joint freedom

Flexion and extension around a transverse axis (left-right); lateral rotation and
medial rotation around a longitudinal axis (along the thigh); and abduction and
adduction around a sagittal axis (forward-backward); as well as a combination of
these movements (i.e. circumduction, a compound movement in which the leg
describes the surface of an irregular cone) [21].
The most important morphological specifications (Fig. 3) that can be measured
on an anteroposterior pelvic radiograph are:
• the femoral neck angle (caput-collum-diaphyseal angle, or CCD angle) - be-

tween the longitudinal axes of the femoral neck and shaft, normally measures
approximately 126° in adults,
• the acetabular inclination (transverse angle of acetabular inlet plane) - the angle
between a line passing from the superior to the inferior acetabular rim and the
horizontal plane, normally measures 40° in adults [25].
Fig. 5. Femoral neck angle and acetabular inclination
The next important morphological specification is the femoral neck antever-

sion, which cannot practically be measured in an AP projection, though it can be
measured well by CT or MR (3D measurement methods).
A perfect AP radiograph of the femur needs to account for the anteversion of
the femoral neck. Patients are required to rotate the leg internally by a mean of
15°. Restricted rotation of the hip in osteoarthritis sometimes makes it difficult to
achieve this position. One study of the radiological dimensions of the femoral
canal showed that the AP width of the medullary canal at the isthmus did not
change significantly from 20° internal to 40° external rotation. At 20 mm below
the lesser trochanter there was no significant change in internal rotation and an
apparent increase of 1.1 mm with 20° external rotation.
4 Digitalization of Preoperative Planning Using CoXaM

Software
The “CoXaM” software was developed in Visual Studio 2005 (Microsoft) in the
Visual C++ programming language at the Department of Biomedical Engineering,
Automation and Measurement at the Faculty of Mechanical Engineering, Technic-
al University of Kosice.
Fig. 6. Overview of main menu
The new sophisticated CoXaM software (See Fig. 6) was designed for preoper-
ative planning and helps to determine the length dimensions, the center of rotation
and the angle values on an X-ray image (See Fig. 7). These parameters are consi-
dered in parallel with guidance lines. The software enables the digitization of plas-
tic templates from several producers who will assess the suitability of the type of
implant. By using digital templates, the surgeon can employ a sequential method
to determine the size of prosthesis to use and where to place the prosthesis within
the bone to ensure optimum function of the joint following surgery. The incorpo-
ration of the various templates into the software in terms of the “magnification
factor” is essential for accurate preoperative templating and planning.
Fig. 7. Example of using the “CoXaM” and control panel
4.1 Possibilities of CoXaM Software
• Calibration circle (See Fig. 8-a) – allows the exact conversion of the marking
dimensions for the given calibration feature on the X-ray. Determining the
three points allows the plotting of a circle whose diameter in millimeters is en-
tered by the user, in this case 28 mm.
Fig. 8. Control panel detail – a- calibration circle, b- center icon, c- measurement of dimen-
sion, d- circle, e- text, f- angle, g- examination of three line parallelism, h- removing the
planning, i- templates
At present, preoperative planning software is used when it is necessary to de-

fine a magnification reference object on an X-ray image (See Fig. 9). The magni-
fication of the real diameter is determined, and the diameter of the reference object
is measured on the X-ray image.
With CoXaM software it is not necessary to know the X-ray image magnifica-
tion. The user defines the value of the reference object diameter, and this elimi-
nates the necessity for measurement (See Fig. 9).
Fig. 9. Software used at the present time (diameter evaluation– real diameter of the calibra-
tion component is 28,00 mm) and CoXaM software (determining of the reference object
diameter).
• Center icon (See Fig. 8-b) – centers the X-ray image into the viewport.
• Measurement of dimension (See Fig. 8-c) – calculates the distance between two
points. If the calibration is done the result is in millimeters; otherwise it is dis-
played in pixels.
• Circle (See Fig. 8- d) – from three points the software calculates a circle (cen-
ter, diameter). If the calibration is done, the result is in millimeters; otherwise it
is in pixels. The circles are used for finding the center of the hip joint and de-
fining the dimension of the femoral head and the diameter of the acetabular
component. Using the circle a user can determine the floatable center of rota-
tion before and after the surgery.
• Text (See Fig. 8-e) – allows the user to enter text. The font used is Arial 12 pt.
• Angle (See Fig. 8-f) – the angle between two lines (created from four points).
The two lines do not have to intersect.
• Examination of three-line parallelism (See Fig. 8-g) – the L. Spotorno and S.
Romagnoli method calculates the parallelism between three lines (created from
six points) – the ischial tuberosities flowline (the base line), the superior aceta-
bular rims flowline and the lesser trochanters flowline.
• Removal of planning (See Fig. 8-h) – removes all these tasks and clears the X-
ray image.
• Templates (See Fig. 8-i) – opens the digital template from the database of
scanned templates from total hip prosthesis producers. This allows the tem-
plates to be calculated and inserted into an X-ray image. The size of the tem-
plate is equal to size of the X-ray, and it’s possible to rotate and move it.
Traditionally, an orthopedic surgeon places an acetate template enlarged to 110%

to 120% over an X-ray film magnified to 110% to 130% [1].
An object (a disc or sphere is recommended) of known size is projected on to
the film in order to determine the magnification.
In digital radiograph templating, the template and radiograph can be scaled to
obtain identical magnifications.
When properly placed, the magnification of the reference object represents the
magnification of the hip (a one-to-one relationship) and thereby enables accurate
preoperative templating. The method requires the reference object to be properly
placed at the same distance from the detector as the center of the hip rotation.
Alternative methods for correcting magnification, including the use of a line as a
magnification reference, of coins placed at various positions, of software to tem-
plate digital radiographs, or software to template CT data have been previously
described.
For preoperative planning interventions of hip-joint arthroplasty (implantation,
re-implantation) plastic templates are commonly used. Each manufacturer offers
its own plastic templates (types and sizes) for their implant product (See Fig. 10).
Fig. 10. Plastic templates
The CoXaM software works with electronic templates, which are placed in a
folder containing the scanned plastic templates. The user selects the required tem-
plate from that folder (See Fig. 11).
If the selected template has not been calibrated, then the next step is to do the
calibration. This process is done in the calibration window, in which the abscissa
is marked on the scale of the selected templates and the user inserts a value into
the box (See Fig. 12, step 2). This value must correspond with the template scale
(e.g., the scale of the template is 1.15:1; if the user marks a 10mm abscissa, then
the inserted value in the box (See Fig. 12, step 2) is 11.5). After selecting the op-
tion OK, the calibration is confirmed. The selected templates are then drawn in
yellow and applied to the X-ray image. The user can also use the tool for template
mirroring, if the preoperative planning is for a counteractive hip joint.
Fig. 11. Selection of required template
The calibration can then be saved before confirmation, meaning that for future
reference, calibration with that particular template is no longer necessary (for each
template it is necessary to perform a calibration at first use).
The user can to move the template to the required location and rotate it around
its center (See Fig. 13).
Selected orthopedic departments used a demonstration version of the CoXaM
software. The respondents learned basic knowledge about working with the
software. Afterward, the orthopedists filled in questionnaires in which they
described their opinion of the CoXaM software. According to the results of the
questionnaires, learning to use the CoXaM software takes from 30 to 60 minutes
and preoperative planning takes from 10 to 15 minutes for each case for a skilled
user.
Fig. 12. Calibration process of template
Fig. 13. Placement of the templates over x-ray image such that optimal fill of both
5 CoXaM Software Verification
It follows from the above-stated that the projection type and the shooting distance
have a significant impact on the resulting X-ray image. In order to determine the
endoprothesis size correctly, it is necessary to use the calibration element with a
given size, which then can be used to recalculate the other necessary data. It is
likewise necessary to establish the appropriate location of the calibration element
for each projection site and type in each patient. Total hip endoprothesis heads
28mm in diameter were used as calibration features. The distribution of the cali-
bration elements depends on the size of the cartridge which is commonly used in
X-ray scanning of the hip-joints in AP projection, whereby one head was placed
higher than the others (a base of 12cm in height). The focal distance used in scan-
ning the calibration elements corresponds to the distance used in scanning the hip-
joints in AP projection, i.e., 1 meter. Overall, three variants for the distribution of
calibration elements on the surface of the image were evaluated.
Element number 1 represents the anatomical site that is primarily measured. Its
location is on all the X-rays (the same images). The distribution of the numbered
elements 1 to 4 represents the greatest distance that can be achieved within the
sensor cartridge. In addition, elements 1 to 4 are in the same plane as the measured
locality. (Fig. 14)
Fig. 14. Evaluated X-ray image for variant 1

For the second and third variant the position of element number 5 was elevated
about 0.12m due to the measuring plane.
The distribution of elements is observed by the horizontal distance between the
elements, and the vertical distance between the X-ray focus and the elements af-
fects the size of the elements.
For the second and third variant the position of element number 5 was elevated
about 0.12m due to the measuring plane.
The distribution of elements is observed by the horizontal distance between the
elements, and the vertical distance between the X-ray focus and the elements af-
fects the size of the elements.
5.1 Variant 1
In the first variant the elements are distributed in one plane without elevation, and
the central beam is directed to the center, where element number 1 is positioned.
From this element the other three elements are distributed at different distances.
The focal length is 1 m. This distribution is designed to identify how much the
size of elements on the X-ray image is affected if they are distributed at different
distances from center point while still lying in the same plane. (Fig. 15)
Fig. 15. Layout of elements and the X-ray image for variant 1
5.2 Variant 2
In the second variant the elements are distributed in a single plane, except element
number 5, which has an elevation of 0.12 m and is then compared to the other
elements (1-4). The final magnification is the sum of the magnifications from the
horizontal distance of element 5 to the central element and the elevation of ele-
ment 5 from the plane in which elements 1 – 4 are placed. (Fig. 16)
5.3 Variant 3
In the third variant the elements are distributed in a single plane, except element
number 5, which has an elevation of 0.12 m and is compared to the other elements
(1-4). Element number 5 is situated near the central element (element 1). The dis-
tortion when compared with variant 2 should be smaller because the angle of the
X-ray beam’s impact is greater than the previous variant. The final magnification
is the sum of the magnifications from the horizontal distance of element 5 to the
central element and the elevation of element 5 from the plane in which elements 1
– 4 are placed. (Fig 17)
6 Methodology and Evaluation of Measurements

Persons who evaluated X-ray images measured the diameters of the given ele-
ments and the distance between the elements. There was a minimum of a 5-minute
time gap between the two evaluated images, while the slide was evaluated 30
times. The mean value was then calculated from these measurements, as were the
standard deviation and the upper and lower expected size, while the normal
(Gaussian distribution) distribution and expansion coefficient k = 3 for 99% of all
values were considered. In order to compare the individual persons who evaluated
the images, a summary table was created which listed the average of the evaluated
dimensions for the given image and the evaluating person. The head in the middle
projection was used as the reference element, and all the other elements were
measured in reference to this head. Tables 1, 2 and 3 are data summary tables for
all evaluating persons for each.variant.
The heads of total joint replacements are primarily evaluated, while the length
dimensions are only of an informative character. Based on the results it can be
stated:
• In the evaluation of X-ray images there were no outliers under the rule
(± 3. standard deviation); all values lie in the 99% interval (for Gaussian distri-
bution).
• Standard deviations for diameters in the summary tables for all X-ray images
are in the range from 0.058 to 0.15 mm.
• The values of the standard deviations are highest for element number 5, which
is elevated.
• Standard deviations for the length dimensions in the summary tables are in the
range from 0.40 to 1.13 mm.
Table 1. Data summary table for all evaluating persons for variant 1
Persons Ø2 [mm] Ø3 [mm] Ø4 [mm] L1-2 [mm] L1-3 [mm] L1-4 [mm] L2-3 [mm]
1 28,17 28,08 28,11 138,78 166,51 88,39 80,41
2 28,29 28,21 28,20 138,14 166,01 88,22 80,30
3 28,09 28,056 28,06 139,97 167,25 89,96 81,79
4 28,22 28,16 28,20 138,076 166,27 88,78 80,39
5 27,99 27,99 28,046 139,023 167,06 88,57 80,95
Average [mm] 28,15 28,10 28,12 138,79 166,62 88,78 80,77
Standard deviation
[mm] 0,11 0,086 0,075 0,77 0,52 0,69 0,63
Maximal estimated
dimension [mm] 28,50 28,36 28,35 141,11 168,19 90,85 82,65
Minimal estimated
dimension [mm] 27,81 27,84 27,89 136,48 165,05 86,72 78,89
Ø2 Ø3 Ø4 Ø5 L1-2 L1-3 L1-4 L1-5 L2-3

Persons [mm] [mm] [mm] [mm] [mm] [mm] [mm] [mm] [mm]
1 28,13 28,15 28,13 31,97 137,93 166,047 88,08 142,83 79,92
2 28,3 28,29 28,18 32,05 138,57 166,64 88,59 143,75 80,29
3 28,11 28,25 28,04 31,73 140,02 166,86 89,46 143,84 80,69
4 28,23 28,21 28,19 32,02 137,69 165,82 87,93 142,84 79,35
5 28,05 28,03 28,11 32,066 137,023 165,66 88,056 142,24 80,12
Average
[mm] 28,17 28,19 28,13 31,97 138,25 166,21 88,43 143,1 80,074
Standard
deviation
[mm] 0,10 0,10 0,061 0,14 1,13 0,52 0,63 0,68 0,49
Xmax [mm] 28,47 28,49 28,31 32,38 141,65 167,77 90,32 145,14 81,57
Xmin [mm] 27,87 27,88 27,94 31,56 134,84 164,64 86,53 141,057 78,58
Persons Ø2 Ø3 Ø4 Ø5 L1-2 L1-3 L1-4 L2-3

1 28,12 28,13 28,09 31,76 138,12 167,54 87,23 80,023
2 28,25 28,22 28,21 31,77 137,77 166,77 86,88 80,007
3 28,1 28,06 28,04 31,56 139,04 168,38 88,18 80,45
4 28,17 28,13 28,15 31,72 136,92 166,01 86,28 79,33
5 28,10 28,11 28,093 31,99 137,02 166,36 86,57 79,84
Average [mm] 28,15 28,13 28,12 31,76 137,77 167,011 87,029 79,93
Standard deviation [mm] 0,063 0,058 0,065 0,15 0,87 0,95 0,73 0,40
Maximal estimated dimension
[mm] 28,34 28,30 28,31 32,22 140,38 169,87 89,22 81,14
Minimal estimated dimension
[mm] 27,96 27,95 27,92 31,30 135,17 164,15 84,83 78,72
The software is able to calculate diameter values to 1 decimal place; greater ac-
curacy is not required because the size of total joint replacement is graded on the
order of mm (e.g. 24, 28, 32).
6 Conclusion
The evaluation table shows that the elements lying in one plane (excluding the
reference element) have negligible differences between them regarding their aver-
age evaluation. Element number 5, which is located closer to the source of radia-
tion, has a larger diameter from the reference element by nearly 4 millimeters.
On X-ray image number 2 the final magnification is a function of the distance
from element 1 and the elevation from the plane in which elements 1 – 4 are
placed. On image number 3 the final magnification is a function of the elevation
from the plane in which elements 1 – 4 are placed.
It is not relevant at which point of the cartridge sensor element number 5 was
located the, as there is still magnification. Therefore, it is important to select the
location of the reference ball – to the sensing area (hip joint).
It follows from the above-stated that in the X-ray scanning of the hip-joint area
using the AP projection, it is necessary when selecting the placement of the cali-
bration element to focus on compliance with the same distance to the X-ray tube,
i.e. the calibration element should lie in one plane with the scanned anatomical
location, which is perpendicular to the central beam. The distance between the
calibration element and the scanned site within the plane will not significantly
affect the resulting distortion.
In the AP projection of the hip-joint area, the lateral location in the location of
the greater femur trochanter meets these criteria the most.
Acknowledgments. This contribution is the result of the project implementation: Creation

and promotion of Technologies in diagnostics of components and junctions with computed
tomography (lTMS:26220220038) supported by the Research & Development Operational
Programme funded by the ERDF.
References
[1] Bencur, J.: Základy RTG diagnostických metód, 2nd edn., vol. 1, 158 p. Vydavateľstvo
Osveta, Bratislava (1962)
[2] Bencur, J.: Základy RTG diagnostických metód, 2nd edn., vol. 2, 158 p. Vydavateľstvo
Osveta, Bratislava (1962)
[3] Blackley, H.R., Howell, G.E., Rorabeck, C.H.: Planning and management of the dif-
ficult primary hip replacement: preoperative planning and technical considerations.
Instr. Course Lect. 49, 3–11 (2000)
[4] Blažek, O., et al.: Obecná radiodiagnostika, 1st edn. Státní pedagogické nakladatelství
(1974)
[5] Bono, J.: MD: Digital Templating in Total Hip Arthroplasty. The Journal of Bone and
Joint Surgery (American) 86, 118–122 (2004); 2004 The Journal of Bone and Joint
Surgery, Inc.
[6] Bono, J.: Digital Templating in Total Hip Replacement: A New Gold Standard? Jour-
nal of Bone and Joint Surgery - British 86-B(suppl. IV), 413 (2004)
[7] Cech, O., Fassbender, M., Kirschner, P., Rozkydal, Z.: Preoperative planning and
surgical technic in achieving stability and leg length equality in total hip joint arthrop-
lasty. Acta Chir. Orthop. Traumatol. Cech. 69, 362–368 (2002) (in Czech)
[8] Eggli, S., Pisan, M., Muller, M.: The value of preoperative planning for total hip arth-
roplasty. J. Bone Joint Surg. Br. 80, 382–390 (1998)
[9] Goldstein, W.M., Gordon, A., Branson, J.J.: Leg length inequality in total hip arth-
roplasty. Orthopedics 28(suppl. 9), 1037–1040 (2005)
[10] Goodman, S.B., Huene, D.S., Imrie, S.: Preoperative templating for the equalization
of leg length in total hip arthroplasty. Contemp. Orthop. 24, 703–710 (1992)
[11] Haddad, F.S., Masri, B.A., Garbuz, D.S., Duncan, C.P.: The prevention of periprosthetic
fractures in total hip and knee arthroplasty. Orthop. Clin. North Am. 30, 191–207
(1999)
[12] Haddad, F.S., Masri, B.A., Garbuz, D.S., Duncan, C.P.: Femoral bone loss in total hip
arthroplasty: classification and preoperative planning. Instr. Course Lect. 49, 83–96
(2000)
[13] Crooijmans, H.J.A., Laumen, A.M.R.P., van Pul, C., van Mourik, J.B.A.: A New
Digital Preoperative Planning Method for Total Hip Arthroplasties. Clin. Orthop. Re-
lat. Res. 467, 909–916 (2009), doi:10.1007/s11999-008-0486-y
[14] Knight, J.L., Atwater, R.D.: Preoperative planning for total hip arthro-plasty: quanti-
tating its utility and precision. J. Arthroplasty. 7, 403–409 (1992)
[15] Kosashvili, Y., Shasha, N., Olschewski, E., Safir, O., White, L., Gross, A., Backstein,
D.: Digital versus conventional templating techniques in preoperative planning for to-
tal hip arthro-plasty. Can. J. Surg. 52(1), 6–11 (2009)
[16] Kulkarni, A., Partington, P., Kelly, D., Muller, S.: Disc calibration for digital templat-
ing in hip replacement. Journal of Bone and Joint Surgery - British 90-B(12), 1623–
1626, doi:10.1302/0301-620X.90B12.20238
[17] Lindgren, J.U., Rysavy, J.: Restoration of femoral offset during hip replacement: a
radiographic cadaver study. Acta Orthop. Scand. 63, 407–410 (1992)
[18] Murzic, J.W.M.D., Glozman, Z.B.S., Lowe, P.R.N.: The Accuracy of Digital (film-
less) Templating in Total Hip Replacement. In: 72nd Annual Meeting of the Ameri-
can Academy of Orthopaedic Surgeons in Washington, DC, February 23-27 (2005)
[19] Müller, M.E.: Lessons of 30 years of total hip arthroplasty. Clin. Orthop. Relat.
Res. 274, 12–21 (1992)
[20] Nekula, J., et al.: Konvenčná rádiológia. selected chapters
[21] Platzer, W.: Color Atlas of Human Anatomy, Locomotor System. 5th revised anden-
larged English edition, vol. 1. Thieme, Stuttgart (2004) ISBN: 3-13-533305-1
[22] Rubash, H.E., Parvataneni, H.K.: The pants too short, the leg too long: leg length in-
equality after THA. Orthopedics 30, 764–765 (2007)
[23] Schiffers, N., Schkommodau, E., Portheine, F., Radermacher, K., Staudte, H.:
[xxPlanning and performance of orthopedic surgery with the help of individual tem-
plates]. Orthopäde. 29, 636–640 (2000) (in German)
[24] Schmalzried, T.P.: Preoperative templating and biomechanics in total hip arthroplas-
ty. Orthopedics 28(suppl. 8), 849–851 (2005)
[25] Schuenke, M., Schulte, E., Schumacher, U., Ross, L.M., Lamperti, E.D.: Thieme At-
las of Anatomy (2006) ISBN: 10: 3131420812
[26] Suh, K.T., Cheon, S.J., Kim, D.W.: Comparison of preoperative templating with
postoperative assessment in cementless total hip arthroplasty. Acta Orthop. Scand. 75,
40–44 (2004)
[27] Svoboda, M.: Základy techniky vyšetrovania rentgenem, 2nd edn. Praha, Avicenum
(1976)
Application of Magnetic Microwires
in Titanium Implants – Conception
of Intelligent Sensoric Implant
Radovan Hudák1, Rastislav Varga2, Jozef Živčák1, Jozef Hudák3, Josef Blažek3,
and Dušan Praslička3
1
Department of biomedical engineering and measurement, Letna 9, 042 00,
Kosice, Slovakia
{radovan.hudak,jozef.zivcak}@tuke.sk
2
Pavol Jozef Safarik university in Kosice, Faculty of science,
Department of Physics of Condensed Matters, Park Angelinum 9, 041 54,
Kosice, Slovakia
rastislav.varga@upjs.sk
3
Technical university of Kosice, Faculty of Aeronautics,
Department of Aviation Technical Studies, Rampová 7, 041 21 Kosice, Slovakia
{jozef.hudak,josef.blazek,dusan.praslicka}@tuke.sk
Abstract. The idea of intelligent sensoric implant which enables to scan parame-
ters from the human body wireless comes from analysis of studies descrbing rea-
sons of implants rejection or loosening. Inflamations and incorrect biomechanical
load are offen the reasons for surgery, where implant has to be removed or re-
placed. Presented study shows a concept of intelligent dental implant, where
magnetic microwires are placed and fixed into titanium dental implant to get pa-
rameters from implant, tissue, or implant-tissue interaction. A part of the study
shows preparation of magnetic microwires, measurement of physical quantities
using bistabile magnetic microwires and realisation of the functional model of the
sensor and experiments. Obtained results show, that utilization of magnetic mi-
crowires in implants for scanning of selected physiological or physical parameters
is promising. The further researches in the field of fabrication technology, magnet-
ic wires preparation and scanning processes to confirm an intelligent sensoric
implant concept is necessary.
Keywords: intelligent implants, metrotomography, direct metal laser sintering,

magnetic microwires.
1 Introduction
Design of new intelligent sensoric implants is the process where integration

of knowledge from more scientific fields is required. Starting with the
developments, design and fabrication of microsensors, design of new implants with
installation space and shapes for application of sensors, ending with realization of

the functional model of the sensor and experiments related to scanning of selected
parameters wireless from the human body.
Idea of intelligent implant is not new; already several studies were published on
application of microsensors into the human body with aim to get physiological
parameters from implant or implant-tissue interface in the wireless, non-invasive
way [6].
Ainslie and Desai in 2008 described several methods of biosensor microfabri-
cation (macromachining, photolithography, electrochemical microfabrication and
micromachining, etc.). Micro-biosensors are able to analyze following parameters:
glucose, neural action potentials, spinal pressure, atherosclerotic pressure, bacteri-
al growth, etc. Mechanical loading biosensors use piezoresistive effect or laser
diffraction [1]. Kubon et. al. fabricated the microsensor implant (MSI) using stan-
dard thin film techniques. Ti/Pt electrodes were deposited and structured on glass.
Micro-sized ion selective electrodes (ISE) were realized by electrodeposition of
iridium oxide as pH electrode and silver/silverchloride as pseudo-reference elec-
trode. In their experimental study dissolved oxygen and pH was measured in three
electrode configuration and potentiometric measurements. Experiments were rea-
lized in quasi-in vivo environments with aim to quantitatively analyze biocompa-
tibility of biomaterials and bioactive coatings [15].
Mouzakis. et. al.in 2009 introduced in their study contact-free magnetoelastic
microsensor for diagnosing failing orthopedic implants. It uses an implant-
embedded 30-µm-thick MetGlas-2826 film and a sensing coil placed 30 mm away
from the film. Implant loading generates thin-film magnetoelastic response, pas-
sively (contact-free) detected as voltage by the coil [18]. A specific integrated
stochastic nonlinear filter performs data denoising, and a decision-making module
concludes (via data assessment) on the health state of the implants. The magneto-
strictive/magnetoelastic materials accepted to belong to the “smart materials”
fadapt their behavior to the outside stimulus. Magnetoelastic sensors are typically
made of amorphous ferromagnetic ribbons or wires, commonly referred to as
“metal glasses”, and usually composed of iron-rich alloys. Libermann and Graham
[17] developed in 1976 a new method of manufacturing thin ribbons of amorphous
metal on a supercooled fastspinning wheel. This was an alloy of iron, nickel,
phosphorus, and boron Fe40Ni38Mo4B18 and is nowadays marketed as MetGlas.
MetGlas-2605 is composed of 80% iron and 20% boron, has a Curie temperature
of 373°C, and a room temperature saturation magnetization of 125.7 mT. An in-
teresting feature of magnetoelastic sensors compared with other well-established
high-accuracy techniques (fiber Bragg gratings (FBGs) [4] or piezoelectric sensors
[16] is that the specimens, parts, or structures need not be connected with electric-
al conducting wires or optical fibers to the signal processing and analyzing system.
Presented study describes a conception of intelligent sensoric implant from im-
plant design, preparation and fabrication point of view together with preparation
of magnetic microwires (amorphous glass-coated microwires) and measurement of
physical quantities using bistabile magnetic microwires as shown on fig.1.
Application of Magnetic Microwires in Titanium Implants 415
Fig. 1. Flowchart of processes essential for design of intelligent sensoric implant
2 CAD Modelling and Laser Sintering of Titanium Implants

First step in development of intelligent sensoric implant was analysis of the
present studies and research and the history. Input data for following CAD analy-
sis were data obtained by industrial computed tomography (Metrotom 1500, Carl
Zeiss, Germany).
After scanning process, the point cloud was obtained and the geometry of the
implant was assessed (fig. 2). By VGstudio MAX software the point cloud of the
implants was transformed to 3D model data and the slices of the implants were
created in two views (right & top) and exported to jpeg format and also exported
to STL format which is standard format for further CAD/CAM analysis.
Fig. 2. Cross-sections of titanium dental implant
Before construction the implant data were modified by Rapidform software to

polygon structure, afterwards diagnostics of the surface parameters was realized,
its optimization and export to IGS and STP format for final CAD/CAM design
process [5][24].
Data stored in IGS and STP format are fully compatible with mainstream
CAD/CAM software, like SolidWorks, and used for common design proposal and
modification (fig. 3).
Fig. 3. Mesh models (left) and *.IGS/ *.STP file formats of the dental implant parts
In SolidWorks 2012 all the implants were constructed by design of the each
part of the implant. Implant parts were constructed based on parameters and
shapes of the implant digitalized by metrotomography (mainly the size).
Fig. 4. CAD models of dental implants
After final construction the model of the implant and all its components were
exported to STL format (fig. 4) due to additive manufacturing technology requests
(volume of the part). As a material for designed implant fabrication was selected
titanium (Ti64).
The popularity of titanium and its alloys in dental and medical fields can be
recognized by counting the manuscripts published in different journals [21]. The
increased trend of published papers may be attributed also to increased researchers
and scientists involved in these medical titanium materials, mainly because of

their good fatigue resistance characteristics, excellent in vivo corrosion resistance,
a feature related to the stable passive oxide layer (TiO2) that rapidly forms, their
lower elastic moduli compared to other metallic biomaterials (100-110 GPa com-
pared to 200-220 GPa) [19].
Fig. 5. Surface structure (left) and microstructure of titanium implant
Alloying of Ti is used to form a two-phase (α+β) alloy of higher strength

(yield, ultimate and fatigue) than commercial purity Ti (CP Ti) while maintaining
excellent corrosion resistance and osseointegration tendency, again because of the
TiO2/OH surface film [19].
In order to biocompatibility the quality of surface of titanium implant is neces-
sary (fig. 5). The analysis of surface properties was realized to compare the sur-
face after installation of sensoric microwires. Magnetic microwires are coated by
glass, which assures biocompatibility after implantation.
There are more technologies available for titanium medical implants fabrica-
tion: casting, machining, Electro Discharge Machining (EDM), Isothermal Form-
ing, Diffusion Bonding (DB), Powder Metallurgy, Metal Injection Molding
(MIM), Soldering, Heat Treatment (HT), etc [21]. One of the novel technologies is
Selective Laser Sintering (SLS), when related to sintering of metal powder and
high energy laser with high heat resistance [2] called Direct Metal Laser Sintering
(DMLS). SLS uses a laser emitting infrared radiation, to selectively heat powder
material spread on a building platform just beyond its melting point. The laser
traces the shape of each cross-section of the model to be built, sintering powder in
a thin layer. After each layer is solidified, the piston over the model retracts to
a new position and a new layer of powder is supplied using a mechanical roller
(fig. 6). The powder that remaines unaffected by the laser acts as a natural support
for the model and remains in place until the model is complete. The resolution of
the SLS system is restricted by the powder particle size, on the one hand, and by
the size of the focal point of the laser, on the other hand [2].
Fig. 6. Principle of Direct Metal Laser Sintering (DMLS) fabrication technology of tita-
nium implants [2]
In our research the EOSINT M280 (EOS, Germany) DMLS machine will be
used. The next activities will direct to verification of the technology for disign of
installation holes for placement of magnetic microsensors.
It is well recognized that the first reaction of vital hard/soft tissue to any type of
biomaterial is a rejection; accordingly, biomaterial is normally recognized as a
„foreign material“ by the host tissue. The biological acceptance of these foreign
materials by the living tissues is essentially controlled by the surface and interfa-
cial reaction between the organic substance and inorganic substance.
Early implant instability has been proposed as a critical factor in the onset and
progresion of aseptic loosening and periprosthetic osteolysis in total joint arthroplas-
ties [21].
Monitoring of parameters and markers showing loosening, rejection, inflama-
tions of implant and surrounding tissue is important after implantation.
Fig. 7. Design of new titanium dental implant with small space for microsensors installation.
Application of Magnetic Miccrowires in Titanium Implants 4119
Figure 7 shows new tittanium dental implant with detail where installation holees
for magnetic microwires are
a placed.
Fig. 8. Details of magnetic microwire (red color) installed in titanium dental implant – a
concept.
The cross section and the transparent model with simulation of installed m mi-
crowires are shown on fig
g. 8. The diameter of microwire is up to 50 µm (red wiree).
3 Preparation of Magnetic
M Microwires
The amorphous glass-coaated microwires are novel material very promising foor
technological application ns because of their small dimensions, insulating glass-
coating and very simple but
b efficient preparation method [30][33]. There are com m-
posite materials that con nsists of metallic nucleus of diameter (from 100 nm tto
50 µm) covered by a glaass-coating with thickness from 2-20 µm. They are pre-
pared by drawing and rap pid quenching of molten master alloys [7]. Due to theeir
amorphous nature, their magnetic
m properties are governed mainly by magnetoelaas-
tic anisotropy. Having positive magnetostriction, their domain structure consists oof
large, axially magnetizedd, single domain that is surrounded by the radial domaiin
structure [20]. Moreover, small closure domains appear at the ends of the wire iin
y fields (fig. 9).
order to decrease the stray
Axial domain
a4
External domain
Closure Glass structure
domain coatinga 2
Fig. 9 SEM microscopy imaage of a glass-coated microwire (left) and schematic magnettic
domain structure of the glasss-coated microwire with positive magnetostriction (right) [28]
4 o
550 C
3
2
1
B(a.u.)
0
-1
-2
-3
-4
-400 -300 -200 -100 0 100 200 300 400
H(A/m)
Fig. 10. Hysteresis loop of glass-coated microwire with positive magnetostriction shows
magnetic bistability [11].
As a result of peculiar domain structure, magnetization process in axial direc-

tion runs through the depining and subsequent propagation of single closure do-
main along entire microwire in a single Barkhausen jump at the level of critical
field. Hysteresis loop of such microwire is perfectly squared (fig. 10) and such
microwires present magnetic bistability (e.g. magnetization has just two values
+Ms and - Ms). The switching between these two states appears when the external
field reaches the value of so-called switching field Hsw.
Magnetic bistability is very well compatible to the PC binary concept. The two
magnetization states corresponds to two logic states 0 and 1 in PC logic. Moreo-
ver, the switching field can be easily estimated by contactless induction method.
Sharp maximum appears at the ends of the pick-up coil (placed close to the mi-
crowire) when the external field approaches the switching field value (fig. 11).
When the exciting field has triangular shape, the time at which the maximum ap-
pears is proportional to the switching field.
Fig. 11. Sharp maximum (blue line) appears at the ends of the pick-up coil placed close to
the microwire when the external field (yellow line) approaches the switching field. Using
triangular shape of the exiting magnetic field, the time at which the switching appears is
proportional to the switching field.
The system of various microwires with different switching field can be used foor
magnetic coding [31]. Figg. 12 shows magnetic binary code 0110001000 that conn-
sists of 3 microwires. In contrary to the classical BAR code, the magnetic codde
can be sense by simple innduction method without having visual contact with thhe
code. The biggest advanttage, however, is the size of the magnetic code (takinng
into account the dimensio
ons of microwires).
0 1 1 0 0 0 1 0 0
Fig. 12. Magnetic code madee from 3 bistable microwires shows binary code 0110001000
Another possibility forr microwire’s application is their utilization for as a senns-

ing element for sensors of magnetic field, temperature, mechanical stress, etcc..
Basic idea arises from strrong dependence of the switching field on various exteer-
nal parameters that can bee used for sensing.
3.1 Frequency Depeendence of the Switching Field

As it was shown earlier [26][25][29], the domain wall potential of the closurre
domain wall in microwirees consists of two contributions: magnetoelastic one annd
structural relaxation of th
he defects introduced into the microwire during produc-
tion. These two contributions can be recognized by measuring the switching fielld
mperatrures and frequencies [25][29]. The structural ree-
distribution at various tem
laxation prevails at low frequences
f (below 100Hz), whereas magnetoelastic conn-
tribution is responsible for magnetic properties at higher frequencies (abovve
100Hz) [29].
Due to the existence of
o two contributions to the domain wall potential of thhe
closure domain wall in amorphous microwires, the switching field of bistable
microwires can also be satisfactory
s described by two contributions [26]. Firstly,
the strongest one is the magnetoelastic contribution that is defined in case oof
amorphous microwires as:
σ
λ sσ
H sw ∝ (1)
μ0 M S
where λσ is the saturation magnetization, μ0 is the magnetic permeability of va-

cuum and σ is given by the sum of the stresses induced during the microwires
production (σi) together with the stresses applied on the microwire by glass-
coating due to different thermal expansion coefficients of metallic nucleus and
glass coating (σa):
σ=σi+σa. (2)
As it was shown earlier, the frequency dependence of the switching field given by
eq.1 can be derived from the equation of domain wall motion and it is given by
equation [34]:
σ
H sw ~ const ( fH 0 ) 1 / n (3)
where n could range from 1 to 3, f is the frequency and H0 is the amplitude of the
exciting magnetic field used during measurement.
On the other hand, there exists another contribution that can be described in
terms of magnetic after-effect. As a result of their preparation, the structure of
amorphous wires is associated with the metastable state of the amorphous struc-
ture, and for this reason quite large relaxation effect can be expected. As the mea-
suring frequency decrease, the measuring time increase and a stabilization of the
domain structure through structural relaxation takes place. Such effect has already
been observed in amorphous glass-coated microwires [26][13].
For relaxation effects due to local structural rearrangements, the pinning field
r is considered:
H sw
1 ε pρ p
2
r
H sw (T ) ∝ G (T , f ) (4)
M S kT
where εp corresponds to the interaction energy of the mobile defects with sponta-
neous magnetization, ρp is the density of the mobile defects and relaxation func-
tion G(T,f) is given:
G(T,f) ~(1-exp(ν0/f)), (5)
being ν0 the relaxation frequency of the defects and f the frequency of a

measurement.
The two contributions given by eq. (3) and (4) can be simply summed giving
the final frequency dependence of the switching field:
σ
H sw = H sw + H r
sw
, (6)
which fits experimental data almost perfectly for wide variety of amorphous mi-
crowires within the range of 10-1000Hz [11][13].
300 Fe77.5Si7.5B15
250
200
Hsw (A/m)
150
100
σ
Hsw
50
r
Hsw
0
10 100 1000
f (Hz)
Fig. 13. Frequency dependence of the switching field measured at 300K. Lines represents
the fits according to eq.6
3.2 Temperature Dependence of the Switching Field

The two contributions to the switching field have also different temperature de-
pendence. This gives us possibility to play with the temperature dependence of the
switching field. Either we can play with the chemical composition or we can play
with the frequency. Figure 6 shows the temperature dependence of the switching
field for two microwires- Fe77.5Si7.5B15 and Fe36Co40Si13B11. The FeSiB-based
microwire has strong relaxation contribution that results in an increase of the
switching field at low temperature. Magnetoelastic contribution prevails in the
temperature range from 150 to 400 K. On the other hand, FeCoSiB alloys show
very low structural relaxation [32]. However, they are characterized by high mag-
netostriction. This results in a high switching field, which temperature dependence
is given mainly by magnetoelastic contribution [27].
FeSiB 3
Total (eq.6) FeCoSiB
3 Total (eq.6)
2
Magnetoelastic
Hsw (Oe)
Hsw (Oe)
Structural 1
relaxation
1
Structural relaxation
Magnetoelastic
0
50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500
T (K) T (K)
Fig. 14. Temperature dependence of the mean switching field for microwires with different
magnetoelastic and structural relaxation contribution to the switching field
The temperature dependence of the switching field can also be varied by ther-
mal treatment. Firstly, the thermal treatment results in the homogeneization of the
microwire’s structure and stress relief that leads to the decrease of magnetoelastic
contribution. On the other hand, the homogeneization of the sample leads to the
decrease of the mobile defect concentration and hence to the decrease of the struc-
tural relaxation contribution to the switching field. Moreover, special composition
(like FeCoMoB- based), leads to the partial crystallization which results in an
extremely stable and magnetically soft nanocrystalline structure. This is the way
how to stabilize the temperature dependence of the switching field. Figure 15
shows the temperature dependence of the switching field for thermally treated
amorphous (annealing temperature below 450oC) and nanocrystalline (annealing
temperature above 425oC) FeCoMoB-based microwires.
Fig. 15. Temperature dependence of the switching field for thermally treated amorphous
(left) and nanocrystalline (right) FeCoMoB microwire
Complex temperature dependence of the switching field for amorphous micro-

wires is given by a complex stress distribution annealed in during the microwire’s
production. Annealing above 300oC leads to the homogeneization of amorphous
structure that leads to almost linear dependence of the switching field in a quite
large range of temperatures. Annealing above 425oC leads to partial crystallization of
structurally very stable FeCo crystalline phase. This results in almost linear tempera-
ture dependence of a switching field that varies very slowly with temperature of
annealing.
1000 Hz
80
60
Hsw (A/m)
350 Hz
40
50 Hz
20
100 200 300 400 500
T (K)
Fig. 16. Temperature dependence of the switching field for three different frequencies for
amorphous FeNbSiB microwires
The next possibility is to use the combination of frequency and temperature de-
pendence of both contributions to the switching field to tailor the overall tempera-
ture dependence of the switching field. As given in fig.5, the structural relaxation
contribution to the switching field prevails at low frequency, whereas the magne-
toelastic contribution is much stronger at higher frequencies. Hence, we can vary
the relative strength of both contributions by simply changing the frequency of the
exciting field. Figure 16 shows the temperature dependence of the switching field
for FeNbSiB amorphous microwire measured at three different frequencies. At
low frequency, the structural relaxation contribution prevails (see also fig. 13) and
the temperature dependence of the switching fields looks like one given in fig. 14
left. At high frequency (1000Hz), magnetoelastic contribution prevails and the
temperature dependence looks like one for FeCoSiB (fig. 14 right). In the inter-
mediate frequencies (350 Hz), the relative strength of both contributions is compa-
rable and one can observe thalmost linear dependence of the switching field on
temperature
3.3 Stress Dependence of the Switching Field

The necessary condition for magnetic bistability to appear is positive magneto-
striction of microwire. Naturally, such microwires will be very sensitive to me-
chanical stresses applied on microwires. The stress dependence of the switching
field based on magnetoelastic properties is given by eq.1.
On the other hand, the stress dependence of the relaxation contribution is
represented by the stress dependence of the interaction energy of the mobile de-
fects with spontaneous magnetization εp (es given in eq.2). The interaction energy
εp is given by a sum of three terms: magnetoelastic (εσ), exchange (εex), and di-
pole-dipole contribution (εd) [14]:
εp = εσ + εex + εd . (7)
The most important in our case is the magnetoelastic contribution, εσ , that is given
by a local magnetoelastic energy:
εσ ∼ λsσ (8)
where σ denotes the local stress that surrounds the relaxing defects.
Finally, the stress dependence of the switching field is given by a sum of two
contributions given by eqs.1, and 4. Fig. 17 shows the stress dependence of the
switching fieldmeasured at different frequencies of excited magnetic field. At low
frequencies, the relaxation contribution is relatively high. Although, the magnetoe-
lastic contribution is responsible for stress dependence at low stress (and switch-
ing field is proportional to √σ). However, at higher stress (above 40 MPa),
relaxation contribution must be taken into account (switching field is proportional
to σ2). Increasing the measuring frequency to 200 Hz, the relaxation contribution
amplitude decreases (see fig. 13) and at 2 kHz the relaxation contribution is neg-
ligible (see also fig. 13) and the magnetoelastic contribution is the only that can
explain the stress dependence of the switching field according to eq.1.
As given in eq. 2, the total stresses introduced on the metallic core of microo-
wires are given by the suum of the stresses induced by drawing and stresses introo-
duced by glass coating. Similarly as temperature dependence of the switchinng
field, the stress dependencce can be tailored by thermal treatment, too.
250
2 kH z
200 H z
50 H z
200
HSW (A/m)
150
100
0 40 80 120 160
σ (M Pa)
Fig. 17. Stress dependencess of the switching field measured at various frequencies of thhe
applied magnetic field
4 Measurement of Physical
P Quantities Using Bistabile Magnetiic
Microwires
4.1 Measurement off External Magnetic Field

The domain structure of amorphous glass-coated magnetic microwire with possi-
tive magnetostriction consists of single large axial domain that occupies more thaan
90 % of the whole volumee in the wire (see fig. 9).
As a result, the hystereesis loops shows bistability: only two values of the magg-
netization (+Ms and -Ms) can be measured in the static state (see fig. 18).
Fig. 18. The hysteresis loops of microwire with positive magnetostriction exhibit magnettic
bistability. The switching beetween the only two magnetization values appears at the switch-
ing field
The switching between n the two states runs through the single Barkhausen jum
mp
at the field called switchiing field. The switching field is easy to be measured bby
the induction method- shaarp maximum appears at the end of the pick-up coil wheen
the domain wall propagatees through it (see fig. 19).
Fig. 19. The switching field

d can easily be measured by the induction method. Sharp maxxi-
mum appears at the end of th
he pick-up coil when switching appears
The advantage of the measuring

m system is that it is easy to be automated: Wheen
the exciting field has triaangular shape, the switching field is proportional to thhe
time t at which the maximmum appears.
Fig. 20. Using triangular sh

hape of exiting field, the switching field is proportional to thhe
time at which maximum app pears
Application of external magnetic field leads to the shift of whole hysteresis

loop (figure 20), which results in the shift of the switching field. This shift is also
proportional to the time t at which the switching field is measured by induction
method. However, the proportionality is valid in the infinite range of the field
from -∞ to +∞ A/m (e.g. there is no saturation, see fig. 21).
It is easy to construct 2D, 3D contact-less miniaturized sensor of magnetic field
based on bistabile microwires with enhanced range for measured magnetic field.
1500
1000
500
Hy [A.m ]
-1
-500
-1000
-1500
-800 -600 -400 -200 0 200 400 600

Δ T [ μ s]
Fig. 21. The external magnetic field Hy is proportional to the time at which the switching
appears. The linear dependence is valid in the range of external field from -∞ to +∞ A/m
4.2 Measurement of External Magnetic Field and Stress

or Temperature
For practical application of sensors based on bistabile magnetic microwires for
measurement of stress in materials, it was, in general, necessary to develop
a suitable method of measurement. From the beginning of the development, great
attention was paid to the fact that this method could be used for contact less
sensing.
In this case, the microwire is built into the measured structure, the exciting and
sensing components of the sensor must be solved as reading devices attached
from outside. In the suggested solution, measurement will be focused on the
external magnetic field H0 and switching field HC of the microwire, which by
the chosen characteristics of the thermo magnetically processed suitable magnet-
ic microwire which carries in itself the information on the measured physical
magnitude such as stress, temperature and the like.
Measurement is suggested to be solved through comparation of the moment of
voltage peaks generated by when the domain wall starts moving in the sensing
winding with the time duration of the given direction excitation on the stimula-
tion triangular signal of the amplitude HM. The carriers of information are times
T+ and T-, as output logical signals of the sensor (Fig. 22).
Fig. 22. Principle of the method of measurement, green color is marking the excitation,
red the band between positive and negative switching fields +/-Hc of the microwire, blue
– time intervals of the output logical signal obtained by comparation of the excitation and
the field +/- Hc . Ho denotes the external magnetic field
From the Figure 22 one can derive that the times T+ and T- are expressed with
the following expressions:
T+ =
T
(H C + H M − H 0 )
2H M
(9)
T− =
T
(H C + H M + H 0 )
2H M
Difference between these times gives us the external magnetic field H0:
T+ − T− = −
T (T − T )
H 0  H 0 = −H M + − (10)
HM T
This expression is independent on the hysteresis and hence on the switching field
HC and as a result, not on the microwire stress and temperature. For example, at a
selected excitation of the sensor and processing of its logical output signal by a
counter, i-e. at parameters: amplitude of excitation.
BM=100μT it is HM= 79A/m, half-period of excitation T=1ms and the coun-
ter with clock 100MHz , the theoretical definition obtained is as much as 1nT.
The sum will give us the value of the switching field, i.e. the magnitude of the
microwire as a function of stress (temperature) and the like:
 H   T + T− 
T+ + T− = T 1 + C   H C = H M  + − 1 (11)
 HM   T 
Now the expression is ind dependent of the external field H0 and enables definitioon
of the switching field, thus, it is the hysteresis of the microwire and the deriveed
physical magnitudes with the definition of 0,00079A/m.
4.3 Realisation of th
he Functional Model of the Sensor
and Experimentts
Experiences of co-authors - researchers from Technical university of Kosice, Fa-
culty of Aeronautics, Department
D of Aviation Technical Studies and EDIIS
Company and their dispo osal with electronics and measurement stands from thhe
previous and parallel running projects as APVV (Slovak Agency for Support oof
Research and Developmeent) with contract numbers 0454-07 (MAGSEN – magg-
netic sensors) and 0266-1
10 (SEMAMID – sensors based on magnetic microwirees)
assures study based on ex
xperiences and knowledge.
Fig. 23. Block diagram of th

he four-channel electronic unit and realized electronic unit
nctional sample of the contact-less sensor it was possible

When realizing the fun
to use the four-channel electronic
e unit of the primary signal processing designeed
for MAGSEN project, wh here it serves for the relaxation flux-gate sensors.
Fig. 24. Presentation of signnals in the oscilloscope screen for sensor in ambient field withh-
out stress: CH1 (yellow) chaaracteristics of the excitation current, CH2 (green) characteriis-
tics of voltage on the sensinng winding (left) Presentation of the some oscilloscope signaals
for sensor in ambient field and
a under stress 0,6MPa : CH1 (yellow) characteristics of thhe
excitation current, CH2 (greeen) characteristics of voltage on the sensing winding, We caan
see the time shift of signal peeks
p (right).
The concept and realization of the unit is presented in Figure 23. Real signals,
before processing in CPLD and IPC Vortex, are shown and commented in fig. 24.
The microwire developed dominantly for measurement of stress was used as a
sensor. It was a microwire with a diameter of 40 micrometers from material
FeSiBP thermo magnetically treated at 300°C/1hour, length of 2 cm, embedded in
hard rubber.
As we apply equations (10 and 11) for measurements according to Figure 24,
we get the results:
For stress-free measuring (times T are expressed in microseconds) we get
(T + − T− )  687 − 625 
H 0 = −H M = −H M   = −H M . 0 , 062 (13)
T  1000 
 T + T−   687 + 625 
HC = HM  + − 1 = H M  − 1 = H M .0,312 (14)
 T   1000 
Excitation field parameters are for HM equal to 300A/m, then the ambient field is
H0 = 18,6A/m (or 23,36 μ Τ) nand Hc = 93,6 A/m.
For measuring under stress 0,6 MPa (times T are expressed in microseconds)
(T + − T− )  812 − 750 
H 0 = −H M = −H M   = −H M . 0 , 062 (15)
T  1000 
 T + T−   812 + 750 
HC = HM  + − 1 = H M  − 1 = H M .0,562 (16)
 T   1000 
At the same excitation HM= 300 A/m the switching field is HC = 168,6 A/m.
It is clear from the results that the ambient magnetic field is equal in both cases.
But switching field of micro-wire in the tested rubber sample under stress is high-
er approximately about eighty percent than in stress free experiment. This gives
the assumption that after the calibration measurements (for the hard rubber sample
with an implanted magnetic micro-wire sensor) can make sufficiently sensitive
and accurate measurement with our contact-less sensor.
5 Results
The initial design of intelligent implant was realized with proposal of installation
holes for magnetic microwires. After analysis, titanium Ti64 appears as a best
material for implants fabrication together with DMLS technology, which allows
preparation of implants with specific shapes and sizes. The further studies will
deal with support analysis and preparation for DMLS and fixation mechsanism of
magnetic microwires will be analyzed.
During the introductory experiments with the contact less sensors, also other
magnetic microwires of different diameters and chemical composition and even
narrow amorphous strips were tested. The shape of the signal on the sensing winding
was changing as a function of the sample used, but in all cases, the sensor was
working reliably, a fact that is promising for application in inteligent implants.
Next studies will solve demand for increasing distance between magnetic mi-
crowire and the sensoric coil, which is recently 5 mm as a maximum. The aim is
to get distance 10 mm for reliable sensing process. The future of presented re-
search rests also in extension of scanned parameters from temperature and stress
to position or distance utilizable in implantology.
Acknowledgement. Presented paper was supported by project Research of New Diagnostic

Methods in Invasive Implantology, MŠSR-3625/2010-11, Stimuls for Reasearch and de-
velopment of Ministry of Education, Science, Research and Sport of the Slovak Republic)
and by the Slovak Agency for Research and Development under contracts APVV-0454-07
and 0266-10.
References
[1] Ainslie, K.M., Desai, T.A.: Microfabricated implants for applications in therapeutic
delivery, tissue engineering, and biosensing. Lab. Chip. 8, 1864–1878 (2008)
[2] Bidanda, B., Bártolo, P.: Virtual Prototyping & Bio Manufacturing in Medical Appli-
cations. Springer Science+Business Media, ISBN: 978-0-0387-33429-5
[3] Bistable amorphous and nanocrystalline FeCoMoB microwires. Acta Physica Poloni-
ca A 118, 809 (2010)
[4] Botsis, J., Humbert, L., Colpo, F., Giaccari, P.: Embedded fiber Bragg grating sensor
for internal strain measurements in polymeric materials. Opt. Lasers Eng. 43(3-5),
491–510 (2005)
[5] Cerny, M., Martinak, L., Penhaker, M., Rosulek, M.: Design and implementation of
textile sensors for biotelemetry applications. In: Proceedings 14th Nordic-Baltic Con-
ference on Biomedical Engineering and Medical Physics, NBC 2008, IFMBE Pro-
ceedings, IFMBE, vol. 20, Riga, June 16-June 20, pp. 194–197 (2008) ISSN:
16800737, ISBN: 9783540693666, doi:10.1007/978-3-540-69367-3-53
[6] Cerny, M., Penhaker, M.: Wireless body sensor network in Health Maintenance sys-
tems. Elektronika ir Elektrotechnika (9), 113–116 (2011) ISSN: 13921215
[7] Chiriac, H., Ovari, T.A.: Amorphous glass-covered magnetic wires: Preparation,
properties, applications. Prog. Mater. Sci. 40, 333 (1996)
[8] Cverha, A., Lipovsky, P., Hudak, J., Blazek, J., Praslicka, D.: Concept of magnetome-
ter with microwire probe. Acta Avionica (25), 22–25 (2011)
[9] Hudak, J., Blazek, J., Cverha, A., Gonda, P., Varga, R.: Improved Sixtus- Tonks
method for sensing the domain wall propagation direction. Sensor and Actuator A,
292–295 (2009)
[10] Klein, P., Varga, R., Vazquez, M.: Temperature dependence of magnetization process
in
[11] Klein, P., Varga, R., Vojtanik, P., Kovac, J., Ziman, J., Badini-Confalonieri, G.A.,
Vazquez, J.: Study of the switching field in amorphous and nanocrystalline FeCo-
MoB microwire. Phys. D: Appl. Phys. 43, 045002 (2010)
[12] Komová, E., Varga, M., Varga, R., Vojtanik, P., Torrejon, J., Provencio, M., Vaz-
quez, M.: Acta Physica Polonica A 113, 135 (2008)
[13] Komová, E., Varga, M., Varga, R., Vojtaník, P., Torrejon, J., Provencio, M., Vaz-
quez, M.: Frequency dependence of the single domain wall switching field in glass-
coated microwires. J. Physics: Condensed Matter 19, 236229 (2007)
[14] Kronmüller, H., Fahnle, M.: Micromagnetism and the Microstructure of the Ferro-
magnetic Solids. Cambridge Univ. Press (2003)
[15] Kubon, M., Moschallski, M., Link, G., Ensslen, T., et al.: A Microsensor System to
Probe Physiologicla Environments and Tissue Response, Sensors. In: 2010 IEEE,
IEEE Sensors 2010 Conference, pp. 2607–2611 (2010)
[16] Lestari, W., Qiao, P., Hanagud, S.: Curvature mode shape-based damage assessment
of carbon/epoxy composite beams. J. Intell. Mater. Syst. Struct. 18(3), 189–208
(2007)
[17] Libermann, H., Graham, C.: Production of amorphous alloy ribbons and effects of
apparatus parameters on ribbon dimensions. IEEE Trans. Magn. MAG-12(6),
921–923 (1976)
[18] Mouzakis, D., Dimogianopoulos, E., Giannikas, D.: Contact-Free Magnetoelastic
Smart Microsensors With Stochastic Noise Filtering for Diagnosing Orthopedic Im-
plant Failures. IEEE Transactions on Industrial Electronics 56(4) (April 2009)
[19] Narayan, R.: Biomedical Materials. Springer Science+Business Media, LLC (2009)
ISBN: 978-0-387-84871-6
[20] Olivera, J., Sanchez, J.L., Prida, V.M., Varga, R., Zhukova, V., Zhukov, A., Hernan-
do, B.: Temperature Dependence of the Magnetization Reversal Process and Domain
structure in Fe77.5-xNixSi7.5B15. Magnetic Microwires IEEE Trans. Magn. 44,
3946 (2008)
[21] Oshida, Y.: Bioscience and Bioengineering of Titanium Materials, 1st edn. Elsevier
(2001) ISBN-13:978-0-08-045142-8
[22] Sabol, R., Varga, R., Blazek, J., Hudak, J., Praslicka, D., et al.: Temperature and fre-
quency dependences of the switching field in glass-coated FeNbSiB microwires. In:
SMM 2011, pp. S03–P237 (2011)
[23] Sabol, R., Varga, R., Blazek, J., Hudak, J., Praslicka, D., et al.: Stress dependence of
switching field in glass-coated microwires. In: ANMM 2011 – Amorphous and Na-
nostructured Magnetics Materials, p. 5 (2011)
[24] Skapa, J., Látal, J., Penhaker, M., Koudelka, P., Hancek, F., Vasinek, V.: Optical fi-
ber distributed temperature sensor in cardiological surgeries. In: Proceedings of SPIE
- The International Society for Optical Engineering, vol. 7726, art. no. 77261V, Spon-
sor: The Society of Photo-Optical Instrumentation Engineers (SPIE); B-PHOT-
Brussels Photonics Team; Brussels-Capital Region; Fonds Wetenschappelijk Onder-
zoek (FWO); International Commission for Optics (ICO); Ville de Bruxelles, April
12-April 15, Brussels (2010) ISSN: 0277786X, ISBN: 9780819481993, doi:
10.1117/12.854309
[25] Varga, R., Garcia, K.L., Vazquez, M., Zhukov, A., Vojtanik, P.: Switching-field dis-
tribution in amorphous magnetic bistable microwires. Physical Review B 70, 024402
(2004)
[26] Varga, R., Garcia, K.L., Zhukov, A., Vazquez, M., Vojtanik, P.: Temperature depen-
dence of the switching field and its distribution function in Fe-based bistable micro-
wires. Appl. Phys. Lett. 83, 2620 (2003)
[27] Varga, R., Garcia, K.L., Luna, C., Zhukov, A., Vojtanik, P., Vazquez, M.: Distribu-
tion and temperature dependence of switching field in bistable magnetic amorphous
microwires. Recent Research Development in Non-Crystalline Solids 3, 85 (2003)
Ed. Transworld Research Network, ISBN: 81-7895-090-1
[28] Varga, R., Zhukov, A., Blanco, J.M., Ipatov, M., Zhukova, V., Gonzalez, J., Vojtaník,
P.: Supersonic domain wall in magnetic microwires. Physical Review B 76, 132406
(2007)
[29] Varga, R., Zhukov, A., Ipatov, A., Blanco, J.M., Gonzalez, J., Zhukova, V., Vojtaník,
P.: Thermal activation over a complex energy barrier in bistable microwires. Physical
Review B 73, 053605 (2006)
[30] Vazquez, M.: Advanced magnetic microwires. In: Kronmüller, H., Parkin, S. (eds.)
Handbook of Magnetism and Advanced Magnetic Materials, p. 221. John Wiley &
Sons (2007)
[31] Vazquez, M., Badini, G., Pirota, K., Torrejon, J., Zhukov, A., Torcunov, A., Pfuetz-
ner, H., Rohn, M., Merlo, A., Marquardt, B., Meydan, T.: Magnetization reversal
process in bistable microwires and its temperature dependence. Int. J. Appl. Electro-
magnetics and Mechanics 25, 441 (2007)
[32] Vojtanik, P., Degro, J., Nielsen, O.V.: Magnetic Aftereffects in CoFeSiB Metallic
Glasses. Acta Phys. Slov. 42, 364 (1992)
[33] Zhukov, A., Gonzalez, J., Vazquez, M., Larin, V., Torcunov, A.: Nanocrystalline and
amorphous magnetic microwires. In: Nalwa, H.S. (ed.) Encyclopedia of Nanoscience
and Nanotechnology, ch. 62, p. 365. American Scientific Publishers, New York
(2004)
[34] Zhukov, A., Vázquez, M., Velázquez, J., Garcia, C., Valenzuela, R., Ponomarev, B.:
Frequency dependence of coercivity in rapidly quenched amorphous materials. Mate-
rials Science and Engineering A 226-228, 753 (1997)
Author Index
Ádám, Norbert 277 Karol’, Tomáš 85

Adamčı́k, František 185 Kártyás, Gyula 135
Ambrus-Somogyi, Kornélia 125 Kat’uch, Peter 335
Andoga, Rudolf 85, 185 Kelemen, Michal 151
Klempous, Ryszard 167
Bakó, András 125 Kracmar, Jan 307
Bednarčı́ková, Lucia 393 Krajňák, Peter 185
Blažek, Josef 413 Krokavec, Dušan 3
Bučko, Marián 135
Madarász, Ladislav 85, 185
Čekanová, Patrı́cia 375
Madoš, Branislav 277
Chaczko, Zenon 167
Medvecký, Štefan 375
Dragoş, Claudia-Adina 25 Michalı́ková, Monika 393
Mihal’, Roman 51
Enache, Florin-Cristian 25
Nikodem, Jan 167
Filasová, Anna 3 Nikodem, Maciej 167
Főző, Ladislav 85
Fullér, Robert 249 Paleček, Juraj 265
Paralič, Ján 207
Gašpar, Vladimı́r 85
Penhaker, Marek 307
Gáti, József 135
Petriu, Emil M. 25
Gmiterko, Alexander 151
Prada, Erik 151
Goban, Bruno 335
Praslička, Dušan 413
Harmati, István Á. 249 Precup, Radu-Emil 25
Hermann, Gyula 39 Preitl, Stefan 25
Hladký, Vratislav 3 Prokop, Lukas 307
Hlavackova, Milada 307
Horváth, László 105 Rădac, Mircea-Bogdan 25
Hošák, Rastislav 51 Rövid, András 221
Hudák, Jozef 413 Rudas, Imre J. 105
Hudák, Radovan 335, 355, 375, 413
Sikora, Tadeusz 307
Jasminská, Natália 63 Spalek, Juraj 265
436 Author Index
Surovec, Róbert 151 Várkonyi-Kóczy, Annamária R. 233

Szeidl, László 221 Várlaki, Péter 221, 249
Vavra, Petr 307
Tick, Andrea 195 Vestenický, Martin 265
Tick, József 293 Vestenický, Peter 265
Tkáčová, Mária 355, 375 Virgala, Ivan 151
Tóth, Teodor 393 Vokorokos, Liberios 277
Tutoky, Gabriel 207
Živčák, Jozef 335, 355, 375, 393, 413
Vacková, Martina 151 Živčáková, Ivana 355
Varga, Rastislav 413 Zolotová, Iveta 51

10.1007/978 3 642 30668 6

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

10.1007/978 3 642 30668 6

Uploaded by

Copyright:

Available Formats

Topics in Intelligent Engineering

Editorial Advisory Board

For further volumes:

ISSN 2193-9411 e-ISSN 2193-942X

Library of Congress Control Number: 2012939521

c Springer-Verlag Berlin Heidelberg 2013

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

The International Conference on Intelligent Engineering Systems (INES) took place

continuous-time nonlinear systems represented by Takagi-Sugeno models; applica-

a full-time PhD student of the Department of Cybernetics and Artificial Intelligence

April 12, 2012 Ladislav Madarász

Part I: Diagnostics, Performance Monitoring and

Part II: New Approaches in Mechatronics, Modeling and

Usage Dependent Rehabilitation and Maintenance Model for Special

Part III: Computational Intelligence, Web Technologies and

Part IV: Methodologies, Methods and Algorithms

Potential Application of P-Graph-Based Workflow in Logistics . . . . . . . . 293

Part V: Softcomputing in Biomedical Engineering

Determining the Calibration Element Location in Planning of

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435

Dušan Krokavec, Anna Filasová, and Vratislav Hladký

Technical University of Košice, Faculty of Electrical Engineering and Informatics,

Abstract. The paper presents the principle of fuzzy-observer-based residual gene-

Keywords: Fault detection, residual generators, fuzzy observers, convex optimi-

L. Madarász et al. (Eds.): Aspects of Computational Intelligence, TIEI 2, pp. 3–23.

q (t ) = a(q (t )) + Bu(t ) , (1)

put, output, and objective variables, respectively, and B ∈  n× r , C ∈  m× n , and

 w lj (θ (t )), j = 1, 2, , k , l = 1, 2, , p,

where k is the number of sector functions, and

Ωh j = Aq(t ) + Ah j q(t ) + Bu(t ) (7)

w1 (θ (t )) = w11 (θ1 (t )) w p1 (θ1 (t )) ,

Constructing the function set {w (θ (t )), i = 1, 2, s, s = 2 }

Ωi = ( A + Ai ) q(t ) + Bu(t ) (10)

is the i-th aggregated normalized membership function satisfying conditions

and can be interpreted as a combination of s linear sub-models through the set of

T > 0, U = U T > 0, V = V T > 0 (17)

Y11 Y12  Y1s 

Hereafter, * denotes the symmetric item in a symmetric matrix.

Writing (13) in the form

v(q(t )) = qT (t ) Pq(t ) > 0 (24)

where P ∈  n× n is a symmetric positive definite matrix, then it yields

v(q(t )) = qT (t ) Pq(t ) + qT (t ) Pq(t ) < 0 (25)

where {X ij = X ijT ∈  n× n , i, j = 1, 2, , s} is the set of symmetric matrices. Now

Since S1, S2 are regular, defining the congruence transform matrix

T1 =  S1−1 S2−1  (30)

then pre-multiplying left-hand side and right-hand side of (29) by T1 gives

and using the substitutions

(31) implies (18). Since evidently (26) is conditioned by

respectively, then (34) implies (19). This concludes the proof.

4 Fuzzy Observer with Measurable Premise Variables

where qe (t ) ∈  , and {J i ∈  n× m , i = 1, 2, , s} is the set of the gain matrices.

( A + Ai ) P + P ( A + Ai ) − ZiT C T − Zi C < 0 ∀i. (38)

Then Ji can be computed as

respectively, where the observer system matrices are introduced as

Aei = A + Ai − J i C , i = 1, 2, , s . (43)

v(e(t )) = eT (t ) Pe(t ) (44)

v(e(t )) = eT (t ) Pe(t ) + eT (t ) Pe(t ) . (45)

v(e (t )) = eT (t )  hi (θ (t )) ( PAei + AeiT P ) e (t ) ,

5 Fuzzy Observer with Un-measurable Premise Variables

5.1 Basic Preliminaries

if γ > 0 is square of the infinity norm of the transfer matrix Gzv ( s ) .

Proof. It is evident that if