!computer Applications in Electrical Engineering - 2011 - 9

COMPUTER APPLICATIONS
IN ELECTRICAL ENGINEERING
POZNAN UNIVERSITY OF TECHNOLOGY
INSTITUTE OF ELECTRICAL ENGINEERING
AND ELECTRONICS
under the auspices of
ELECTRICAL ENGINEERING COMMITTEE
OF POLISH ACADEMY OF SCIENCES
and IEEE POLAND SECTION
COMPUTER APPLICATIONS
IN ELECTRICAL ENGINEERING
VOLUME 9
Edited by
Ryszard Nawrowski
Poznan University of Technology
Published by ARP Promocja21 Sp. z o.o.

Poznan 2011
COMPUTER APPLICATIONS IN ELECTRICAL ENGINEERING
Chairman of the Programming Committee

Prof. Stanisław Bolkowski, DSc
Chairman of the Organising Committee
Prof. Ryszard Nawrowski, DSc
Members of the Organising Committee
Prof. Wojciech Machczyński, DSc
Prof. Konrad Skowronek, DSc
Krystyna Horemska, MSc
Leszek Kasprzyk, PhD
Scientific Secretary of the Conference
Andrzej Tomczewski, PhD
Organising Secretary of the Conference
Dorota Warchalewska-Hauser, MSc
Jarosław Jajczyk, PhD
Reviewers
Arkadiusz Dobrzycki Lech Nowak
Konrad Domke Andrzej Odon
Ireneusz Grządzielski Władysław Opydo
Michał Gwóźdź Ryszard Porada
Jarosław Jajczyk Konrad Skowronek
Leszek Kasprzyk Zbigniew Stein
Wojciech Lipiński Wojciech Szeląg
Ryszard Nawrowski Andrzej Tomczewski
Secretary of the Conference

Poznan University of Technology, Institute of Electrical Engineering and Electronics
60-965 Poznań, Piotrowo 3a
tel. (061) 6652388, fax. (061) 6652389
e-mail: zkwe@put.poznan.pl, http://www.iee.put.poznan.pl/zkwe/
ISSN 1508-4248
ISBN 978-83-86912-59-9
Copyright by Promocja21
Agencja Promocyjno-Reklamowa Sp. z o.o.
ul. Dolina 7A, 61-551 POZNAŃ
All rights reserved
Poznań 2011
Contents
Preface
I. Doležel, V. Kotlan, B. Ulrych
Electromagnetic friction clutch for fixing accurate position of sliding
elements 1
J. Purczyński
Application of trapezoidal integration method in determining discrete time
windows 12
T. Rymarczyk, S. F. Filipowicz, J. Sikora
Comparing methods of image reconstruction in electrical impedance
tomography 23
Z. Piątek, T. Szczegielniak, D Kusiak
Influence of the external proximity effect on the power losses in flat three
phase high current busduct 34
K. Kluszczyński, T. Trawiński, M. Szczygieł, E. Boeren, A. Skibniewski
Software for quality evaluation of a PS plate edge 45
M. Wołoszyn, J. Łubkowski, M. Chomnicki,
Naval mine detection system based of FPGA circuit 60
K. Budnik, W. Machczyński
Modeling magnetic field in the vicinity of single wire helix 69
K. Jakubiuk, P. Zimny, M. Wołoszyn
Multisource model of ship electric field 78
J. Tchórzewski
Searching of regularities in the development of electrical power system.
Part I – Identification and evaluation 85
J. Tchórzewski
Searching of regularities in the development of electrical power system.
Part II – Analysis and interpretation of changes 102
R. Frąckowiak, T. Gałan
Electrical energy consumption by the household – type consumers 112
M. Wesołowski, R. Niedbała, D. Kucharski, J. Hauser
Numerical modeling of dynamic heating processes 120
M. Łukaniszyn, M. Kowol, J. Kołodziej
An analysis and verification of a dynamical model of a transverse flux motor 129
L. Nowak, K. Kowalski, D. Stachowiak, J. Mikołajewicz, W. Pietrowski,
K. Radziuk
Determining the optimal working point and the loading characteristic of
synchronous machine for fan and pump drive system 139
J. Wiśniewski
Influence of disturbances in power network on torsional torques in big
power turbines-generator shafts 150
L. Kasprzyk, Z. Stein, M. Zielińska
Basic equations for analysis of a three-phase three-winding transformer
subjected to asymmetric load 161
P. J. Serkies, K. Szabat
Model predictive control of two-mass drive system with mechanical
backlash 170
J. Walczak, R. Stępień
Discrete models of the NLFSR generators 181
M. Gwóźdź
Sub-phase power electronics inverters 190
K. Okarma
Assessment of the spatio-temporal vector median filtering algorithms using
colour video quality metrics 200
E. Kornatowski
Sound recording with the application of microphone arrays 213
P. Mazurek
Analysis of noise suppression techniques for Track-Before-Detect
algorithms using spatial noise estimators 223
A. Jastiebow, G. Słoń
Parametric identification of dynamic imprecise models based on relational
fuzzy cognitive maps 234
R. Jędrychowski
The significance of wireless communication for the metering data
transmission via Smart Grids 245
D. Łuczak
Remote laboratory with WEB interface 257
AUTHORS INDEX 269
Preface
The publication includes contents of selected lectures delivered during the
debates of the Scientific Conference on Computer Application in Electrical
Engineering that was held in Poznan on April 11-13, 2011.
The Institute of Electrical Engineering and Electronics of the Poznan University
of Technology organized the Conference on Computer Application in Electrical
Engineering for the 16th time. The first Conference was held in 1996 and, since that
time, has been held every year. Total number of 2742 lectures have been published
from 1996 to 2011. During the past twelve years about 2940 persons participated to
the Conferences, inclusive of the workers of universities, research centres, and
industry, also from Czech, Germany, Romania, and Ukraine.
The Conference is aimed at presenting the applications of existing computer
software and original programs in the field of modelling, simulation,
measurements, graphics, databases, and computer-aided scientific and engineering
works related to electrical engineering.
The following thematic groups are foreseen:
1. ELECTRICAL ENGINEERING
a. Electromagnetic field, electromagnetic compatibility
b. Theory of circuits and signals
c. Bioelectromagnetism
d. Power engineering, renewable energy
e. Electronics and power electronics
f. Electrical engineering of vehicles
g. Electrical heating
h. Electrical machines, electrical drive
i. Materials technology
j. Mechatronics
k. Electrical and electronic metrology
l. Microprocessor technology and control systems
m. Lighting technology
2. COMPUTER APPLICATION IN ENGINEERING
a. Databases
b. Numerical methods and algorithms
c. Parallel computation
d. Optimisation, genetic algorithms, neural networks and expert systems
e. Computer networks, information networks
f. Programming techniques
g. Internet technologies
h. Visualization, CAE/CAD Systems
3. DIDACTICS, EDUCATION AND SCIENTIFIC INFORMATION
4. SCIENTIFIC STUDENTS TEAM
Detailed information related to the ZKwE Conference are available on
http://www.iee.put.poznan.pl/zkwe/.
Computer Applications in Electrical Engineering
Electromagnetic friction clutch for fixing accurate position

of sliding elements
Ivo Doležel
Czech Technical University
166 27 Praha 6, Technicka 2, e-mail: dolezel@fel.cvut.cz)
Václav Kotlan, Bohus Ulrych
University of West Bohemia
306 14 Plzen, Univerzitni 26, e-mail: vkotlan@kte.zcu.cz, ulrych@kte.zcu.cz)
A complete mathematical model of an electromagnetic friction clutch is presented,

whose numerical solution provides its basic operation characteristics. The clutch represents
a part of various devices working with the movable plunger and its purpose is to fix it in a
prescribed accurate position. Its application is illustrated by a typical example.
1. Introduction
A lot of technical applications (linear motors, various actuators and other

devices, see, for example [1–2]) requires fixing of their movable elements (for
example, a plunger of a thermoelastic actuator) at a prescribed position x , see Fig.
1. This position of plunger 1 from a given interval a, b is controlled by s
photoelectric sensor 4. A supplementary condition is that the axial pulse of force
Fa during the process of fixing is as low as possible. On the other hand, the
perpendicular magnetic force Fm should be as high as possible in order to secure
the fast fixation of the movable element 1. This can be realized by a magnetic
friction clutch 3 with a very low mechanical inertia.
Fig. 1. Fixing of prescribed position of movable element by friction clutch:

1–movable element (for example the plunger of a thermoelastic actuator),
2–device generating the movement of element 1 (for example a thermoelastic actuator),
3–friction clutch, 4–photoelectric sensor
1
I. Doležel, V. Kotlan, B. Ulrych / Electromagnetic friction clutch…
The paper deals with the technical description of such a friction clutch and its
basic operation properties. i.e., produced magnetic and friction forces and
temperature rise (that must not exceed the prescribed value). It is organized in the
following manner: first we present the structural and physical formulations of the
problem and then its mathematical model whose solution provided a number of
results. The crucial part of the paper contains the selected results and their
discussion together with final evaluation. Attention is mainly paid to the
circumstances affecting the basic properties of the clutch – field current, materials
and geometrical configuration (ratio between the length and diameter of the
clutch). These data form the basis for its eventual subsequent optimization.
2. Structural and physical formulation of the problem
The starting structural arrangement (geometrical configuration) of the
considered clutch is depicted in Fig. 2. The clutch represents, in fact, a simple
electromagnetic actuator whose ferromagnetic plunger 2 and shell 3.1 are drilled
through by coaxial holes perpendicular to the axis of the core.
Fig. 2. Arrangement of the considered friction clutch:

1–field coil, 2–core (carbon steel 12 040) with a hole, 3.1 and 3.2–ferromagnetic shells (carbon
steel 12 040), 4–leading shell of the movable core, 5–braked movable element (for example plunger
of the thermoelastic actuator), 6–fixing ring, 7–front of the device producing the movement (for
example thermoelastic actuator), AB, EF – contact surfaces between the movable element 5
and shell 3.1, CD –contact surface between the movable element 5 and core 2
These holes are covered by friction fillings made of metal-ceramic friction

material Diafrikt K5 [3] of friction coefficient f ≈ 0.5 . The movable element 5
goes through the fillings and is (in the starting position) freely braked by them. At
the moment when the field coil 1 is connected to the source of direct current I ext ,
the core 2 is drawn into it by electromagnetic force Fm . But the element 5 makes it
impossible to continue this process of drawing and starts to act on the core 2 (in the
area CD ) by the same reaction force. And the same force appears between the
2
element 5 and shell 3.1 in the areas AB and EF . These forces acting in the direction
of the normal with respect to the corresponding areas (with high accuracy Fm ≈ Fm,n )
generate the friction force Ff,a = f Fm,n acting in the axial direction and preventing
these areas from any mutual movement. In this way the resultant friction force fixes the
position of element 5 with respect to the shell 3.1 of the friction clutch.
The values of physical parameters of particular elements of the clutch are given
in Tab. 1 and Figs. 3 and 4. These values were used for all computations whose
results are presented and discussed in sections 4 and 5.
3. Mathematical model of the problem
The mathematical model of the clutch consists of two partial differential
equations describing the distribution of the magnetic and temperature fields.
The stationary magnetic field in the clutch is expressed in terms of the magnetic
vector potential A and obeys the equation [6]
1
curl curl A = J ext , (1)
μ
where symbol μ stands for the magnetic permeability and J ext is the field current
density. The field produces magnetic force Fm acting between the feromagnetic
core 2 and shell 3.2 (see Fig. 2) whose value is expressed by the integral
1
Fm =
2 ∫ [ H (n ⋅ B )
Ń + B (n ⋅ H ) − n (H ⋅ B ) ] dS , (2)
S2
where H and B are vectors of the magnetic field and n denotes the unit vector
of the outward normal to the surface S 2 . Finally, integration is performed over the
whole core 2 of the clutch.
The force Fm then produces the axial friction force Ff,a between the clutch
and braked movable element 5, whose value is
Ff,a = f Fm . (3)
The stationary temperature field in the clutch is described by the equation [7]
div ( λ grad T ) = − wJ , (4)
where λ stands for the thermal conductivity, and wJ denotes the specific ohmic
losses produced in the field coil given by the formula
2
J ext
wJ = , (5)
γ
γ being the electrical conductivity of material of the coil (see Tab. 1).
3
Table 1. Physical parameters of particular elements of the friction clutch
Element Material Parameter Value Unit

1 field coil Cu [5] diameter of conductor Dv 1 mm
length of coil Δz 30 mm
thickness of coil Δr 7 mm
—
number of turns Nz 210
filling coefficient κ 0.785
—
relative permeability μr 1
thermal conductivity* λ 306.1 W/m oC
electrical conductivity* γ 4.474×107 S/m
2 ferromagnetic core steels: [4] curve B(H) see Fig. 3
3.1
3.2 ferromagnetic shell CSN 11 375 thermal conductivity λ see Fig. 4
CSN 12 040 electrical conductivity γ 4.5×106 S/m
CSN 42 2709
—
4 leading shell Teflon [5] relative permeability μr 1
thermal conductivity λ 1.6 W/m oC
—
5 movable element aluminium [5] relative permeability μr 1
electrical conductivity γ 3.5× 107 S/m
thermal conductivity λ 229 W/m oC
—
6 fixing ring Teflon [5] relative permeability μr 1
thermal conductivity λ 1.6 W/m oC
* respected is the filling coefficient κ = 0.785.
2.0
1.5
B [T]
1.0 I.
II.
0.5 III.
0.0
0 5 H [103 A/m] 10 15
Fig. 3. Magnetization curves of particular steel materials:

I – steel CSN 11375, II – steel CSN 12 040, III – cast steel CSN 42 2709
4
60
50
40 I.
λ [W/mK]
II.
30
III.
20
10
0
0 1 2 3 4 5 6 7
2
T [10 °C]
Fig. 4. Temperature dependencies of thermal conductivities of particular steel materials: I – steel
CSN 11375, II – steel CSN 12 040, III – cast steel CSN 42 2709
The model is solved in the axisymmetric arrangement. Unambiguousness of the

solutions of (1) and (4) is secured by correct boundary conditions. For (1) we apply
the condition of antisymmetry along the axis of the field coil and the Dirichlet
condition of the zero magnetic vector potential along an artificial boundary
approximating the situation in infinity. As for equation (4), because of a relatively
low temperature of the ferromagnetic shell of the clutch we can neglect the
influence of radiation, so that we apply there only the condition of convection of
heat into ambient air.
4. Computer model
The system of equations (1) and (4) was solved by the professional commercial
code QuickField [8] as a nonlinear 2D coupled problem. The selected results were
then validated by our own fully adaptive code Agros2D [9] that is based on finite
elements of higher order of accuracy.
Selected results are presented in
- Figure 5 that depicts a typical distribution of magnetic field in the clutch. It is
obvious that magnetic circuit of the device is proposed correctly – it is
characteristic by practically no undesirable leakage of this field.
- Figure 6 that shows the corresponding steady-state temperature field in the
clutch. The metal parts are heated relatively uniformly (and, consequently, there
is no danger of undesirable thermoelastic strains and stresses), but the
temperature is rather high. This confines the current loadability of the system
and also the force effects of the clutch.
In the course of both electromagnetic and thermal computations we carefully
checked the convergence of the numerical solution. Some relevant data given in
Tab. 2 provide a good idea about the reached accuracy.
5
Fig. 5. Magnetic field distribution in the system (plunger and shell are made of steel CSN 12 040):
δ = 1 mm, J ext = 2.5 ×106 A/m2, scale 2 × 10−5 Wb
Fig. 6. Steady-state temperature field distribution in the system (plunger and shell are made of
steel: CSN 12 040): δ = 1 mm, J ext = 2.5 × 106 A/m2, Tmin = 120 °C, Tmax = 122 °C,
ΔT = 0.06 °C
Table 2: Convergence of the numerical solution (plunger and shell are made of steel CSN
12 040): J ext = 2.5 × 106 A/m2, wJ,coil = 1.388 × 106 W/m3
Electromagnetic field
nodes hmax hmin Fm
mm mm N
1 918 3 1 32.633
7 385 1.5 0.5 31.717
29 260 0.75 0.25 31.625
119 724 0.375 0.125 31.570
Temperature field
nodes h Tmax Tcoil,avrg
o o
mm C C
1 383 1 122 121.70
4 918 0.5 122 121.63
19 304 0.25 122 121.59
6
5. Results of illustrative computations and their discussion
The crucial part of the paper is a set of results of particular computations (and
their discussion) whose aim is to evaluate
- available materials for manufacturing the magnetic circuit,
- geometrical configuration of the magnetic circuit (ratio of its length and
diameter).
Another aim of the computations is to find the highest possible current density
J ext in the field coil, for which the maximum temperature of the coil Tmax,coil
would not exceed 150 °C while the friction force Ff,a would reach the maximum.
5.1. Evaluation of available materials for the magnetic circuit
For the initial geometrical configuration of the clutch (see Fig. 2) this evaluation
is performed in Figs. 7 and 8.
200
175
150
150
T avrg,coil [°C]
125
100 I. ≈ II. ≈ III.
75
50
25
8.84
0
0 2 4 6 8 10
6 2
J ext [10 A/m ]
Fig. 7. Dependence of the average temperature Tavrg,coil of the field coil on external field current
density J ext and material of the magnetic circuit ( δ = 1 mm):
I – steel CSN 11375, II – steel CSN 12 040, III – cast steel CSN 42 2709
7
160
139.5
140
120
100
F m [N]
80 I.
II.
60 III.
40
20
8.84
0
0.0 2.5 5.0 7.5 10.0
6 2
J ext [10 A/m ]
Fig. 8. Dependence of the magnetic force Fm on external field current density J ext and material
of the magnetic circuit ( δ = 1 mm): I – steel CSN 11375, II – steel CSN 12 040,
III – cast steel CSN 42 2709
Figure 7 indicates that the average temperature of the coil Tavrg,coil practically
does not depend on the material of its core and shell of the clutch. The reason is
that that their thermal conductivities (see Fig. 4) are of the same order. This
temperature, however, strongly grows with the field current density J ext . The
highest permissible temperature Tmax,coil = 150 °C is reached for field current
density J max,ext ≈ 8.84 ×106 A/m2. Theoretically, we should expect the highest
temperatures in case of the cast steel CSN 42 2709 (see Fig. 4). Nevertheless, very
detailed computations proved that this growth (in comparison with other materials)
does not exceed about 3 °C even in case of the highest possible current.
Figure 8 shows that the magnetic force Fm also strongly depends on the field
current density J ext , while the dependence on the materials used is much less
significant. This is caused by their similar magnetization curves (see Fig. 3). But
the results lead to the conclusion that for field current density
J max,ext ≈ 8.84 ×106 A/m2 the most advantageous is the classical carbon steel
CSN 12 040, for which we obtain Fm ≈ 139.5 N and, consequently, Ff,a ≈ 70 N.
5.2. Evaluation of geometrical configuration of the clutch
This evaluation is carried out in Figs. 9 and 10. The particular geometrical
configuration is expressed by the coefficient
8
l1,real l2,real
kG = = ,
l1,start l2,start
where (see Fig. 2) l1,start = 30 mm is the length of coil 1 in the starting geometrical
configuration of the clutch, l2,start = 40 mm is the length of the core 2 in the same
configuration, l1,real is the length of coil 1 in the newly investigated geometrical
configuration and l2,real is the length of the core 2 in this new configuration. The
volumes V1 and V2 of both these parts, however, remain the same.
Fig. 9. Dependence of the average temperature Tavrg,coil of the field coil on external field current
density J ext and geometrical configuration expressed by coefficient kG ( δ = 1 mm, steel CSN
11375): I – kG = 0.8 , II – kG = 1.0 , III – kG = 1.2
Figure 9 shows that the average temperature of the coil Tavrg,coil again
practically does not depend on the geometrical configuration of the body of clutch.
This temperature, however, similarly as in the previous case, strongly depends on
the field current density J ext . The highest permissible temperature Tmax,coil = 150
°C is (for material CSN 12 040) reached for J max,ext ≈ 8.93 × 106 A/m2.
Theoretically, we should expect higher temperatures in case of shorter body of the
clutch ( kG = 0.8 ). The cooling surface of the clutch shell is somewhat smaller in
this case, while the volume of the coil remains unchanged, so that the Joule losses
produced in it are the same. Nevertheless, detailed computations proved that this
9
growth (in comparison with the case kG = 1.2 ) does not exceed about 1 °C even in
case of the highest possible field current.
225
193
200
175
150
F m [N]
125 I.
100 II.
III.
75
50
25
8.93
0
0.0 2.5 5.0 7.5 10.0
6 2
J ext [10 A/m ]
Fig. 10. Dependence of the magnetic force Fm on external field current density J ext and
geometrical configuration expressed by coefficient kG ( δ = 1 mm, steel CSN 11375):
I – kG = 0.8 , II – kG = 1.0 , III – kG = 1.2
Figure 10 indicates that the magnetic force Fm strongly depends not only on
the field current density J ext , but also on the value of kG . A shorter body of the
clutch ( kG = 0.8 ) with greater diameter is from the viewpoint of force Fm (for a
given value J ext ) more advantageous. This is obviously caused by the fact that the
front of the core 2 (its diameter is higher), where the substantial part of the force is
produced, is greater (see also eq. (2)). For field current density
J max,ext ≈ 8.93 ×106 A/m2 corresponding to Tmax,coil = 150 °C, and for classical
carbon steel CSN 12 040 we obtain Fm ≈ 193 N and, consequently,
Ff,a ≈ 96.5 N.
6. Conclusion
The paper is devoted to one of possible arrangements of an axial

electromagnetic friction clutch and its basic operation characteristics. The results of
computations show that for the proposed configuration of the clutch the required
friction force Ff,a :
10
- does not practically depend on the material of the clutch – the clutch can be
manufactured of any type of construction steel,
- strongly depends on the geometrical configuration of the clutch (ratio of its
length and diameter at the same volume of material) – a shorter body of the
clutch of higher diameter is more advantageous from the viewpoint of force
Ff,a than a longer body of smaller diameter.
But as this geometrical configuration of the clutch is not so good from the
viewpoint of temperature rise of the clutch, there is a possibility of optimizing its
geometrical configuration, for example, from the simultaneous viewpoints of axial
friction force Ff,a and temperature Tavrg,coil of the coil producing the magnetic
field.
Acknowledgment
The financial support of the Grant Agency of the Czech Republic (project No.
102/09/1305), Research Plan MSM 6840770017 and project KONTAKT MEB051041 is
gratefully acknowledged.
References
[1] Doležel, I., Karban, P., Ulrych, B., Pantelyat, M., Matyukin, Y., Gontarowskiy, P.:
Numerical Model of a Thermoelastic Actuator Solved as a Coupled Contact Problem.
COMPEL 26 (2007), No. 4., pp.1063–1072.
[2] Doležel, I., Kotlan, V., Krónerová, E., Ulrych B.: Induction Thermoelastic Actuator
with Controllable Operation Regime. COMPEL 29 (2010), No. 4, pp. 1004–1014.
[3] www.diafrikt.cz, metal-ceramic friction material K5.
[4] Company standard ŠKODA 00 6004.
[5] www.jahm.com, database of material parameters and their temperature dependencies.
[6] Stratton, J. A.: Electromagnetic Theory. John Wiley & Sons, Inc., Hoboken, NJ, 2007.
[7] Holman, J. P.: Heat Transfer. McGrawHill, New York, NY, 2001.
[8] www.quickfield.com, SW QuickField.
[9] agros2d.org, SW Agros.
11
Application of trapezoidal integration method

in determining discrete time windows
Jan Purczyński
West Pomeranian University of Technology
71-126 Szczecin, ul. 26 Kwietnia 10, e-mail: janpurczynski@ps.pl
1. Modified Dirichlet window determined from the trapezoidal rule
In one of the essential works on time windows, the rectangle (Dirichlet) window
is defined in the following way [2]:
- for the purpose of the finite Fourier transform (equation (21a)):
N N
w(n) = 1 for n = − ,...,−1,0,1,... (1)
2 2
- for the purpose of the discrete Fourier transform (DFT) ( equation (21b) ):
w(n) = 1 for n = 0,1,..., N − 1 (2)
Equation (1) includes N+1 points, whereas equation (2) only N points. It stems
from the fact that while switching from the Fourier transform (expressed by an
integral) to the DFT, none of the known numerical integration procedures is
applied, but only the last element of the sum is deleted.
A classic method of numerical integration are Newton-Cotes quadratures with
equally spaced points [1]. It consists in calculating an integral of f(t) function
defined in points
t n = nh; h = const; n = 0,1,..., N ; f n = f (t n )
Assuming a linear interpolation of the function the so called trapezoidal formula is
obtained:
⎛ f f ⎞
I T = h ⎜⎜ 0 + f 1 + f 2 + .. + f N −1 + N ⎟⎟ (3)
⎝ 2 2 ⎠
From equation (3) the following form of a modified Dirichlet window is
determined:
⎧ 1 for n = 1,2,..., N − 1
wm(n) = ⎨ (4)
⎩0.5 for n = 0 and n = N
By analogy with equation (4), the Dirichlet window (equation(2)) is given in the
form:
⎧1 for n = 0,1,2,..., N − 1
w(n) = ⎨ (5)
⎩ 0 for n = N
By applying equation (4) the trapezoidal rule (3) can be given as:
12
J. Purczyński / Application of trapezoidal integration method in …
N
IT = h ∑ wm(n) ⋅ f
n=0
n (6)
where: wm(n) is defined by equation (4).

The form of the Dirichlet window w(n) (equation(5)) as well as the modified
Dirichlet window wm(n) (equation(4)) for N=16 are shown in Fig.1.
1.5
1
w( n)
wm( n)
0.5
0
0 2 4 6 8 10 12 14 16
.
n
Fig. 1. Sample (N=16) Dirichlet window w(n) (equation (5)) - marked with circles - and modified
Dirichlet window wm(n) (equation (4)) – marked with bars
Applying the DFT to the Dirichlet window:

N −1
W (θ ) = ∑ w(n) exp(− jnθ )
n =0
(7)
a complex spectrum form is obtained ([2],equation (21c)):

⎛ Nθ ⎞
sin ⎜ ⎟
N −1 ⎝ 2 ⎠
W (θ ) = exp(− j θ )⋅ (8)
2 ⎛θ ⎞
sin ⎜ ⎟
⎝2⎠
Whereas when using the trapezoidal rule the following equation is applied:
N
WM (θ ) = ∑ wm(n) exp(− jnθ )
n =0
(9)
Having done the calculations the following equation is obtained

⎛ Nθ ⎞
sin ⎜ ⎟
Nθ ⎝ 2 ⎠ cos⎛ θ ⎞
WM (θ ) = exp(− j )⋅ ⎜ ⎟ (10)
2 ⎛θ ⎞ ⎝2⎠
sin ⎜ ⎟
2
⎝ ⎠
Figure 2 shows logarithmic characteristics (in decibels) of discrete time
windows w(n) and wm(n):
13
⎡ W (θ ) ⎤ ⎡ WZ (θ ) ⎤
LW (θ ) = 20 log ⎢ ⎥ ; LWM (θ ) = 20 log ⎢ ⎥ (11)
⎣ W (0) ⎦ ⎣ WZ (0) ⎦
0
10
20
LW ( θ ) 30
dB
LWM ( θ )
40
50
60
70
0 2 4 6 8 10 12 14 16
N .
θ⋅
2⋅π
Fig. 2. Logarithmic characteristics (in decibels) of time windows for N=32: the Dirichlet window
LW (θ ) (dotted line) and the modified Dirichlet window LWM (θ ) (solid line)
The amplitude characteristic of the modified window (solid line) demonstrates

faster falling than the classic window characteristic. An analogous situation occurs
for all time windows which at interval edges take the value different from zero.
2. Hamming window
By applying the DFT:

N −1
XH (θ ) = ∑ Hm(n) exp(− jnθ )
n =0
(12)
to the Hamming window

⎛ 2π n ⎞
Hm(n) = 0,54 − 0,46 ⋅ cos⎜ ⎟ (13)
⎝ N ⎠
we obtain:
⎛ ⎛ 2π ⎞ ⎛ 2π ⎞⎞
XH (θ ) = 0,54 ⋅ W (θ ) − 0,23 ⋅ ⎜⎜W ⎜θ − ⎟ + W ⎜θ + ⎟ ⎟⎟ (14)
⎝ ⎝ N ⎠ ⎝ N ⎠⎠
where W (θ ) - the rectangle window spectrum (equation(8)).
By implementing the trapezoidal rule:
N
XHM (θ ) = ∑ wm(n)Hm(n) exp(− jnθ )
n =0
(15)
we obtain:
14
⎛ ⎛ 2π ⎞ ⎛ 2π ⎞ ⎞
XHM (θ ) = 0.54 ⋅ WM (θ ) − 0.23 ⋅ ⎜⎜WM ⎜θ − ⎟ + WM ⎜θ + ⎟⎟ (16)
⎝ ⎝ N ⎠ ⎝ N ⎠ ⎟⎠
where WM (θ ) - spectrum of a modified rectangle window (equation (10)).
Figure 3 shows logarithmic characteristics (in decibels) of the Hamming
window LH (θ ) and the modified Hamming window LHM (θ ) .
11.43
22.86
LH( θ ) 34.29
dB
LHM( θ )
45.71
57.14
68.57
.
80
0 2 4 6 8 10 12 14 16
N
θ⋅
2⋅π
Fig. 3. Logarithmic characteristics (in decibels) of time windows for N=32: the Hamming window
LH (θ ) (dotted line) and the modified Hamming window LHM (θ ) (solid line)
3. Blackman-Harris window
As another example the Blackman-Harris window will be examined [2]:

⎛ 2π n ⎞ ⎛ 4π n ⎞
B(n) = a0 − a1 ⋅ cos ⎜ ⎟ + a2 cos⎜ ⎟ (17)
⎝ N ⎠ ⎝ N ⎠
In paper [2] the following coefficient values were provided
a 0 = 0.44959 ; a1 = 0.49364 ; a 2 = 0.05677 (18)
for which window (17) is characterized by the fact that the maximum of the first
sidelobe equals –61 dB.
In accordance with equations:
N −1
B (θ ) = ∑ B(n) exp(− jnθ ) (18)
n =0
N
BM (θ ) = ∑ wm(n)B(n) exp(− jnθ )
n =0
(19)
a complex spectra of the Blackman-Harris window (equation (18)) and the

modified Blackman-Harris window (equation (19)) were derived. Figure 3 presents
the logarithmic characteristics (equation (11)) of the Blackman-Harris
window LB(θ ) and the modified Blackman-Harris window LBM (θ ) .
15
Much like in Fig. 4, the LBM (θ ) amplitude characteristic demonstrates

substantially faster falling than the LB(θ ) characteristic.
0
10
20
30
40
LB ( θ )
50
LBM ( θ )
60 .
70
80
90
100
0 2 4 6 8 10 12 14 16
N
θ⋅
2⋅π
Fig. 4. Logarithmic characteristics (in decibels) of time windows for N=32: the Blackman-Harris
window LB(θ ) (dotted line) and the modified Blackman-Harris window LBM (θ ) (solid line)
Paper [2] provides yet another set of coefficients:

a 0 = 0.42323 ; a1 = 0.49755 ; a 2 = 0.07922 (20)
yielding the window characterized by the maximum of the first sidelobe equal–67
dB. Calculations yield amplitude characteristics demonstrated in Fig. 5.
Paper [3] draws attention to an inaccuracy concerning sidelobe maximum
values given in paper [2]. Namely, instead of –61 dB there should be –62,05 dB
and instead of –67 dB there should be –70,83 dB. Therefore in this paper the value
of the sidelobe maximum was determined in the function of window length N. The
results of the calculations are presented in Table 1.
0
10
20
30
40
LB( θ )
50
LBM( θ )
60 .
70
80
90
100
0 2 4 6 8 10 12 14 16
N
θ⋅
2⋅π
Fig. 5. Logarithmic characteristics (in decibels) of time windows for N=32: the Blackman-Harris
window LB (θ ) (dotted line, equations (17),(20)) and the modified Blackman-Harris window
LBM (θ ) (solid line)
16
Table 1. Maximum values of the first sidelobe in relation to window length N
Window length - 61 dB - 67 dB
N Maximum of the first sidelobe
16 -53,7 -62
32 -58,2 -66,7
64 -60,7 -70,6
128 -61,7 -70,8
256 -62 -70,8
Table 1 recognizes the fact that the window under consideration is a discrete
one – there is a considerable dependence of the maximum value on the window
length N. The values proposed in paper [3] are obtained for N=256, i.e. they refer
to continuous time windows.
4. Polynomial windows
Paper [4] describes the rules of a polynomial window construction. A relation

describing a polynomial window of the properties similar to those of the Hamming
window is given below - assuming an identical maximum value of the first
sidelobe (-42.87 dB for N=32):
⎡ 2n
3
⎛ 2n ⎞ ⎤
2
P(n) = 1 + 0.93494 ⎢2 ⋅ 1 − − 3⎜1 − ⎟ ⎥ (21)
⎢⎣ N ⎝ N ⎠ ⎥⎦
Figure 6 demonstrates the logarithmic characteristics (in decibels) of time

windows for N = 32: the Hamming window LH (θ ) (dotted line) and the polynomial
window LP(θ ) (solid line).
0
20
LH( θ )
LP ( θ )
40
.
60
0 2 4 6 8 10 12 14 16
N
θ⋅
2⋅π
Fig. 6. Logarithmic characteristics (in decibels) of time windows for N=32: the Hamming
window LH (θ ) (dotted line) and the polynomial window LP (θ ) (solid line)
17
In accordance with the accepted assumption, the equality of maximum values of

the first sidelobe is observed, however, for the following lobes the maximum for
the polynomial window ranges between 1.5 and 2 dB below the maximum for the
Hamming window.
Figure 7 shows the logarithmic characteristics (in decibels) of the polynomial
window LP (θ ) ( N=32) and the modified polynomial window LPM (θ ) . Faster
falling of the characteristic of the window determined on the basis of the
trapezoidal formula is observed.
0
20
LP ( θ )
40
LPM ( θ )
60
.
80
0 2 4 6 8 10 12 14 16
N
θ⋅
2⋅π
Fig. 7. Logarithmic characteristics (in decibels) of time windows for N=32: the polynomial
window LP (θ ) (dotted line) and the modified polynomial window LPM (θ ) (solid line)
Figure 8 demonstrates the comparison between the logarithmic characteristics (in

decibels) of time windows for N=32: the modified Hamming window LHM (θ ) (dotted
line) and the modified polynomial window LPM (θ ) (solid line). The maximum of the
following sidelobes for the modified polynomial window falls about 2 dB below the
maximum of the lobes for the modified Hamming window.
0
20
LHM( θ )
40
LPM ( θ )
60
80
0 2 4 6 8 10 12 14 16
.
N
θ⋅
2⋅π
Fig. 8. Logarithmic characteristics (in decibels) of the modified Hamming window LHM (θ )
(dotted line) and the modified polynomial window LPM (θ ) (solid line) for N = 32
18
5. Gaussian window
As yet another example the Gaussian window was considered [2]:

⎛ ⎛ ⎛ 2n ⎞ ⎞ 2 ⎞
G (n) = exp⎜ − ⎜⎜ 2⎜1 − ⎟ ⎟⎟ ⎟ (22)
⎜ ⎝ ⎝ N ⎠⎠ ⎟
⎝ ⎠
Figure 9 shows characteristics determined by the following methods: the DFT-
dotted line LG (θ ) , and the trapezoidal formula – solid line LGM (θ ) . Similarly to
all the previously examined cases, the logarithmic characteristic of the modified
Gaussian window ( LGM (θ ) ) falls faster than the logarithmic characteristic of the
Gaussian window ( LG (θ ) ).
0
10
20
30
LG( θ )
40
LGM( θ ) .
50
60
70
80
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
N
θ⋅
2⋅π
Fig. 9. Logarithmic characteristics (in decibels) of the Gaussian window LG (θ ) (dotted line)
and the modified Gaussian window LGM (θ ) (solid line) for N=32
6. Harmonic signal
The final example is a harmonic signal of the form:

x(n) = cos(n ⋅ θ 1 + φ ) (23)
By applying the following expressions:
N −1
X (θ ) = ∑ x(n) exp(− jnθ )
n =0
(24)
N
XM (θ ) = ∑ wm(n) x(n) exp(− jnθ )
n =0
(25)
a complex spectra X (θ ) and XM (θ ) are determined.
19
Figure 10 presents the amplitude of the complex spectrum X (θ ) (dotted line)

and XM (θ ) (solid line). For the complex spectrum X (θ ) and XM (θ ) logarithmic
amplitude characteristics are introduced:
⎡ X (θ ) ⎤ ⎡ XM (θ ) ⎤
LX (θ ) = 20 log ⎢ ⎥ ; LXM (θ ) = 20 log ⎢ ⎥ (26)
⎢⎣ X (θ 1) ⎥⎦ ⎢⎣ XM (θ 1) ⎥⎦
Figure 11demostrates the logarithmic amplitude characteristics of a harmonic
signal x(n) = cos(0,25π n) determined on the basis of equation (26): the dotted
line LX (θ ) ; the solid line LXM (θ ) .
15 θ1
X(θ) 10
XM( θ )
0
0 0.5 1 1.5 2 2.5 3
θ
Fig. 10. Complex spectrum amplitude of the harmonic signal x( n) = cos(0,25π n) : dotted line
X (θ ) determined as a DFT; solid line XM (θ ) - on the basis of equation (25)- the modified
complex spectrum of the signal
0
θ1
20
LX ( θ )
LXM( θ )
40
.
0 0.5 1 1.5 2 2.5 3

θ
Fig. 11. Logarithmic amplitude characteristics of the harmonic signal

x(n) = cos(0,25π n) determined on the basis of equation (26): dotted line LX (θ ) ; solid line
LXM (θ )
20
Both figures 10 and 11 demonstrate faster falling of the modified spectrum of

the harmonic signal determined using the trapezoidal formula (equation (25)).
Figure 12 compares the logarithmic amplitude characteristics of the harmonic
signal x(n) = cos(0,5π n) determined using equation (26): dotted line LX (θ )
(equation (24)) and solid line LXM (θ ) (equation (25)).
Figure 13 shows the logarithmic amplitude characteristics of the harmonic
signal x(n) = cos(0,75π n) determined using equation (26): the dotted line LX (θ ) -
in accordance with the DFT, and the solid line LXM (θ ) - in accordance with the
trapezoidal formula.
0
θ1
20
LX ( θ )
LXM( θ )
40
0 0.5 1 1.5 2 2.5 3

θ
Fig. 12. Logarithmic amplitude characteristics of the harmonic signal

x(n) = cos(0,5π n) determined using equation (26): dotted line LX (θ ) ; solid line LXM (θ )
0
θ1
20
LX ( θ )
LXM( θ )
40
.
0 0.5 1 1.5 2 2.5 3
θ
Fig. 13. Logarithmic amplitude characteristics of the harmonic signal x( n) = cos(0,75π n)

determined using equation (26): dotted line LX (θ ) ; solid line LXM (θ )
21
7. Summary
For all the examined time windows: Dirichlet, Hamming, Blackman-Harris ( -

61 dB, - 67dB), polynomial and Gaussian, faster falling of frequency
characteristics of the modified window was observed (determined on the basis of
the trapezoidal formula) compared to the frequency characteristics of the classic
window (determined using the DFT method).
In case of the harmonic signal, also faster falling of the modified spectrum of
the harmonic signal, determined by applying the trapezoidal formula (equation
(25)), was observed compared to the spectrum of the signal determined by applying
the DFT (equation (24)). However, it should be noted that Figures 10, 11, 12 were
created for the harmonic signal (equation (23)) with phase φ = 0 .
When changing the value of the phase φ , the obtained results do not
demonstrate such a clear advantage of the modified method over the classic one.
The author of this article has not considered all the aspects resulting from the
application of the trapezoidal formula, however, his aim was to draw attention to
the possibilities of using the trapezoidal formula in signal frequency analysis.
References
[1] Fortuna Z., Macukow B., Wąsowski J.: Metody numeryczne. Warszawa, WNT 1993,
s.166.
[2] Harris F.J.: On the Use of Windows for Harmonic Analysis with the Discrete Fourier
Transform. Proc. IEEE, vol.66, NO 1 January 1978, pp.51-83.
[3] Nuttall A.H.: Some Windows with Very Good Sidelobe Behavior. IEEE Tran. on
Acoustics, Speech, and Sign. Proc. V.ASSP-29,NO1, Feb. 1981,84-91.
[4] Purczyński J.: Wybrane zagadnienia teorii sygnałów. Wydawnictwo Naukowe
Politechniki Szczecińskiej, Szczecin 2000.
22
Comparing methods of image reconstruction

in electrical impedance tomography
Tomasz Rymarczyk, Stefan F. Filipowicz, Jan Sikora
Electrical Engineering Institute
04-703 Warszawa, ul. Pożaryskiego 28, e-mail: tomasz@rymarczyk.com
The problem of the image reconstruction in Electrical Impedance Tomography (EIT) is

a highly ill-posed inverse problem. There are mainly two categories of image reconstruction
algorithms, the direct algorithm and the iterative algorithm which was used in this
publication. This paper presents the applications of the level set function for identification
the unknown shape of an interface motivated by Electrical Impedance Tomography (EIT)
by using a several level set algorithms. The conductivity values in different regions are
determined by the finite element method. The applications were based on the level set
method, the variational level set algorithm and the Mumford-Shah algorithm to solve the
inverse problem.
1. Introduction
There are many algorithm of the image reconstruction in the electrical
impedance tomography (Fig. 1). In this paper was proposed a method based on the
combination level set idea and a few numerical methods to solve the inverse
problem [2, 8]. The representation of the boundary shape and its evolution during
an iterative reconstruction process is achieved by the level set method [5, 6, 7]. The
shape derivatives of this problem involve the normal derivative of the potential
along the unknown boundary. Numerical algorithm is a combination of the level
set method. Mumford-Shah model [4] and variational level set method [3] to solve
the inverse problem in the electrical impedance tomography (Fg.2). The
conductivity values in different regions are determined by the finite element
method. Numerical iterative algorithm is a combination of the level set methods for
following the evolving step edges and the finite element method for computing the
velocity. The objective function is defined as the difference between the potential
due to the applied current and the measured potential. This function is minimized.
Methods of the image reconstruction in the electrical impedance tomography
are following [1]:
A. Deterministic methods
- back-projection method,
- perturbation method,
- layer stripping,
- Kaczmarza method,
- Wexlera method,
23
T. Rymarczyk, S. F. Filipowicz, J. Sikora / Comparing methods of image reconstruction…
- gradients mehod:
- Newtona-Rapsona method,
- steepest descent method,
- Levenberg–Marquardt algorithm,
- conjugate gradient method,
- bell functions,
- level set methods.
B. Stochastic methods
- genetic algorithm (GA),
- simulated annealing (SA),
- Monte-Carlo method (MC).
C. Neural Networks.
D. Hybrid model of the optimization.
a) b) c)
d) e) f)
Fig. 1. The image reconstruction: a) model, b) conjugate gradient method, c) simulated annealing,
d) genetic algorithm, e) bell functions, f) level set method
24
Fig. 2. The reconstruction methods
2. Level set method
The level set method tracks the motion of an interface by embedding the
interface as the zero level set of the signed distance function. The motion of the
interface is matched with the zero level set, and the resulting initial value partial
differential equation for the evolution of the level set function. The idea is merely
to define a smooth function φ( x, t ) , that represents the interface as the set where
φ( x, t ) = 0 . The motion is analyzed by the convection the ϕ values (levels) with
the velocity field. The Hamilton-Jacobi equation of the form [6]:
∂φ
+ ν ∇φ = 0 (1)
∂t
where ν is the velocity on the interface.
When flat or steep regions complicate the determination of the contour, the
reinitialization is necessary. This reinitialization procedure is based by replacing by
another function that has the same zero level set but behaves better. This is based
on following partial differential equation:
∂
φ + S(φ)(∇φ − 1) = 0 (2)
∂t
where S(ϕ) is defined as:
⎧ - 1 for φ < 0
⎪
S(φ) = ⎨ 0 for φ = 0
⎪1 for φ > 0
⎩
25
Initialization –
Input Determination
zero level set
START data - conductivity
function
Forward problem
(finite element method)
Adjoint
equation
Velocity - calculation
Update function ϕ
Conductivity
- calculation
Reinitialization
NO
Condition
INVERSE PROBLEM
YES
STOP Results
Fig. 3. The iterative algorithm
26
The iterative algorithm was shown in Figure 3. The Figure 4 presents the
images of reconstruction by using the level set methods and the finite element
method.
a) b)
c) d)
e) f)
Fig. 4. The image reconstruction (mesh size 16×16) – the level set method: a) the original object and
the zero level set function, b) after 30 iterations, c) after 100 iterations, d) after 250 iterations,
e) the final reconstruction, f) the process reconstruction
27
3. Mumford Shah model
The Mumford-Shah algorithm sets formulation and minimization problem in

image processing, to compute piecewise-smooth optimal approximations of a given
image. The proposed model follows and fully generalizes work [4], where there
was proposed an active contour model without edges based on a 2-phase
segmentation and level sets γ 1 and γ 2 .
a) b)
c) d)
e) f)
Fig. 5. The image reconstruction – Mumford-Shah model: a) the original object and the zero level
set function, b) after 30 iterations, c) after 100 iterations, d) after 250 iterations,
28
Conductivity γ is represented as:

γ = γ I H(φ) + γ II (1 − H(φ)) (3)
where H is the Heaviside function.
The derivative of F with respect to γ is given by
⎡ ∂F ⎤ p
⎢ ⎥ = − ∑ ∇u j ∇p j (4)
⎣ ∂γ ⎦ j =1
Level set function is updated the following iterative scheme:

⎡ ∂F ⎤
φ k +1 = φ k − μ ⎢ ⎥ (5)
⎣ ∂φ ⎦
where coefficient μ>0 and
⎡ ∂F ⎤ ⎡ ∂F ⎤ ∂γ ⎡ ∂F ⎤
⎢ ⎥=⎢ ⎥ = ⎢ ⎥ ( γ I − γ II ) δ(φ) (6)
⎣ ∂φ ⎦ ⎣ ∂γ ⎦ ∂φ ⎣ ∂γ ⎦
where δ is the Dirac delta function.
Figure 5 presents model of computer simulation an image reconstruction by
using the Mumford-Shah model.
4. Variational level set method
The formulation of the variational level set method consists of an internal

energy term that penalizes the deviation of the level set function and an external
energy term that drives the motion of the zero level set toward the desired image
features. Variational formulation for geometric active contours that forces the level
set function to be close to a signed distance function, and therefore completely
eliminates the need of the costly reinitialization procedure. The resulting evolution
of the level set function is the gradient flow that minimizes the overall energy
functional [3]:
1
P (φ) = ∫ ( ∇φ − 1) 2 dxdy (7)
Ω2
An external energy for a function ϕ(x, y) is defined as below:
Ε ( φ ) = μ Ρ (φ ) + Ε m ( φ ) (8)
P (φ) – internal energy,
Ε m (φ) – external energy.
∂Ε
Denoting by the Gateaux derivative of the functional E receiving the
∂φ
following evolution equation:
29
∂φ ∂Ε
=− (9)
∂t ∂φ
The process for minimization of the functional E is the following:
∂φ ⎡ ∇φ ⎤
= μ ⎢Δφ − div( )⎥ + νδ(φ) (10)
∂t ⎢⎣ ∇φ ⎦⎥
using the variational level set methods.
a) b) c)
d) e) f)
Fig. 6. The image reconstruction – the variational level set method: a) the original object and the zero
level set function, b) after 30 iterations, c) after 100 iterations, d) after 250 iterations,
5. Variational level set method with Mumford-Shah model
The method is based on the Mumford-Shah algorithm the variational level set
method.
using the variational level set methods and Mumford-Shah model.
30
a) b)
c) d)
e) f)
Fig. 7. The image reconstruction – the variational level set method with Mumford-Shah model:
a) the original object and the zero level set function, b) after 30 iterations, c) after 100 iterations,
d) after 250 iterations, e) the final reconstruction, f) the process reconstruction
31
6. Comparing methods
Table 1 shows values of the objective function for one and two objects by using
the various algorithms. The Figure 8 presents an image reconstruction by using the
level set methods, the variational level set method and Mumford-Shah model. The
Figures 9 shows minimum values of the objective function.
Table 1. Values of the objective function
Objective function (min)

ALGORITHMS
1 object 2 objects
Level Set Method 145 611
Mumford-Shah model 194 332
Variational Level Set Method 90 1598
Variational Level Set Method and Mumford-Shah Model 1455 961
a) b)
c) d)
Fig. 8. Comparing methods – 2 obiects: a) the level set method, b) Mumford-Shah model,
c) the variational level set method, d) the variational level set method with Mumford-Shah model
32
Fig. 9. The objective function - 2 objects
7. Conclusion
This paper has introduced the level set algorithms of approximation of material coefficient.
The applications of the level set function, the variational level set method, Mumford-Shah
model and the finite element method for the electrical impedance tomography were presented.
The level set idea is known to be a powerful and versatile tool to model evolution of interfaces.
Variational formulation is the faster the traditional level set because completely eliminates the
need of the costly reinitialization procedure. The Mumford-Shah model gives the best quality
reconstruction unknown areas with many objects.
References
[1] Filipowicz S.F., Rymarczyk T.: Tomografia Impedancyjna, pomiary, konstrukcje i
metody tworzenia obrazu. BelStudio, Warsaw, 2003.
[2] Ito K., Kunish K., Li Z.: The Level-Set Function Approach to an Inverse Interface
Problem. Inverse Problems, Vol. 17, No. 5, pp. 1225-1242, 2001.
[3] Li C., Xu C., Gui C., and M. D. Fox., “Level set evolution without re-initialization: A
new variational formulation”, In IEEE Conference on Computer Vision and Pattern
Recognition (CVPR), volume 1, pages 430–436, 2005.
[4] Mumford D., Shah J.: Optimal approximation by piecewise smooth functions and
associated variational problems. Comm. Pure Appl. Math., (42):577–685, 1989.
[5] Osher S., Fedkiw R.: Level Set Methods and Dynamic Implicit Surfaces. Springer, New
York, 2003.
[6] Osher S., Sethian J.A.: Fronts Propagating with Curvature Dependent Speed:
Algorithms Based on Hamilton-Jacobi Formulations. Journal of Computational
Physics, 79, 12-49, 1988.
[7] Osher S., Santosa F.: Level set methods for optimization problems involving geometry
and constraints. Frequencies of a two-density inhomogeneous drum. Journal of
Computational Physics, 171, pp. 272-288, 2001.
[8] Rymarczyk T., Filipowicz S.F., Sikora J.: Numerical algorithms for the image
reconstruction in electrical impedance tomography. COMPUMAG 2009, Florianópolis,
Brasil, November 22-26, 2009.
33
Influence of the external proximity effect on the power losses in

flat three phase high current busduct
Zygmunt Piątek, Tomasz Szczegielniak, Dariusz Kusiak
Częstochowa University of Technology
42-200 Częstochowa, ul. Brzeźnicka 60a, e-mail: zygmunt.piatek@interia.pl,
dariuszkusiak@wp.pl, szczegielniak@interia.pl
In the paper the influence of the external proximity effect on the active and reactive
power in flat three phase high current busduct were presented. Calculations were made
using the Poynting theorem and Joule-Lenz law.
1. Introduction
Flat high current busduct with tubular conductors are applied mostly as an
unsheathed in switching station, low and high voltages, and sheathed as a
connection of big generators and transformers, and also in power transmission in
long distance for high value of nominal currents – Fig. 1 and Fig. 2 [1-9].
The design of the busducts used for high currents and voltages causes a necessity of
precise describing of electromagnetic, dynamic and thermal effects. Knowledge of the
relations between electrodynamics and constructional parameters is necessary in the
optimization construction process of the high current busducts [1, 2].
Information about distribution electromagnetic field and power losses is a base
into analysis of electrodynamics and thermal effects in the high current busducts.
Determination of the power losses into high current busducts enables the
calculation of temperature of these devices, which is a basic constructional
parameter [1, 2].
Fig. 1. Power station [10]
34
Z. Piątek, T. Szczegielniak, D. Kusiak / Influence of the external proximity...
y y’ y’’
H1 X(r,Θ,z) H2 H3
J2 J3
J1 r
rXZ
Θ rXY x’ x’’
μ0 1 μ0 2 μ0 3 x
L1 L2 I3 L3
γ γ I2 γ
I1
R2 R1 R1 R2 R1 R2
d d
Fig. 2. Flat three phase high current busduct
Power losses depend on value of currents, but for the large cross-sectional
dimensions of the phase conductor, even for industrial frequency, skin and external
proximity effect (Fig. 3) should be taken into account [5-9].
x
z
X(r,Θ) R2
Je1
Θ Hw I2
e r
y
Je1
μ0 γ1 R1
R3
R4
d
Fig. 3. Eddy currents induced in the screen by the magnetic field of the neighboring phase conductor
2. Electromagnetic field in flat three phase high current busduct
Let us consider the electromagnetic field in flat three phase high current busduct
presented in the Fig. 2.
Total density current in the first conductor [5, 7]:
J 1 (r , Θ )  J 11 (r )  J 12 (r , Θ)  J 13 (r , Θ)  1 z J 11 (r )  J 123 (r , Θ)  1 z J 1 (r , Θ )
(1)
where current density J 11 (r ) take into account the skin effect and has a form
Γ I 1 K1  Γ R1  I 0  Γ r   I1  Γ R1  K 0  Γ r  Γ I1
J 11 r    j (r ) (2)
2 πR2 I1  Γ R2  K1  Γ R1   I1  Γ R1  K1  Γ R2  2 πR2
while
35
K1  Γ R1  I 0  Γ r   I 1  Γ R1  K 0  Γ r 
j (r )  (2a)
I 1  Γ R2  K1  Γ R1   I 1  Γ R1  K 1  Γ R2 
and current density J 12 (r , Θ ) is defined by formula
 n
Γ I2 R 
J 12 (r , Θ )  
π R2
  d2  f n (r ) cos nΘ (3)
n 1
where
K n1 ( Γ R1 ) I n ( Γ r )  I n1 ( Γ R1 ) K n ( Γ r )
f n (r )  (3a)
I n1 ( Γ R2 ) K n1 ( Γ R1 )  I n1 ( Γ R1 ) K n1 ( Γ R2 )
but current density J 13 (r , Θ)
 n
Γ I3 R 
J 13 (r , Θ)  
π R2
  2 d2  f n (r ) cos nΘ  J13 (r , Θ) exp[j J 13 (r , Θ)] (4)
n 1
If currents
2 2
I 2  exp[  j π ] I 1 oraz I 3  exp[j π ] I 1 (5)
3 3
then total current density in first conductor has a form
  n 
Γ I1  R2 
J 1 (r , Θ)  J 11 (r )  J 123 (r , Θ)   j (r )  2  An   f n (r ) cos nΘ 
2 π R2  n 1  d  
(6)
where
1
An  
2
  
1  2 n  j 3 1  2  n  An exp[j  n ] (6a)
An  1  2 n  4  n (6b)
 n   π  arctg

3 1 2 n
 (6c)
1  2 n
Total eddy currents in second conductor is described by formula (6) in which
current I 1 should be replaced by I 2 and constant A n by constant
1
Bn 
2
n
n

  1  1  j 3  1  1  Bn exp[j n ]  (7)
For the third conductor, current I 1 should be replaced by I 3 , but constant A n by
Cn 
 1n  1  2 n  j  
3 1  2  n  C n exp[j n ] (8)
2
36
Total magnetic field in the first conductor has a form [6, 7]

H 1 (r , Θ )  H 11 (r )  H 12 (r , Θ )  H 13 (r , Θ )  1 r H 1r (r , Θ)  1Θ H 1Θ (r , Θ) (9)
where radial component
H 1r (r , Θ )  H 12r (r , Θ)  H 13r (r , Θ)  H 123r (r , Θ) (10)
while
 n
I2  R2 
H 12 r (r , Θ)    
π Γ R2 r n1  d 
 n f n (r ) sin nΘ (10a)
and
 n
I3  R2 
H 13r (r , Θ )      n f n (r ) sin nΘ
π Γ R2 r n1  2 d 
(10b)
Tangent component of magnetic field in first conductor

H 1Θ (r , Θ)  H 11Θ (r )  H 12Θ (r , Θ)  H 13Θ (r , Θ)  H 11Θ (r )  H 123Θ (r , Θ)
(11)
where
I1 K1 ( Γ R1 ) I1 ( Γ r )  I1 ( Γ R1 ) K1 ( Γ r ) I
H 11Θ (r )   1 h(r ) (11a)
2 πR2 I1 ( Γ R2 ) K1 ( Γ R1 )  I1 ( Γ R1 ) K1 ( Γ R2 ) 2πR2
K1 ( Γ R1 ) I1 ( Γ r )  I1 ( Γ R1 ) K1 ( Γ r )
h(r )  (11b)
I1 ( Γ R2 ) K1 ( Γ R1 )  I1 ( Γ R1 ) K1 ( Γ R2 )
and
 n
I1 R 
H 123Θ (r , Θ)   
π Γ R2 r n1  d 
 
A n  2   n f n (r )  g n (r ) cosnΘ (11c)
K n1 ( Γ R1 ) I n1 ( Γ r )  I n1 ( Γ R1 ) K n1 ( Γ r )

g n (r )  Γ r (11d)
I n1 ( Γ R2 ) K n1 ( Γ R1 )  I n1 ( Γ R1 ) K n1 ( Γ R2 )
In the same way we can determinate the magnetic field in second and third
conductor.
In the above formulas I 0 ( Γ r ) , K 0 ( Γ r ) , I1 ( Γ r ) , K1 ( Γ r ) , I n ( Γ r ) ,
K n ( Γ r ) , I n1 ( Γr) , Kn1 ( Γr) , I n1 ( Γr) and Kn1 ( Γr) are modified Bessel’s
functions, 0, 1, n, n-1 and n+1 order, calculated for r  R1 and r  R2 , and the
complex propagation constant of electromagnetic wave
π
Γ j   exp[j ]  k  j k  2 j k (12)
4
in which attenuation constant
37
 1
k  (12a)
2 
where δ is the electrical skin depth of the electromagnetic wave penetration into
the conducting environment,  is an angular frequency,  means conductivity of
conductor, and μ 0  4 10 7 H  m -1 is magnetic permeability of the vacuum.
3. Influence of the external proximity effect on power losses

in flat three phase high current busduct
Stream of the complex power penetrating into first conductor is defined by the
formula [11]
*
S 123 (r , Θ)  E 123 (r , Θ)  H 123 ( r , Θ) (13)
Apparent power of the first conductor is equal
 *
S 123   E 123 (r )  H 123 
(r )  dS  P123  j Q123 (14)
S
from where a complex apparent power S 123 in the first conductor has a form
j l I12  2n   * *
 R2  n f n ( R1 ) f n ( R1 )  f n ( R2 ) f n ( R2 ) 
S 123   An2     (15)
π  R22 n 1  d   f ( R2 ) g * ( R2 )  f ( R1 ) g * ( R1 ) 
 n n n n
In the above formula we can not isolate the real part (as an active power) and
the imaginary part (as a reactive power). It is impossible on account of the complex
propagation constant and complex modified Bessel’s functions. Therefore the
active power will be calculated from formula [11]
l 2 R2
1 * 1 *
P123   J 123 ( r ) J 123 ( r ) dV     J 123 (r ) J 123 (r ) r dr dΘ dz (16)
V
  0 0 R1
From the formula (16) we get
2n
Γ * l I12 
 R2  an
P123   An2   (17)
2 π  R2 n 1  d  b n b *n
where

a n  I n 1 ( Γ R2 ) K n 1 ( Γ R1 ) I n*1 ( Γ R1 ) K n* ( Γ R 2 )  I n* ( Γ R2 ) K n*1 ( Γ R1 )  
(17a)
 j I n ( Γ R2 ) K n 1 ( Γ R1 ) I n*1 ( Γ R2 ) K n*1 ( Γ R1 ) - I n*1 ( Γ R1 ) 
K n*1 ( Γ R 2 )
 I n 1 ( Γ R1 ) 

 K n*1 ( Γ R1 ) I n* ( Γ R2 ) K n 1 ( Γ R2 )  j I n*1 ( Γ R ) K ( Γ R )  
2 n 2


 I n*1 ( Γ R1 ) K n* ( Γ R2 ) K n 1 ( Γ R2 )  j K n*1 ( Γ R ) K ( Γ R )  
2 n 2
38
b n  I n1 ( Γ R2 ) K n1 ( Γ R1 )  I n1 ( Γ R1 ) K n1 ( Γ R2 ) (17b)

b *n  I n*1 ( Γ R2 ) K n*1 ( Γ R1 )  I n*1 ( Γ R1 ) K n*1 ( Γ R2 ) (17c)
If we introduce the reference active power
l I12
P0  (18)
π  ( R22  R12 )
then the relative active power in the first conductor has a form
 2n
P - j 2j 1 a
(P)
k123  123 
P0 2
1  2 
    b nb*  An2 (19)
n 1 n n
R R d
where   2  k R2 ,   1 ( 0    1 ) and    1.
 R2 R2
Dependence of the coefficient (19) on parameter α for different values of the
relative walls thickness β of the first conductor and of relative distance between
conductors λ is presented in the Fig. 4.
a)
b)
Fig. 4. Dependence of the relative active power in the first conductor on parameter α:
a) for constant value of the parameter β, b) for constant of the parameter λ
39
The reactive power in the first conductor

2n
l I 22 
 R2 
Q123   j ( S 123  P123 )   An2   
π  R2 n 1  d 
 n

n f ( R1 ) f *
n
f n ( R2 ) f *n ( R2 ) 
( R1 )  

(20)
 * * Γ R2 a n 
 f n ( R2 ) g n ( R2 )  f n ( R1 ) g n ( R1 )  
 2 b n b*n 
If we introduce the reference reactive power
μ0 l  R14 R2 1 3R12  R22  2
Q0  X 0 w I12    ln   I1 (21)

2 π  R2  R2
 2 1 
2
R1 4 R22  R12 

then the relative reactive power of the first conductor has a form
 2n
(Q ) Q 1 1
k123  123   An2   
Q0  4 1 1 3 2  1 n1  
2  ln  

 1   2 
2
 4 1  2 

(22)
n

n f ( R1 ) f * ( R1 )  f ( R2 ) f * ( R2 ) 
n n n
 
 
 * * 2 j  a 
n
 f n ( R2 ) g n ( R2 )  f n ( R1 ) g n ( R1 )  * 
 2 b n b n 
relative walls thickness β of the first conductor and of relative distance between
In the same way we can calculated power losses in the second conductor. The
relative active power in the second conductor has a form
 2n
P - j 2j 1 an
( P)
k 213  213  1  2   Bn2   (23)
P0 2 n 1  b n b*n
relative walls thickness β of the second conductor and of relative distance between
The relative reactive power of the second conductor has a form
40
 2n
(Q ) Q 1 1
k 213  213   Bn2   
Q0  4
2 1 1 3 2  1 n1  
  ln  

 1   2 
2
 4 1  2 

(24)


n f ( R1 ) f * ( R1 )  f ( R2 ) f * ( R2 ) 
n n n n
 

 * * 2 j  an 
 f n ( R2 ) g n ( R2 )  f n ( R1 ) g n ( R1 )  
 2 b n b *n 
relative walls thickness β of the second conductor and of relative distance between
a)
b)
Fig. 5. Dependence of the relative reactive power of the first conductor on parameter α:
41
a)
b)
Fig. 6. Dependence of the relative active power in the second conductor on parameter α:
42
a)
b)
Fig. 7. Dependence of the relative reactive power of the second conductor on parameter α: a) for
constant value of the parameter β, b) for constant of the parameter λ
In the case of third conductor we have

C n  An  1  2 n  4 n (25)
hence
(P) P312 (P)
k 312   k123 (26)
P0
and
(Q ) Q312 (Q)
k 312   k123 (27)
Q0
43
4. Conclusions
In produced high current busducts, for industrial frequency value of parameter

 is included from 5 to 20. It means that active power in conductors of the flat
three phase high current busduct can reach value of active power connected with
phase current (Fig. 4). In the second conductor (Fig. 6) active power connected
with proximity effect can be two times higher than active power of phase current.
Similarly, reactive power connected with proximity effect in first and third
conductor of three phase high current busduct can reach value reactive power of
phase current (Fig. 5). But in second conductor reactive power connected with
proximity effect is few times higher than reactive power of phase current (Fig. 7).
In all determined causes proximity effect is stronger when distance between
conductors is shorter.
References
[1] Nawrowski R.: Tory wielkoprądowe izolowane powietrzem lub SF6. Wyd. Pol.
Poznańskiej, Poznań 1998.
[2] CIGRE Brochure No 218.: Gas Insulated Transmission Lines (GIL). WG 23/21/33-15,
CIGRE, Paris, 2003.
[3] Piątek Z., Szczegielniak T., Kusiak D.: Moc czynna i bierna bifilarnego toru
wielkoprądowego, XV ZKwE’2010, s. 35-36, Poznań 2010.
[4] Piątek Z., Szczegielniak T., Kusiak D.: Wpływ zewnętrznego zjawiska zbliżenia na
straty mocy w trójfazowym płaskim torze wielkoprądowym, XVI ZKwE’2011, s. 35-
36, Poznań 2011.
[5] Piątek Z.: Impedances of Tubular High Current Busducts. Series Progress in High-
Voltage technique, Vol. 28, Polish Academy of Sciences, Committee of Electrical
Engineering, Wyd. Pol. Częst., Czestochowa 2008.
[6] Kusiak D.: Pole magnetyczne dwu i trójbiegunowych torów wielkoprądowych.
Rozprawa doktorska, Pol. Częst., Wydz. El., Częstochowa 2008.
[7] Piątek Z., Szczegielniak T., Kusiak D.: Straty mocy w płaskim rurowym trójfazowym
torze wielkopradowym. Wiadomości Elektrotechniczne, LXXVII 2009 nr 11.
[8] Piątek Z., Kusiak D., Szczegielniak T.: Pole magnetyczne trójfazowego płaskiego toru
wielkoprądowego, Zesz. Nauk. Pol. Śl. 2009, Elektryka, z. 1(209), ss. 51-65.
[9] Piątek Z., Kusiak D., Szczegielniak T.: Reverse Reaction Magnetic Field in Flat
Three Phase High Current Busduct, XXXII IC-SPETO, s. 17-18, Gliwice-Ustroń
2009.
[10] Szuba M. i inni: Linie i stacje elektroenergetyczne w środowisku człowieka,
Informator PSE-Operator S.A., Warszawa 2008.
[11] Krakowski M.: Elektrotechnika teoretyczna. Pole elektromagnetyczne. WN PWN,
Warszawa 1995.
44
Software for quality evaluation of a PS plate edge

Krzysztof Kluszczyński, Tomasz Trawiński, Marcin Szczygieł
Silesian University of Technology
44-100 Gliwice, ul. Akademicka 10A, e-mail: tomasz.trawinski@polsl.pl
Erik Boeren, André Skibniewski

FUJIFILM Manufacturing Europe B.V.
5000 LJ Tilburg, Oudenstaart 1, P.O. Box 90156,
e-mail: andre_skibniewski@fujifilm.eu
Pre-sensitized (PS) plates are specially treated and coated aluminium sheets, widely
used for offset printing. PS plates used in the Computer to Plate technology require strict
quality control of the plate’s edges, disturbed during various finishing processes like cutting
on guillotines or slitting. For this purpose a special tests stand had been built by the Silesian
University of Technology, in cooperation with one of the largest PS plates manufacturer in
the world – FUJIFILM. The paper describes the software, written for the purpose of
evaluation of various defects caused during finishing processes.
1. Introduction
Pre-sensitized (PS) plates are approximately 0.15 – 0.4 mm thick aluminium

sheets, covered by several layers of a light- or heat- sensitive coating. Hundreds of
millions square meters of these materials are used yearly in the offset printing
industry. The State of Art – Computer to Plate (CtP) technology requires from
manufacturers adherence to various quality requirements, more strict and
demanding than the applicable Standard (1). A very important part of every PS
plate are the edges, sometimes severely disturbed during finishing processes,
especially during cutting and slitting. Defective products cannot be repaired.
Defects to the plate’s edges can have detrimental effect on the printing quality (i.e.
“edge toning”) and can cause damages to the machine parts of the printing press.
Excessive cutting burrs can also cause injuries during plate handling. Some typical
defects introduced during finishing processes to the PS plate are shown in Fig. 1.
Extend of some defects, like burrs and edge fold can easily be determined by
the use of a standard profile measuring apparatus. Other, like plate’s “roughness”
or so called “edge ripple” have to be estimated organoleptically. To overcome this
difficulty a special testing stand had been built and tested at the Silesian University
of Technology at the Department of Mechatronics. The PS Plate Edge Scanner
allows quality control of the PS plate’s edge, during a non-contact, non-destructive
measurement of the selected sample area. Major advantage of this apparatus is
able to estimate the plate edge “roughness” as no other test method is available.
45
K. Kluszczyński, T. Trawiński, M. Szczygieł / Software for quality evaluation …
a)
b)
Fig. 1. Typical defects to PS plate after cutting: a) Cutting burrs (B ), Edge Bending (EB)
and Edge “Roughness”, b) Severe form of a PS plate defect caused by cutting on a guillotine
2. Definition of the PS plate edge “roughness”.
“Roughness” on the plate’s edge might be defined as a rough unevenness of the

plates’ surface. It is caused by aluminium and aluminium oxide particles adhering
to it. There is no equipment for evaluation of this defect, and consequently it is
estimated by the technician during examination of the product’s edge after cutting.
Some “rough” particles can be removed easily with a fingernail, causing difficulty
to obtain and to preserve reference materials. This defect has, however, serious
consequences for plate’s quality. It causes potential risk of jamming and pollution
with fine particles during exposure and development of the plate and can have
detrimental influence on the quality of the printed image.
Origin of the “roughness” phenomenon has been investigated in detail by the
team of scientists led by A. Mężyk and Z. Rdzawski [2 – 4]. A typical view of a
plate surface after cutting shows Fig. 2.
46
Fig. 2. Particles causing „edge roughness” of the PS plate
In order to overcome the difficulty in evaluation of the severity of this defect, a

prototype of a measuring stand has been constructed at the Department of
Mechatronics, Faculty of Electrical Engineering of the Silesian University of
Technology in Gliwice, Poland [5]. Two laser sensors moving alongside the edge
measure the size of particles adhering to the plate. The “roughness” phenomenon
can be investigated further, allowing to take proper countermeasures in order to
limit the extend of this defect.
Roughness
Plate thickness
Burr
Roughness
and Burr
Fig. 3. Exemplary defects of plate edge
The magnified region of PS plates edge area are presented in Fig. 4.

The newly constructed at the Silesian University of Technology apparatus can
estimate also severity of various other defects, like cutting burrs, edge folding,
edge ripples etc. [6]. Results of organoleptic judgements of the operators and
47
surface profile measurements were combined and integrated, resulting in one,

objective judgement regarding the PS plate edge quality.
Fig. 4. Magnified region of the PS plate edge area after cutting
3. A Prototype of the PS plate edge scanner
The general view of prototype PS Plates Edge Scanner is presented in Fig. 5.

The main components of this device are as follows:
Fig. 5. A general view of the PS Plate Edge Scanner
- Laser scanning heads – two Keyence LJ-G030 laser scanning heads were used.
The view of a single laser head is shown in Fig. 6. The two heads work
simultaneously and scan top and bottom parts of PS plate with standard vertical
resolution of 1 μm. The width of the scanning heads laser beam is 22 mm at a
reference distance of 30 mm.
48
Fig. 6. Exemplary view of the Keyence LJ-G030 laser scanning head used for the Project
The two laser heads create accessible measuring (scanning) area, consisting
of two trapezoidal regions. The measuring area and its fundamental
dimensions are shown in Fig. 7a. The tested PS plate should be placed close
to the central part of measuring area, - it assures high accuracy of the
measurement. The way how the PS plate should be placed in the measuring
area is shown in Fig. 7b.
a) b) Centre
of measuring
Head A range
Reference Scanning
Distance area
30 mm
20mm plate
Head B Measuring range

– 22 mm
Fig. 7. Basic dimensions of the accessible measuring area of the PS Plate Scanner
- Scanning laser heads controller – this device is responsible for controlling the
heads and collecting the temporary measurement data. The LJ-G5001P laser
heads controller is able to collect in its internal memory data from 1024 scanning
series. Naturally, it is possible to extend the number of storage series by adding
an external memory card.
49
Fig. 8. Scanning head LJ-G50001P controller
- Servo linear module – this device is responsible for transporting the sample plate
between the scanning heads in a discreet way. The maximum (stable) discreet
resolution of horizontal linear motion is 200 μm,
- PLC controller – it is responsible for controlling the linear servo module and
keeps the linear translation of measured sample in the specified region,
- Touch panel – allows operator to control the measurement (scanning) process. It
allows to set the test conditions – like number of scans, selected/specified
scanning region etc.
- Personal computer – it is used for collecting measurement data and storing them
in the data base. Installed software allows evaluation of the data collected during
testing. Finally, the quality of the PS plate sample can be estimated.
4. Operation algorithm of the PS plate scanner
The PS plate scanner consists of six main hardware components, which are
necessary for proper operations. They include: scanning heads, LJ-G5001P laser
heads controller, linear servo module with servo controller, FESTO PLC controller,
a touch panel and a personal computer. Mutual interactions between main
components and flow signals are shown in Fig. 9.
The touch panel allows for setting position of the table for suitable PS plate
sample placed on it. In addition, it allows for choosing the number of desired scans
and the scanning region. From the touch panel all mentioned settings are sent to the
PLC controller. The PLC controller sets the servo module to the desired home
50
position. After reaching it, the strobe signal is sent to the laser scanning head
controller. The controller denotes permission to begin single scan of the PS plate
sample. Results of the measurements are sent to the personal computer and are
presented by the LJ-Navigator software. The LJ-Navigator software sends the
second signal to the laser scanning heads controller, containing information if
further storing of the data is required. Exemplary raw data after a single scan are
presented in Fig. 10. The top curve corresponds to the shape (single scan from edge
to interior part of PS plate) of the coated side of the examined PS plate; the bottom
curve corresponds to the uncoated side of the PS plate.
PS
PS plate
plate sample
sample
Laser
Laser Laser Laser
PLC Servo
scanning
scanning scanning scanning
controller module
heads
heads heads heads
Laser Laser
Touch
PC
scanning scanning
panel
heads heads
Fig. 9. Block structure and mutual interactions between the hardware components
of the PS Plate Scanner
Fig. 10. Exemplary data representing single scan of both sides of a PS plate
51
5. Algorithm of operation of PS plate quality evaluation software
The Algorithm of operation of a PS plate quality evaluation software consists of

twelve steps, shown in Fig. 11.
Fig. 11. The Algorithm of operation for a PS plate quality evaluation

52
There are following steps of the Algorithm:

1. *.csv file processing - all scanning data are stored into the internal memory of
LJ-G5001P laser heads controller. This data should be finally saved in the
desired folder, using the LJ-Navigator software with "csv" extension. The
"*.csv" file contains the data representing the shape of the top and the bottom
of the PS plate surface (recorded with the desired resolution, put-in by the user
through the touch panel).
2. Selecting the method of measurements – the measurements method is based
on collecting data from the upper and the lower laser head in the measurement
range (tested area) for an adequate number of scanning points.
3. Separation of top and bottom data - at this step the data are split into two
sets, representing only the data related to the top and the bottom surface of the
PS plate. Furthermore, the data are cut into 200 points in the direction towards
the interior part of PS plate. It limits the data for further analysis to a stripe of
an average width about 5.4 mm and the standard length defined and set by user
from touch panel. Exemplary data representing the shape of the top and the
bottom side of a single scan after data clipping are shown in Fig. 12.
3.5
thickness [ mm ]
3.4
3.3
3.2
3.1
0 1 2 3 4 5 6
width [ mm ]
Fig. 12. Exemplary shape of the top and the bottom fragment of a PS plate after single scan
4. 3D surface plot - in this step the processed data from the top and the bottom
side of a PS plate are prepared for three-dimensional plot presentation.
Exemplary 3D plot of the scanned plate is shown in Fig. 13. A severe
deformation between point 60 and point 70 on the plate edge is clearly visible.
5. Preparing the set of cross sections of the examined plate - at this step the
reference area for the base is selected and the angle between the reference area
and the rest of the plate is calculated.
6. Linear approximation - at this step the reference area for the base is selected.
From all the data set of each individual scan (for every side of the plate), thirty
samples near the edge of the base region are omitted. The remaining results are
used for linear approximation and for the calculation of the empirical
53
“Roughness Coefficient”. The reference regions used the in data from Fig. 10
are shown in Fig. 14.
thickness [ mm ]
length [ samples ] width [ mm ]
Fig. 13. The view of a reconstructed fragment of the scanned PS plate
3.5
thickness [ mm ]
3.4
3.3 Reference
regions
3.2
3.1
0 1 2 3 4 5 6
width [ mm ]
Fig. 14. Regions used for linear approximation, necessary for establishing artificial reference region
7. Linear extrapolation to the suspected “roughness” and the “edge bending”

regions - at this step the interpolation line from step 4 is extrapolated to the
region suspected for “roughness” and “bending” defects. The distance between
the extrapolating line and the line where measured data are collected is
calculated.
8. Linear approximation to the “roughness” region - at this step the samples
close to the plate’s edge are approximated by a line, however every first
measurement result is omitted from approximation (because it may be
influenced by a single roughness particle). Once again the distances between
samples and the approximation line are calculated.
9. Calculation of the “Roughness Classification Factor” - at this step, based on
a calculated distance between the measured distances from the approximation
line at the roughness region an empirical so-called “Roughness Classification
Factor” is calculated according to the following formula:
54
∑S Ak + ∑ S Bk
RF = k k
k ⋅ 100 , (1)
Z −k
where: RF – „Roughness Factor”, SAk and SBk – height of the single bar from the
lower and the upper side, Z – total number of scans (i.e. 200), k – number of
measurements with bars higher than 9 μm.
10. Calculation of the “Edge Bending (Edge Folding) Classification Factor”
- calculating the bending factor is based on the value of the angle between data
from linear approximation at bending region and the data from linear
approximation from the remaining measurement points beyond the reference area.
11. Saving the measurement results - at this step all measured and analyzed
data are saved on the hard disk of the server of the PC unit.
12. Comparison of the data from “Roughness” and the “Bending” analysis -
at this step the saved results are compared with the historical data for
statistical analysis.
6. PS plate quality evaluation software
The program for quality evaluation of PS plates was created in the Matlab
environment, with the use of the GUI (Graphical User Interface) module. Finally it
was compiled to an executable version. The program is dedicated to users with two
different access levels – the “Basic” and the “Advanced” level (Fig. 16).
Fig. 15. Starting window of the program for quality evaluation of PS plates
The Basic Level version is dedicated to all machine operators. It enables

obtaining quick information about the Roughness Factor and can be used directly at
the workstation in the factory. The graphical interface of basic level is divided into
four main parts, as shown in Fig. 17. In the area marked “A” the users observe the
final results of the analysis. General information about the outcome: actual date of
measurement, numeric value of the “Roughness Factor” (between 0 and 200) and
the classification symbol which represents the quality of the plate are displayed.
There are four classes describing the quality of the plate’s edge: A (very good), B
55
(good), C (questionable) and D (bad). Classes C and D require further analysis of

the origin of the defect and proper countermeasures must be taken.
The area “B” contains a fields dedicated for inputting personal data of the
operator – like his name or nick name and the symbol or a number of the working
team. In the area “C” the user has a possibility to choose the device (i.e. guillotine)
number. All the data in B and C areas are typed by operator. The D area is a control
panel for opening the source directory file, for starting analysis of the process, for
saving data or for printing results of the measurements.
Fig. 16. Main window of the “Basic Level” mode
The most sophisticated tools are included in the Advanced Access Level mode.
In this case the software is dedicated to specially trained operators and engineers,
interested in much more detailed analysis. The general view of the Advanced Level
window is shown in Fig. 18.
The main window displayed on the computer screen for the advanced user can
be divided into five areas. Window marked as area “E”, placed in the central part
of the screen, enables visualization of the measurement results. In allows free
choice between the 2D and 3D view of the displayed data. This tool is very useful
when more detailed analysis (i.e. trouble shooting) of a PS plate is required. The
visualization includes the form of the 3D plots: top and bottom surface plot (Fig.
19) and the deformation bar plots of the edge area (Fig. 21).
56
Fig. 17. A view of the main window of the “Advanced Level” mode
Fig. 18. A view of the main window of the “Advanced Level” – top and bottom surface plot
Also 2D plots are enabled, such as: line plots of edge area (Fig. 20) and the bar
plot of local Roughness Factor distribution (Fig. 21).
The all settings: the type of plot and the range of presented data are placed in
the A area. The remaining information, like the date of measurement, Roughness
Factor and the quality of plate is basically the same as described previously. In the
down left corner (area B) the operator puts in his personal data. On the right hand
57
side of the main window (area C), the graph the navigation icons are placed. This
panel contains tools for choosing isometric view, for rotating the view and for
resetting the view settings. The Area D is the same as described previously – it
allows the operator to select the file folders, type of charts to be displayed, print out
formats and contains file saving options. The software does not require re-entry of
data after working with the basic level or advanced level, because the procedure
analyses saves the data in text files which are stored in a folder on the hard disk
drive. If necessary, the old files can be analyzed again and compared with the more
recent ones.
Fig. 19. A view of the main window of the “Advanced Level” – line plot
Fig. 20. A view of the main window of the “Advanced Level” – 2D bar plot
58
7. Conclusions
After a period of an extensive research and repeatedly carried out scanning of

over 500 PS plate samples, a very good correlation had been reached between the
organoleptic, human judgment of PS plates edge quality and the test result obtained
by the PS Plate Edge Scanner. The program will be extended for all other useful
modules, like estimation of edge bending and burrs alongside the plates edge. In
addition, some recently observed phenomenon like – the “edge ripple”, related to
newest generation of the PS plates used in CtP technology, will be measured and
classified, adding a new dimension to the quality control in the PS plate industry.
References
[1] ISO 12635, 2008. Graphic Technology - Plates for Offset Printing - Dimensions. s.l. :
International Standards Organization, 2008.
[2] Kaczmarczyk J., Gąsiorek D., Mężyk A.: Analiza numeryczna przyczyn powstawania
defektów w ustalonym procesie cięcia płyt na gilotynach. Modelowanie Inżynierskie.
Grudzień 2007, pp. 61-66.
[3] Kaczmarczyk J., Gąsiorek D., Mężyk A.; Skibniewski A.: Connection Between the
Defect Shape and Stresses Which Cause it in The Bundle of Sheets Cut on Guillotines.
Gliwice : Politechnika Śląska, 2007. Modelling and Optimization of Physical Systems.
pp. 81-84. ISBN 9788360102466.
[4] Mężyk A., et al.: Badania doświadczalne procesu cięcia pakietu blach na gilotynie.
Przegląd Mechaniczny. Maj 2010, pp. 36-38.
[5] Skibniewski A., Boeren E., Kluszczyński K., Trawiński T., Szczygieł M., Kielan P.,
Pilch Z., Podworski K.: Laser Scanner for Quality assessment of Offset Plates (in
Polish) . Warsaw : World Convention of Polish Engineers, 2010.
[6] Skibniewski A., Kluszczyński K., Trawiński T., Pilch Z., Szczygieł M., Kielan P.,
Laser Beam Scanner fo Quality Evaluation of Pre-sensitized Offset Plates. Acta
Technica Jaurinensis. 2010, Vol. Vol.3 No. 2.
59
Naval mine detection system based of FPGA circuit

Mirosław Wołoszyn, Jacek Łubkowski, Marek Chomnicki
Gdańsk University of Technology
80-952 Gdańsk, ul. Narutowicza 11/12, e-mail: mwoloszyn@ely.pg.gda.pl
1. Introduction
Electrochemical processes take place in a metal object immersed into sea water
even if an anticorrosive coating is applied [1]. As a result, flowing field appears
around the object. There are naval mines between many other objects situated in
the sea. Naval mines can be put in the seabed in order to be more difficult to detect
by sonars. Such a mine is located on the line demarking two environments of
different electrical conductivity. The results of a computer analysis of the influence
of seabed’s electrical conductivity on the electric field’s disposition around a mine
model are shown below. Also, a naval mine detection system based on the FPGA
circuit is presented.
2. Electric field of an object
Determination of the most effective electric field mine detecting sensors layout
was one of the goals of a numeric electric field analysis or a naval mine model’s
electric potential disposition. A mine-shaped object was modeled with two sources
(Fig. 1) of potential difference equaling U. The distance between the sources equals
L. All the quantities are presented as dimensionless according to the following
formulas:
ϕ
V= (1)
U
R
r= (2)
L
where: V – dimensionless potential, ϕ – electric potential, U – tension between
electric field point sources, R – distance from measurement point to the middle of
mine model, L – distance between the electric field sources equaling the length of
the mine-shaped object, r – dimensionless distance measurement point to the
middle of mine model.
The values of all the coordinates as well as the distance between the electric
field sensors are given as dimensionless values referred to the distance between the
L electric field sources equaling the length of the analyzed mine-shaped object.
The simulations were performed for the tension between electric field point sources
U = 1V and the distance between the sources L = 1m.
60
M. Wołoszyn, J. Łubkowski, M. Chomnicki / Naval mine detection system based...
Fig. 1. Naval mine model in numeric analysis
A dimensionless electrical conductivity defined with the correlation below was

applied in the research:
σd
σ= (3)
σw
where: σw – electrical conductivity of water, σd – electrical conductivity of the
seabed.
On the basis of analysis of many computer simulation results it was stated that
in the case where the electric field source is located above the seabed or on the
borderline between the seabed and water, the seabed electrical conductivity value
does not have a significant impact on the disposition of the potential in the water.
Such an impact is observable when the electric field is located beneath the seabed.
In Fig. 2 the dependence of maximum dimensionless potential value on the
dimensionless “z” height above the object hidden on a dimensionless depth of
z = 0.25 is presented.
As shown in Fig. 2 and Fig. 3, the influence of seabed’s electrical conductivity
is particularly relevant for short distances between the electric sensor and the
object. For distances longer than three times maximal dimension of the object the
influence of the seabed is not big. The influence of seabed’s electrical conductivity
on the electric field disposition in space is insignificant. If the flowing field source
is placed on the borderline between the seabed and water or if it is placed above the
seabed, the influence of seabed’s electrical conductivity is not relevant for the
values and the disposition of the potential above the object. The dependence of the
61
dimensionless potential on the dimensionless depth “z” follows the curve shown in
Fig. 2 for σ = 1.
Fig. 2. Dependence of the maximum dimensionless electric potential value on the dimensionless “z”
height above the object (the electric field source is located beneath the seabed for z=0.25)
Fig. 3 Dependence of the maximal dimensionless electric potential value in logarithmic scale on
dimensionless “z” height above the object, for an object located beneath the seabed
3. Analysis of electric potential disposition

in sensors arranged in a circle
An analysis of electric potential disposition of a naval mine model was carried

out for different sensor configurations. Eventually, as a result of the carried
research, it has been stated that arranging the sensors in a circle is the most
effective. A dimensionless electric potential disposition and the disposition of
62
potential differences between each sensor set in a plane system where the sensors
are located on a circle of dimensionless diameter equaling 2 is presented below.
Fig. 4. The sensors’ layout in a circle and explanation of the α angle
The dispositions of electric potential were made along straight lines parallel to
the x axis. It was assumed that the electric field source lies in the middle of the
coordinate system (x = 0, y = 0, z = 0), while the straight line linking the two
spherical electric fields sources form a α angle with the x axis (Fig. 4). Examples
of electric potential disposition in each sonar for angles α = 0° and α = 90° are
presented in Fig. 5 and Fig. 6.
The following conclusions result from a numeric analysis of dimensionless
electric potential disposition around a mine-shaped object located on the seabed or
beneath the seabed:
• arranging sensors in a circle is the most effective for determining the
location of a naval mine-shaped objects and for their identification;
• in order to localize and identify naval mine-shaped objects a sensor arrangement
in two concentric circles set in one plane should be considered. The reference
sensor should be placed in the middle of the two circles;
• the diameters of the circles where the electric field sensors will be arranged
depend on the length L of the searched object. The suggested diameters should
equal 2L and 4L, respectively. The sensor arrangement with the smaller diameter
shall localize a mine from a shorter distance while the sensor arrangement with
the bigger diameter shall localize a mine from a longer distance. The plane of the
sensor arrangement should be parallel to the seabed. For technical reasons, the
maximal diameter of the sensor arrangement must not be too big;
63
0.02
mas
Vs1( x)
Vs2( x)
0.01
Vs3( x)
Vs4( x)
Vs5( x)
Vs6( x)
Vs7( x)
0
Vs8( x)
Vs9( x)
Vs10( x)
Vs11( x)
Vs12( x)
− 0.01
mis
− 0.02
− 10 −5 0 5 10
− 10 x 10
Fig. 5. Dispositions of dimensionless electric potential for each sensor in Fig. 55 for α = 0°, z = 1, y = 0
mas
0.02
Vs1( x)
Vs2( x)
Vs3( x)
Vs4( x)
Vs5( x)
Vs6( x)
Vs7( x)
0
Vs8( x)
Vs9( x)
Vs10( x)
Vs11( x)
Vs12( x)
− 0.02
mis
− 10 −5 0 5 10
− 10 x 10
Fig. 6. Dispositions of electric potential for each sensor for α = 90°, z = 1, y = 0
• the distance from which a mine is possible to localize depends on the resolution
of the measurement instrument (the resolution of the electric field sensor
arrangement plus an amplifier);
64
• other than circular geometric configurations of the sensor arrangement, such as

spherical or hemispherical etc., show worse properties for naval mine
localization and identification possibilities.
4. Mine detection system based of FPGA circuit

Mine detection system requires a continuous and fast signal processing and
operation of peripherals. These tasks can be performed by using an FPGA circuit.
A parallel structure of such a circuit enables a fast hardware implementation of the
introduced algorithms. Logic gate array with a possibility to reconfigure the
connections between the gates is the basic element of the FPGA circuit (Field
Programmable Gate Array). After programming the gates constitute a hardware
version of the designed algorithm or function. VHDL of Verilog hardware
description language is used to describe the gates. Besides the possibility to design
parallel structures, the FPGA circuits also enable implementing complete
microcontrollers and microprocessors together with cooperating digital peripherals.
Every producer offers his own processor destined for his FPGA circuit, e.g. Xilinx
company provides a 8-bit PicoBlaze or a 32-bit MicroBlaze [3]. The designer
chooses the appropriate microprocessor, available as the so-called IP-core
described with VHDL or Verilog language [2]. The company offers many other IP-
cores easily addible to one’s own project as a block.
Fig. 7. Flowchart of the measurement system
A flowchart of a measurement system integrated into the naval mine detection

system is presented in Fig. 7. The electric potentials of electric field sensors
measured regarding the reference sensor potential after amplification with a high
65
precision, small voltage offset, low noise amplifiers are measured with A/D
converters of high resolution (24 bits). The voltage values are decoded by a FPGA
processor, undergo a digital filtration to be sent to a computer on the water surface
by a RS485 system. The FPGA circuit (Fig. 8) also enables implementation of
object detection algorithm.
Electric field sensors are an important element of the measurement system.
Metrological parameter values should be identical for all the employed sensors,
particularly with regard to the changes of the parameters in time.
Fig.8. Photograph of FPGA circuit
Fig. 9. Electric field sensors in a measurement capsule
In Fig. 9 there are electric field sensors located in a small waterproof

measurement capsule. Inside the capsule the FPGA circuit is installed together with
66
other electronic devices. Fig. 10 shows the sensor disposition in two circles in a
naval mine detection technology demonstrator. Fig. 11 presents the biggest
measured potential difference between the sensors No 9 and No 14 (Fig. 10) of the
object detected in the sea water.
Fig. 10. Sensor disposition in the naval mine detection technology demonstrator
Fig. 11. Course of changes of the electric potential differences between the sensors No 9 and No 14
Silver chloride electrodes were used in the arrangement as the electric field
sensors., that is metallic silver covered with a layer of silver chloride AgCl
immersed in a saturated solution of potassium chloride. Chemical reactions occur
in the electrode between the surface of the membrane and the solutions on both
sides of the membrane under the influence of electric field changes. Besides a low
value of the potential, the Ag-AgCl electrode also shows stability of the potential
and lack of toxic effects.
67
5. Summary and final conclusions
The paper presents the results of a numeric analysis of the influence of seabed’s
electrical conductivity on the disposition of the electric potential around the electric
field source hidden in the seabed and the electric potential disposition of the
electric field. The structure of a naval mine detection measurement system with
particular regard to the implemented PFGA circuit was described.
Research done under the development project OR00007009

of Ministry of Science and Higher Education .
References
[1] Ochrona elektrochemiczna przed korozją. WNT Warszawa 1991.

[2] Spartan-6, Virtex-5, and Virtex-6 FPGAs, UG129 (v2.0) January 28, 2010.
[3] Microbalze Processor Reference Guide, Embedded Development Kit EDK 10.1i,
UG081 (v9.0) January17, 2008.
Computer Applications in Electrical Engeenering
Modeling magnetic field in the vicinity of single wire helix

Krzysztof Budnik, Wojciech Machczyński
Poznań University of Technology
60 – 965 Poznań, ul. Piotrowo 3A, e-mail: Wojciech.Machczynski@put.poznan.pl
Transmission of the electric power is accompanied with generation of low – frequency

electromagnetic fields. Electromagnetic compatibility studies require that the fields from
sources of electric power be well known. Many of these sources are not defined to the
desired degree of accuracy. This applies e.g. to the case of helical current conductors of
finite length. The paper presents an analytical-numerical method of the calculation of the
3D magnetic fields in vicinity of conductors having helical structure. The method is based
on the Biot-Savart law and a software tool to model the magnetic fields generated by e.g.
twisted wires, helical coils, etc can use the formulas obtained.
1. Introduction
Electromagnetic compatibility studies require that the fields from sources of

electric power be well known. Many of these sources are not defined to the desired
degree of accuracy. Twisting of conductors pairs is a well known method used in
telephone communications to minimize crosstalk among lines in bundled cables.
Generally, the effect of twisting can be utilized when the magnetic field is a
problem. Already practiced is twisting of insulated high-voltage three phase power
cables and single-phase distribution cables as well [1 – 8].
The paper presents a method of the calculation of magnetic fields in vicinity of
finite length conductors having helical structure. The method is based on the Biot-
Savart law.
2. Magnetic field produced by helical line current of finite length
The realistic model of a twisted cable should be based on the theory of a helical
line current of finite length instead on the theory of infinitely long one [1]. The
magnetic flux density in the observation point due to the current path, as in Fig. 1,
can be computed using the Biot-Savart law [9]:
μ0 I (dl × 1r )
B ( x, y , z ) = ∫ (1)
4π c r2
In (1) I is a phasor current, the vector element dl coincides with the direction
of the current I, 1r is a unit vector in the direction of the vector r , r is the
69
K. Budnik, W. Machczyński / Modeling magnetic field in the vicinity...
distance between the source point Ps(X,Y,Z) and the observation point P(x,y,z) and
µ0 is the magnetic permeability of the vacuum.
P(x,y,z)
dl
r
1r
Ps(X,Y,Z)
I
c
Fig. 1. Current path generating the magnetic field
Consider the helical line of finite length L located in the Cartesian co-ordinate
system (x,y,z), Fig. 2.
z
I P(x, y, z)
h dl
Ps(X(φ), Y(φ), Z(φ))
r
0
a φ0 y
x φ
Fig. 2. Helical line current of finite length
Denoting the radius of the helix by a, h the pitch distance of the helix, ϕ0 the
ϕ co-ordinate of the point where the helix intersects the plane z = 0 , the
parametric equations of the helical line with respect to the parameter φ
( ϕ0 ≤ ϕ ≤ 2πL / h + ϕ0 ) indicated on Fig.2 and with ϕ0 ≠ 0 are:
70
X (ϕ ) = a cos ϕ
Y (ϕ ) = a sin ϕ (2)
h
Z (ϕ ) = (ϕ − ϕ0 )
2π
In order to apply the formula (1), we have to find suitable expressions 1r (ϕ ) and
dl (ϕ ) . By looking at Fig.2, if X (ϕ ) , Y (ϕ ) , Z (ϕ ) are the coordinates of the
generic element dl (ϕ ) and 1x ,1y ,1z are rectangular unit vectors, we have
( x − X (ϕ ))1x + ( y − Y (ϕ ))1y + ( z − Z (ϕ ))1z
1r (ϕ ) = (3)
( x − X (ϕ )) 2 + ( y − Y (ϕ )) 2 + ( z − Z (ϕ )) 2
and taking into account the eqn.(2):
h
( x − a cos ϕ )1x + ( y − a sin ϕ )1y + [ z − (ϕ − ϕ0 )]1z
1r (ϕ ) = 2π (4)
h
( x − a cos ϕ )]2 + ( y − a sin ϕ )]2 + [ z − (ϕ − ϕ0 )]2
2π
h
dl (ϕ ) = − a sin ϕ dϕ1x + a cos ϕ dϕ1y + dϕ1z (5)
2π
Taking into account the formulas (4) and (5) at finding the product of
dl (ϕ ) × 1r (ϕ ) and remembering that
r 2 (ϕ ) = ( x − X (ϕ ))2 + ( y − Y (ϕ ))2 + ( z − Z (ϕ ))2
(6)
we obtain from the eqn.(1) the three components of the magnetic flux density field:
2πL h h
+ϕ 0 a cos ϕ[ z − (ϕ − ϕ0 )] − ( y − a sin ϕ )
μ I h
2π 2π
Bx ( x , y , z ) = 0 ∫ dϕ
4π ϕ0
3
⎡ 2 2 h ⎤
2 2
⎢⎣( x − a cos ϕ )] + ( y − a sin ϕ )] + [ z − 2π (ϕ − ϕ0 )] ⎥⎦
(7)
2πL h h
+ϕ 0 a sin ϕ[ z − (ϕ − ϕ0 )] + ( x − a cos ϕ )
μ I h
2π 2π
B y ( x, y , z ) = 0 ∫ dϕ
4π ϕ0
3
⎡ 2 2 h 2⎤2
⎢⎣( x − a cos ϕ )] + ( y − a sin ϕ )] + [ z − 2π (ϕ − ϕ0 )] ⎥⎦
(8)
71
2πL
+ϕ 0
μ0 I h − a sin ϕ ( y − a sin ϕ ) − a cos ϕ ( x − a cos ϕ )
Bz ( x, y, z ) = ∫ dϕ
4π ϕ0
3
⎡ 2 2 h 2⎤2
⎢⎣ ( x − a cos ϕ )] + ( y − a sin ϕ )] + [ z − (ϕ − ϕ0 )] ⎥
2π ⎦
(9)
where L is the length of a conductor considered and
h
L= ϕ (10)
2π
The integrals in formulas (7) – (9) have to be solved numerically.
Finally, the modulus of the magnetic flux density field can be obtained from the
formula:
2 2 2
B ( x, y , z ) = B x ( x , y , z ) + B y ( x , y , z ) + Bz ( x , y , z ) (11)
It should be pointed out that in case h → 0 , the helical line current degenerates
to a conducting circular loop of radius a, placed symmetrically in the xy plane with
the center located at the origin 0, and curried the current I. Evaluation of the
integrals (7) – (9) shows that for x = 0, y = 0 and z ≠ 0, the x and y components of
the magnetic flux density vanish and well-known
μ0 I a 2
Bz = (12)
2(a 2 + z 2 )3 / 2
remains as the only nonvanishing component.
It should be noted, that the formulas derived enables one to analyze the
magnetic field produced by twisted-wire pair as well as by three-core cable
considering the conductor twist. The twisted-pair cable can be represented
mathematically as a double helix that consists of two helices having the same
radius and pitch; the helices are located 180 spatial degrees from each other. In the
three-wire helix structure the location of conductors in the z = 0 plane is fixed by
angles ϕ0i , where ϕ0i = (i − 1) 2π / 3 .
It should be also mentioned, that an alternative way to the direct application of
formulas derived consists in the discretization of the helix by means of a suitable
number of rectilinear segments. Thus, the field is evaluated by superposition of the
contributions produced by each segment [8].
3. Examples of calculations
Verification of the method presented lies in comparison of calculation results

with calculation and measurement results available in some papers. In order to
verify the correctness of the analytical calculations presented in the paper, first
72
comparison has been made with an analytical solution in form of infinite series
containing Bessel functions obtained in [5] for infinitely long helical conductor.
Figures 3 – 5 show Bx (Br), By (Bφ) and Bz, respectively, along the axial direction at
radial distance from helix axis r = 1 cm, whereas the behavior of the module of the
magnetic flux density as a function of radial distance from the axis of the helical
conductor is shown in Fig.6. The calculations refer to I = 1 A, a = 1 cm, h = 2 cm
and ϕ0 = 0 for helix length ranging from 8 m to 0.08 m.
Fig. 3. Bx component versus z (y = 0)
Fig. 4. By component versus z (x = 0)
73
Fig. 5. Bz component versus z
Fig. 6. Module of B versus x
It follows from the calculations, that independent of helix length the agreement
was excellent, for field components except of the By (Bφ), what is evident from
physical point of view. As shown in Fig. 2 the distribution of the component of the
magnetic flux density along the axial direction for short helix is evidently different
as compared with the same component produced by a long helical line.
Discrepancy between the results obtained by the use of the method appropriate for
the infinitely long helical conductor and by use of the method for conductor finite
74
in the length shows that for short helical conductors the proposed method shall be
applied.
Further verification of the method described is made by comparison with results
of measurements on an experimental helical arrangement. In the laboratory
experiment presented in [6] a rig consisting of plastic coated 50 mm2 stranded
copper wires helically wound on an 11 m long plastic pipe of 0.2 m diameter has
been constructed. The wires were fed by a balanced 200 A, 50 Hz 3-phase current.
The three components of the magnetic flux density were measured at various
positions by use of a one-coil magnetic field meter. For each distance, the field at
six points was measured and the average value was calculated and presented in [6]
for the 1.0 m pitch. Figure 7 presents the calculation results obtained according to
the proposed method. The calculations have been curried out in the middle part of
the three-wire helix structure for each radial distance at six points spaced 0.1 m
(analogical to the test [6]) and the average values of field components are shown in
Fig.7 as functions of radial distance. The agreement between the experimental
results and the theoretical solution is excellent.
Fig.7. Components of the magnetic field density as function of radial distance from axis of the 3-
phase cable
The usefulness of the presented method shall be next shown through its
application for the calculation of the 3D magnetic field produced in the vicinity of
helical conductor. Fig.8 shows the 3D distribution of the magnetic flux density in
the xz-plane located over the helical line current at distance y = 2 cm from helix
axis. The calculations have been curried out according to formulas derived for the
helix with length L = 6 cm, and refer to I = 1 A, a = 1 cm, h = 2 cm and ϕ 0 = 0 .
75
Fig. 8. Distribution of the magnetic field density in the xz- plane over the helical line current
4. Final remarks
The design of installation generating low-frequency magnetic field requires

access to effective analytical and computational tools.
A method of the determination of the magnetic field for finite length helical
conductors is presented. The analytical formulas for calculating the 3D magnetic
field are derived. For the helical conductor with finite length the method is based
on the Biot-Savart law. Verification of the method presented lies in comparison of
calculation results with calculation and measurement results available in some
papers. Discrepancy between the results obtained by the use of the method
appropriate for the infinitely long helical conductor and by use of the method for
conductor finite in the length shows that for short helical conductors the proposed
method shall be applied.
The integration of formulas describing the magnetic flux density must be
performed numerically. The problem arises that in case of decreasing pitch distance
the integrand oscillates increasingly which brings about numerical difficulties.
The formulas allow tackling the magnetic field of the two-wire helix, as well as
for the three-wire helix and can be used by a software tool to model the magnetic
fields generated by e.g. twisted-wire pairs, twisted three-phase power cables,
triplex service cables, helical coils, etc.
The results derived can be used as the foundation for almost every study on the
magnetic fields for conditions that are almost always satisfied for power
engineering applications.
76
References
[1] Buchholz H.: Elektrische Strömungsfelder mit Schraubenstruktur. Elektrische

Nachrichtentechnik, Band 14, 1937, Heft 8. S. 264 – 284.
[2] Moser J.R., Spencer R.F.: Predicting the magnetic fields from a twisted-pair cable.
IEEE Trans. on Electromagnetic Compatibility, Vol. EMC-10, No. 3, September 1968,
pp. 324 – 329.
[3] Shenfeld S.: Magnetic fields of twisted-wire pairs. IEEE Trans. on Electromagnetic
Compatibility, Vol. EMC-11, No. 4, November 1969, pp. 164 – 169.
[4] Haber F. The magnetic field in the vicinity of parallel and twisted three-wire cable
carrying balanced three-phased current. IEEE Trans. on Electromagnetic Compatibility,
Vol. EMC-16, No. 2, May 1974, pp. 76 – 82.
[5] Hagel R., Gong L., Unbehauen R.: On the magnetic field of an infinitely long helical
line current. IEEE Trans. on Magnetics, Vol. 30, No. 1, January 1994, pp. 80 - 84.
[6] Pettersson P., Schönborg N.: Reduction of power system magnetic field by
configuration twist. IEEE Trans. on Power Delivery, Vol. 12, No. 4, October 1997, pp.
1678 – 1683.
[7] Budnik K., Machczynski W., Haubrich H.-J.: Magnetic field of helical conductors with
finite length. COMPUMAG2011, 12-15 July, Sydney, Australia, PA2.8 (ID 150) CD.
[8] Holbert K.E., Karady G.G., Adhikari S.G., Dyer M.L.: Magnetic fields produced by
underground residential distribution system. IEEE Trans. on Power Delivery, Vol. 24,
No. 3, July 2009, pp. 1616 - 1622.
[9] Griffiths D.J.: Introduction to Electrodynamics. Prentice Hall, Englewood Cliffs, N.J.,
1989.
77
Multisource model of ship electric field

Kazimierz Jakubiuk, Paweł Zimny, Mirosław Wołoszyn
Gdańsk University of Technology
80-952 Gdańsk, ul. Narutowicza 11/12, e-mail: mwoloszyn@ely.pg.gda.pl
A steel-hulled ship in sea water produces an electric field around itself. The source of
this field are electrochemical processes and ship’s cathodic protection. The analysis of the
electric field around the ship is important by the reason of the ship’s identification and
counter-mine protection. The paper presents a simplified model of the ship’s multisource
electric field, which allows to calculate the electric field distribution in the so-called distant
zone around the ship in an approximate way using a PC, during time lasting from few up to
several dozen of minutes.
1. Introduction
An electric field is formed around the steel hull of the ship in sea water. This
field is created due to the electrochemical processes occurring on the corrosion
damaged fragments of the protective coating of the hull and on the ship's propeller,
which is made of a different material than the hull, and due to the existing systems
of the passive or active cathodic protection in the form of electrodes placed on the
ship’s hull. In the sea water surrounding the ship, there are currents associated with
the enviroment physicochemical processes. The existence of these processes causes
formation of an electric field around the underwater section of the hull. The
existence of this field is detected by measuring facilities,and their task is to
identify, and often, to destroy the ship. Therefore, there is a great need to study the
electrical field around the ship.
The physical and chemical processes on the interface metal - sea water, which
is the electrolyte, do not play a significant role in the electric field distribution
around the submerged ship’s hull, especially when it is measured at a distance
greater than the dimensions of the electrodes or the dimensions of the anticorrosive
coating defects. Even, if the potential distribution on the surface of such electrodes
has a slight fluctuation, it does not affect the distribution of the electric field at a
distance comparable to the dimensions of the electrode surface. On that basis, it is
assumed that the metal electrodes, where a direct contact between the electrolyte
and the metal surface occures, are characterized by a constant potential Ve on the
whole surface of the electrode. The size of this potential results from the
parameters of the electrochemical series of metals or is forced by an external
source used in the active cathodic protection. Fragments of the surface, where the
protective coating has been destroyed, become electrodes and their potential results
from the electrochemical series of metals. The surface covered with anticorrosive
78
K. Jakubiuk, P. Zimny, M. Wołoszyn / Multisource model of ship electric field
coating can be treated as an insulating surface, and that means the inability of
current flow through this surface.
In the case of the calculations of the distribution of the vessel's own electrical
field induced by electrochemical processes occurring on its surface, it is necessary
to take into account the propeller. The propeller plays a significant role as an
electrode because it is made of a material with a different electrochemical potential
than the ship’s hull.
The paper presents a new method of calculating the electric field around the
ship allowing, with satisfactory accuracy, to calculate the distribution of the
electric field around the ship in much shorter time. The idea of the method relies on
that, that the ship is replaced by the system of multiple sources from which the
potential on the surface, identical as the surface of the ship, is compatible (with an
acceptable approximation) with the potential for a real ship’s hull surface.
2. Mathematical model of ship's electric fields
In order to obtain a tolerably simple description of the analytical model of the

ship’s hull, it was assumed that it consists of three parts. The stern is a cone with
the generatrix described by the second degree curve, which, while preserving the
continuity of the first derivative, goes into a central part, which has a cylindrical
shape. The ship’s bow is also a cone with the generatrix described by the second-
degree curve, which guarantees the continuity of the first derivative at the
transition from the ship’s middle part to the bow (Fig. 1).
Fig. 1. The model of the submerged hull with a superimposed division grid of the hull surface
The mathematical model of the ship’s propeller shape is assumed in the form of a
cylinder with a diameter equal to the diameter of the propeller base and a height
equal to the thickness of the propeller. The propeller shaft is omitted because the
applied alternative sources model is not able to take into account elements, which
surfaces are much smaller than of the remaining elements, with sufficient accuracy.
79
It is worth to emphasize that the full model of the ship with all its details is possible
while using the mathematical apparatus allowing for a detailed description of the
hull shape. Integral equations are certainly such an apparatus. Application of this
method allows for a more accurate description of the ship’s electric field [2 ÷ 8].
Unfortunately, the detailed calculations using the boundary element method (BEM)
are time consuming and to obtain an acceptable computation time, they should be
conducted by large multiprocessor numeric machines, and not on personal
computers. The authors’ aim was to build a mathematical model allowing to
calculate the electric field distribution around the vessel in an approximate manner
using a personal computer, and during several to several dozen of minutes.
The paper proposes a mathematical model allowing to calculate the electric
field distribution around the ship in an approximate manner. To achieve this
objective, point sources have been placed inside the hull in xi, yi, zi points and they
are described the relation:
1 1
G i (x , y, z, x i , yi , zi ) = +
(x − x i )
2
+ (y − yi ) + (z − zi )
2 2
(x − x i )
2
+ (y + yi )2 + (z − zi )2
(1)
The selection of the source function in the form of (1) allows to satisfy the
boundary condition on the surface of the conductive medium that fills the half
space y < 0:
∂G i
=0 (2)
∂y y =0
The boundary condition results from zeroing of the normal component of the
current density on the border between the conductive medium and the insulator,
which is air in the area of y > 0 (Fig. 1).
The potential φ (x, y, z) of the flow field in the conductive half-space y <0 is
taken as the sum of potentials of different sources:
Nz
ϕ(x, y, z ) = ∑ a n G n (x , y, z, x1n , y1n , z1n ) (3)
n =1
where: an – the n-th source efficiency.
In order to determine the efficiency of each sources Trefftz method has been
used. By virtue of Green's symmetric formula, this function fulfills the equation in
the half-space y <0:
∫ (ϕ∇ ) ⎛ ∂G i ∂ϕ ⎞
2
G i − G i ∇ 2 ϕ dV = ∫ ⎜ ϕ − Gi ⎟dS (4)
V S ⎝ ∂n ∂n ⎠
Function Gi described by the formula (1) satisfies the Laplace’s equation, as its
primary function, so the volume integral on the left side of equation (4) equals
zero. On the other hand, the surface integral on the right side (4) has been written
80
as a sum of integrals, taking into account the boundary conditions on particular

boundaries, and in result the equation has been obtained:
∂G i ∂ϕ ∂G i
∫ϕ ∂n
dS + ∫ G i
∂n
dS = Ve ∫
∂n
dS (5)
Siz S e S e
where: n – the outer normal to the surface limiting electrolyte, Siz – the border
∂ϕ
insulating surface, for which the boundary condition is met =0 .
∂n Siz
It is worth to point out that due to adopting the function of sources in the
form (1), there is no integral over the surface y = 0 To avoid difficulties related
with the calculation of integrals, the sources have been placed on a line parallel to
the z-axis (Fig. 1) and going through the gravity center of the arc obtained as the
intersection of the hull’s cylindrical part with the plane z = const.
Fig. 2. The potential distribution on the hull surface along the generatrix (Fig. 1) for 10 sources.
Line 21 corresponds to the hull cross-section with plane x = 0
To examine the quality of the above approximation, it has been assumed that
the propeller’s potential equals -2, and there are two electrodes on the hull (one in
the bow part and the other in the stern part) and their potential equals 2. The area
between the electrodes is partially devoid of the anticorrosive coating and its
potential is 0, and the remaining part of the hull is covered with the anticorrosive
coating. The potential distribution on the hull’s surface calculated by Mathcad is
shown in Figure 2. As results from the potential distribution on the hull’s surface
presented in Figure 2, its largest value along the keel line Ψs (21, υ) differs from
81
the assumed potential of the electrode almost 30%. On the other hand, a good
approximation of the potential without anticorrosive coating is obtained. Increasing
the number of sources to 20 does not make substantial adjustments but greatly
extends the time of calculation. The calculation results can be improved using a
more dense division grid of the hull surface while keeping the same number of
sources, and that does not affect the calculation time so significantly as increasing
the number of sources.
The comparison of the components distribution of the electric field, produced
by 20 and 10 sources 4m below the keel, is shown in Figure 3 - component Ex,
Figure 4 - component Ey and Figure 5 - Ez component.
Fig. 3. The distribution of component Ex (x, y, z) of the electric field 6m below the water line
in the surface x = const for a) 20 sources and b) 10 sources
Fig. 4. The distribution of component Ey (x, y, z) of the electric field 6m below the water line
82
Comparing the distributions of the particular components of the electric field

presented in Figure 3 to 5, one can conclude that the differences in these
distributions decrease along with the increase of the measurement distance, which
is reflected by the various distances x. The distributions of the electric field at x =
7.5m or 10m from the centreline of the hull have slightly different maximum
values, and, virtually, identical shape of the waveform.
Fig. 5. The distribution of component Ez (x, y, z) of the electric field 6m below the water line
3. Summary and conclusions
Based on the calculations of the electric field distribution around the ship, it can
be concluded that the point source system described by formula (1) can be applied
for the purposes of modelling the electric field at a distance greater than the width
of the ship and the depths at least twice larger than the depth of immersion. The
defined above area is called a distant zone. The accuracy of calculations can be
increased by using a more dense division grid of the surfaces of the hull and the
electrodes. Increasing the number of sources, above the limit number - for the
given example above 10, does not affect the accuracy of calculations significantly.
In the zone close to the ship more accurate calculation methods should be applied -
based on the BEM, preferably. It is worth to emphasize that the precision of
calculations using the BEM may be so accurate that it allows to detect relatively
minor damage to the aniticorrossive layer [5]. Thus, the choice of method should
be based on the purpose of the analysis.
The work done within the project development OR00007009 Polish Ministry of
Science and Higer Education.
83
References
[1] Collective work: Protection against electrochemical corrosion. Theory and practice (in
Polish). WNT Warszawa 1991.
[2] Wrobel L. C.: The Boundary Element Method. John Wiley&Sons 2002.
[3] DeGiorgi V. G., Wimmer S.A.: Geometric details and modeling accuracy requirements
for shipboard impressed current cathodic protection system modeling. Engineering
Analysis with Boundary Elements 29 (2005), pp. 15-28.
[4] Riemer D. P., Orazem M. E.: A mathematical model for the cathodic protection of tank
bottoms. Corrosion Science 47 (2005), pp. 849-868.
[5] DeGiorgi V.G.: Evaluation of perfect paint assumptions in modeling of cathodic
protection systems. Engineering Analysis with Boundary Elements 26 (2002), pp.435-
445.
[6] Wrobel L. C., Miltiadou P.: Genetic algorithms for inverse cathodic protection
problems. Engineering Analysis with Boundary Elements. 28 (2004), pp. 267-277.
[7] Amaya K., Aoki S.: Effective boundary element methods in corrosion analysis.
Engineering Analysis with Boundary Elements 27 (2003), pp. 505-519.
[8] Diaz E. S., Adey R.: Optimising the location of anodes in cathodic protection systems
to smooth potential distribution. Advances in Engineering Software 36 (2005), pp. 591-
598.
Computer Applications of Electrical Engineering
Searching of regularities in the development

of electrical power system. Part I - Identification and evaluation
Jerzy Tchórzewski
Siedlce University of Natural Sciences and Humanities
08-110 Siedlce, ul. 3-Maja 54, e-mail: jtchorzewski@interia.pl
The paper contains selected results of research on identification of the development of

the electrical power system, presented in the table in the form of development models th,
matrices of state variables and characteristic polynomials. An algorithm of development
was also presented along with assumptions concerning the evaluation and examination of
regularities in development.
1. Introduction
Examination of regularities in the development of the Electrical Power System

(hereinafter called the EPS system or EPS) comes down to the examination of
structural changes and parametric changes of the system in long time θ [years]. For
this purpose, it is necessary to begin with the identification of the development in
order to obtain a model of development, and then, bind the problem of the system
development with the movement of roots on the complex variable plane s using the
Evans root locus method [35].
Parametric changes on the complex variable plane s are connected with
movement of roots along the existing root loci, and structural changes are
connected with the appearance of new loci or disappearance of the old ones [ 26-
33]. Therefore, the goal of the programming development technique is to find not
only the model of development [2-4, 22, 34] but also the rules governing its
development in long time θ [23]. In the case examined in this paper, the model of
development is a model expressed in the form of the arx model matrices or in the
form of the state variable model, with the parametric changes being connected with
the changes of values of coefficients in polynomials A(q) and B(q) or with changes
of the elements of matrices A, B, C, D in state equations, and the structural changes
being connected with the change of the degree of the polynomial or with the
change of the dimensions of the matrices, especially matrix A, which is responsible
for internal organization of the EPS system. While examining changes in the
control system, it is convenient to study elements of matrix B, which is responsible
for the control effect expressed through it by the influence of control quantities on
the EPS system development. Examination of changes in both matrices A and B is
especially significant from the point of view of technological changes leading to
more flexible functioning of systems and their robotization, as matrix A is a carrier
85
J. Tchórzewski / Searching of regularities in the development of …
of knowledge about the degree of internal organization of the EPS system

development, and matrix B - is a carrier of knowledge about the level of the EPS
system development control.
2. Identification of the EPS system development
To find a new model of the EPS system, identification of the EPS system
development ought to be performed in a specified period of time to define so far used
models of development. Then, appropriate assessment ought to be carried out in
accordance with the theory of system development engineering [26-33]. The
development of a mathematical model of the EPS for the purpose of studying
regularities in the development, using modelling is not possible due to the fact that the
EPS system is a great and highly complex system. Thus, the only way to obtain a
model for the above mentioned purpose is by means of identification, e.g. by means of
interval identification in the stepping system [37]. As a result of identification, th
matrices are obtained, and, following their transformation into the state space in the
MATLAB environment, state equations and output equations that describe the EPS for
the studied periods of development are obtained. From this perspective, states of the
EPS system are described by state vectors, which change their dimension and
composition as well as values in each period of development.
As a result of identification, models of the EPS system development are obtained
for appropriate periods of the EPS system development, which may then be used to
generate a metamodel of development, which is a specific information code of the EPS
system development - an artificial genetic code in the context of genetic algorithms
(fig. 1). The operation of coding development results in obtaining the above mentioned
development code, which, in fact, is a specific genetic model of development, and the
operation of decoding the development of the system leads to obtaining a new state of
the EPS system on the basis of the new artificial genetic code that has been obtained. In
works [26-33], examples how to use the proposed system development engineering as
regards the electrical power system were presented.
The developed conception of system development engineering was verified in
practice. First, a series of processes of identification of the electrical power system
and its sub-systems was performed. Then, the obtained model was used to study
regularities in the development by performing a series of experimental tests in the
MATLAB and Simulink environments using its toolboxes, especially the System
Identification Toolbox and the Control system Toolbox.
Identification of the system is a fundamental issue as regards the theory of
control and systems, especially in a situation, in which a model to be developed is a
model of a great system, and such a model is the electrical power system. When
developing a model of the system in the state space, a state vector that describes the
state of the system is sought. A state of the system may change in long time θ, and,
in this situation, it is possible to speak about a state of the electrical power system
86
(EPS) development in the state space, expressed by the vector of state variables
x(θ), which is the vector with the smallest set of elements xi(θ), which fully define
the results of past influences on the system. If the initial state and the input function
in a certain interval of long time (Δθ) is given, it is sufficient to define states of the
system and its output in this interval of time, and the state of the system
development may not be directly observed.
Systems of development, and similarly, dynamic systems, for which the state of
the system development in long time θ is measured by the derivative of the state
vector x(θ), may be described using a system of n linear differential equations. E.g.
a system of such equations may be presented in the following general form for the
SISO1 object:
dx ( K , θ , t = const )
= f1 ( x ( K , θ , t = const ), u ( K , θ , t = const ), z ( K , θ , t = const ), θ ), (1)
dθ
and supplement it with the output equation:
y ( K , θ , t = const ) = f 2 ( x( K , θ , t = const ), u ( K , θ , t = const ), z ( K , θ , t = const ), θ ), (2)
where: x(K,θ) – system development state variables, u(K,θ) – input variables (input
signal, inputs, input functions), z(K,θ) – interference, K – a set of characteristic
quantities for the electrical power system, which, in case of a developing system
(development of the system) depend on long time θ, e.g. K1(θ), K2(θ), … , Kn(θ) -
they occur in the form of e.g. components of vector K(θ) - a vector of electrical
power system development state, which, i.a. indicates the dynamics of the
electrical power system development. Thus, in dynamic systems as well as in the
systems of development, the current value of the output variable depends not only
on the current values of external stimulating signals but also on the state of the EPS
system development.
In subsequent periods, the examined electrical power system undergoes changes
as regards values Ki(θ) (i=1,2,…,n) in time θ as a result of external interactions
caused e.g. by the inflow of new technical and technological information allowing
for the upgrade of the EPS system, and, consequently, improvement of its safety;
the inflow of new economic information allowing to increase the efficiency of the
electrical power system and information about new requirements that allow to
improve reliability supplied by electrical energy recipients who wish to be supplied
with power that meets quality requirements and parameters on time. For these
reasons, changes that the electrical power system undergoes may conveniently be
measured using the derivative of the vector of the system development state.
A mathematical model may be obtained using modelling provided that it is
possible to use e.g. physical laws related to circuits theory and electro engineering.
In case of difficulties in obtaining a phenomenological model a process of
experimental identification is applied when the inputs and outputs of the system are
available for measuring.
1
SISO – Single Input Single Output, i.e. a control system described using one input variable and one output
variable.
87
Fig. 1. A diagram showing how to obtain models of development and a metamodel

of the DEP system development. Explanations in the text
88
Differential equations (or their discrete equivalents - difference equations) are

used to describe dynamic phenomena in modelled real systems. There exists a
finite number of parameters used in the models, and because of that such models
are called parametric models. In the MATLAB's library named the System
Identification Toolbox differences between individual parametric models result
from the method of considering the influence of interference on the work of the
system. One of these models used in this work is an autoregressive model with an
exogenous input - ARX [37], described by the following equation:
A(q) ⋅ y(θ ) = B(q) ⋅ u (θ − nk ) + e(θ ), (3)
which may be written down in the following form of a difference equation:
y(θ ) = −a1 y(θ −1) − a2 y(θ − 2) − ... − ana y(θ − na) + b1u(θ − 1) + ... + bnb−nku(θ − nb − nk) + e(θ )
e(θ ) = ε (θ ) + a1ε (θ −1) + ... + anaε (θ − na), (4)
where: u(θ) - input of the system observed in a discrete instant of time θ, y(θ) -
output of the system observed in a discrete instant of time θ, v(θ) - a replacement
interference that occurs in a discrete instant of time θ (non-correlated randomly
generated signal of the "white noise" type with a normal distribution and the
average value equal zero and with constant variance), ε - a random variable with
the average value equal zero and constant variance, na - a number of factors
connected with the output signal y, nb - a number of factors connected with the
input signal u, nk - a number of factors connected with the lag between signal y
and signal u, A(q) = 1 + a1q −1 + a2 q −2 + ... + ana q − na - rational functions of the lag
operator q-i of a polynomial connected with y(θ), which have constant coefficients
representing parameters of the model, Bi (q) = b0 + b1q −1 + b2 q −2 + ... + bnb−nk q − ( nb + nk ) -
rational functions of the lag operator q-i of a polynomial connected with u(θ),
which have constant coefficients representing parameters of the model.
ARX model (AutoRegressive with eXogenous input) is defined with the help of
the model structure used for modelling technological object as in practice the ratio
of noise to the signal is small. The function call has the following form: th=arx([y
u],[na nb nk]), where: y, u - i/o column vectors, [na nb nk] - coefficients of the
equation describing the model, 'na' specifies the number of poles (denominator's
roots), 'nb-1' specifies the number of zeros (enumerator's roots), nk - number of
factors connected with the lag between signal y and signal u, th - a matrix of
identification factors with the THETA format. It contains the whole information
about a model, its structure and estimators of parameters together with their
estimation using covariance. This matrix has a strictly specified dimension. The
elements of the first row contain in the following sequence: estimators of the
variance of the parameters, sampling interval, the values of parameters na, nb, nk,
etc. The second row contains: FPE (Final Prediction Error) index, year, month,
day, minute, and numerical code of the command used to generate a specific
model. The third row contains estimators of the model parameters in the
alphabetical order: a1, a2, …, b0, b1, …(with zeros and ones at the beginning of the
89
polynomial being disregarded). Rows 4 to 3+n contain estimations of the

covariance matrix [37].
Therefore, the current value of the output signal depends on a certain number
na of previous values of the output signal y (autoregression), a certain number nb
of successive values of the input signal u, on the time nk between the lag y and u
and on the value of the interference signal (denoted by ε). The structure of the
model is unequivocally specified by three parameters na, nb, nk. The arx model is
stable when all its poles are inside the unit circle on the complex variable plane Z,
and it is minimum phase if this condition is also satisfied by its zeros. It is also
necessary to examine the roots lying on the unit circle. The parametric method used
in the paper is the least squares method. In order to find a model that specifies the
relations between the output signal y and the input signal u, based on N samples
corresponding to successive instants of time kTP (TP – sampling period,
k=1,2,...,N), coefficients of the model expressed as equation (4), which is only
satisfied in approximation when the input signal, the output signal and the
interference are intercorrelated, and estimator Θ for the least squares method is
biased [37]. Due to the fact that the coefficients vector [na nb nk] specifying the
rank of the enumerator and denominator of the transmittance and the lag is
unknown, it is sought according to the criterion of minimization of the error
determined during the verification of the model. In order to analyze a greater
number of the structures of models in a relatively short time, a method which
automatically generates the best results of verification performed on all possible
structures with parameters [na, nb, nk] within the interval <1,10> was used. From
all the structures, the system automatically selects parameters [na, nb, nk] so that
the obtained arx model approximates the expected output signal to the real output
in the best possible way.
3. Evaluation and examination of regularities in the development
A model for the evaluation of the system development contains, i.a.:

- information about the EPS system related to the structure, parameters and
characteristics of the system,
- criterion or a set of criteria for the evaluation of the EPS system development
quality, which contains information about energy and power as a product of the
system,
- algorithm allowing to determine the value of the criterion, which contains
information about the EPS system and about energy and power as a product of
the system.
In the technique of programming development of systems there occur two
groups of problems, namely: problems connected with the formulation of the
model of development and problems connected with determination of laws
governing the changes it undergoes in long time θ (examining regularities in the
90
development of the system). This work deals with problems connected with
working out models of development as well as problems connected with the
specification of parametric and structural changes of the electrical power system in
long time θ. In the control and systems theory, and especially in systems development
engineering, evaluation involves working out a model of development on the basis of
appropriate information about changes in the system. Evaluation criteria, the subject
and the object of evaluation as well as the essence of the evaluation are elements
connected by a certain relation, which make up evaluation, and the relation between the
subject and the objects of evaluation is called a system of evaluation. Different
evaluation criteria are adopted, and these criteria are always tightly connected with the
value of the system. In his work, [21] Piotr Sienkiewicz emphasizes that evaluation
applies to certain values due to specific needs of the system. In the studied case of the
EPS system development, the value of the system is connected with the demand for
power and electric energy, thus, in the evaluation phase it is possible to examine the
difference between the demand for the system development usefulness and the
obtained results as possibilities of the development of the system.
Therefore, evaluation of the development can be brought down to the evaluation
of needs and evaluation of the development potential, and needs of the system of
development as regards the security stream are equal to the financial outlay it can
afford, and the capability of the system of development is equal to its operational
potential as regards power and electrical energy production. Ultimately, the
evaluation of needs connected with development may be reduced to the evaluation
of the system inputs (usefulness income and the potential connected with securing
development), and the evaluation of the development potential may be reduced to
the evaluation of the system outputs (operational potential and outlay on
usefulness) (4). The evaluation of needs and potential of development may be
supplemented with the evaluation of development strength based on the theory of
systems and control (evaluation of power- and information-related complex
quantities). Assuming that the EPS system state is defined by the following
formula:
s( K ,θ , t ) =< Z ( K ,θ , t , u ( K ,θ , t ) >, (5)
and the function Z(K,θ,t) and the function u(K,θ,t) are called indexes of the system
state, and the change of the system state is called system's movement, the sequence
of system changes for the selected instants of time is called a system trajectory [6-
7]. The subject of the study is the EPS system as a technical system or technical-
economic system, in which, over the years, parameters and structure undergo
changes2. An independent variable of the EPS system development is time θ, in the
long aspect, measured in months, years and even decades. It was assumed that the
2
Jacek Malko is of the opinion that "apart from the complexity of the system itself and its relations with the
neighbourhood it is also necessary to consider development dynamics while studying the problem of planning",
and he emphasizes that ..."a phenomenon of object structure change occurs in time, and is connected with the
dynamics of increase in power demand (and with installing new power-generating units) and the process of
ageing and taking operating objects out of active use”[100].
91
EPS system is a system with a developing structure, which evolves in time, at

successive stages of development (1, 2,...,N), corresponding to individual states of
the EPS system development, and each stage comprises the identification phase
and development assessment phase. Successive stages of the EPS system
development should show improvement in time, though it was not necessarily
reflected in reality [18, 20, 22, 24, 26-33].
In order to describe a developing system at each stage of development θ we
assume the theory of developing systems [23] after Robert Staniszewski, and in
case of systems that belong to the same class as the EPS system, it is not possible
to directly provide characteristic quantities because of the fact that a model of
development obtainable using modelling development method is not available as
well. However, a model of development may be obtained as a result of the
identification of the EPS system development, and an artificial genetic code may
be extracted from it, and written in the form of model (evolutionary model), whose
genes are specific cumulated characteristic quantities describing the system.
If one is in possession of the artificial genetic code (evolutionary model), i.e. if one
is in possession of a finite set of values of characteristic quantities of the EPS
system expressed as the set K(θ )=<K1(θ), K2(θ), ..., Kn(θ)>, one should seek a
function of the system quality J(K1, K2, ..., Kn) as a uniform criterion of evaluation
of individual stages of the EPS system development. A set K(θ ) may be expressed
in the state space as a multi-dimensional vector of the EPS system development
state in the following form [17, 26-33]:
K i (θ ) = [ K1 (θ ) K 2 (θ ) ... K n (θ )]T ( 6)
and some components of the vector of the EPS system development state may
constitute probability distributions.
The above formulation gives the possibility to use dynamics of systems
development [1, 4, 6-20, 26-34], with the assumption that the researcher is not
interested in the functioning of the system but in its development and,
consequently, methods that allow to study and examine regularities in the
development. Thus, as such, the EPS system and its subsystems are interesting
from the point of view of the changes in the values of Ki(θ) (i=1-n) in long time θ,
which took place because of deliberate actions taken by the designers, producers,
users, etc. of the system. These actions were taken in the form of control influences
s(θ) as technical influences, economic influences or influences in the form of user
requirements. Changes in the structure of the EPS system may be measured using
the derivative of the state development vector Ki(θ), i.e.
dK i (θ )
= J [( K1 , K 2 ,..., K n ), K i (θ ) = Ci ], (7)
dθ
and: Ci (i=1,2,...,n) specifies the initial state of the EPS system structure, θ – long
time as an independent variable of the EPS system development process, Ki(θ) -
components of the EPS system development state vector.
92
The development of the EPS system in the years 1946-2007 was multi-stage in
nature. Thus, successive states of the EPS system were influenced by previous
states, e.g. resulting from the stream of technical and economic information
connected with various epochs. The pace of changes of the EPS system state vector
also results from the multi-stage nature of changes, and it depends not only on the
current state but also on the form of the system in the past, which may be expressed
as follows:
dK i (θ )
= J [( K (θ ), K1 (θ − θ1 ),..., K n (θ − θ1 )), K 0 (θ ) = C ]. (8)
dθ
It is necessary to emphasize that that the change of the transition process in time
θ, and not the course of the transition process as it is considered in the dynamics of
systems functioning, is important for the processes of the EPS system
development. In the EPS development, there occur two opposing processes,
namely a process of the EPS system operation and a related process of ageing of
the system as well as a process of improvement of the system and related process
of modernization (a process of positive development of the system, modernization).
Thus, if, as a result of the EPS system development, new states of the system
characterized by transition processes gradually becoming gentler are obtained, the
system seems to gradually improve3. Therefore, from the point of view of the
development of the system, the functioning and development of the EPS system
may be expressed jointly in the following way:
dK i ( K , θ , t )
= J [K ( K , θ , t ), K 0 ( K , θ , t ) = C ], (9)
dθ
where: K – a set of parameters of the electrical power system development, θ –
long time of the electrical power system development, t – short time of the
electrical power system operation in one period of development θi.
Thus, the vector of state development in (9) in relation to θ provides
information about development, and in relation to t - information about dynamic
properties connected with the system operation. Thus, there occurs one dynamic
function K(K,θ,t), jointly comprising the dynamics and the development of the
EPS system, and examination of the system was brought down in this work to
states expressed in the following way:
dK i ( K , θ , t = const)
= J [ K ( K , θ , t = const), K 0 ( K , θ , t = const) = C ], (10)
dθ
What seems interesting in this case is the change in the quality of successive
states of the EPS system. And this work focuses on this case of the EPS system
development, i.e. the case connected with its structural and parametric changes for
the situation t=0 or t=const seen from this perspective.
3
In the theory of control and systems it is possible to speak about gentler processes when the shapes of
characteristics become less steep and, consequently, the anticipated responses are more predictable. In the case
of the SEE system operation, the quality of the system becomes worse as θ increases, which is reflected in the
worse shape of the characteristics of the transition process (in short an ageing process occurs).
93
To show a clear-cut difference between the behaviour of characteristics of the

EPS system in the function of operation time t and the behaviour of characteristics
of the system in the function of time of the system structure development, it is
possible to make a simple comparison of behaviour of such characteristics as
deflection, time of the initial process establishment (transition process), or the
number of pulsations in the initial process. It is also worth mentioning that
characteristic quantities are not characteristics of the system operation, and which,
in turn, are not characteristics of the EPS system development. Characteristic
quantities of physical systems are quantities which assume appropriate values and
shape the system, influence and decide about its weight, dimensions, physical
properties, functional usefulness, etc.
In the process of operation (a course of physical process in short time t) these
quantities change as a result of the operation of the system (its wear) and they
change their values. As a result of development (long time θ) the structure of
machines, devices and systems changes, and sometimes, as a result of technology
upgrade, physical process also changes, which results in the changes in the set of
these quantities (the existing quantities disappear and new quantities appear,
properties of the existing quantities change, etc.). Sets of these quantities are
bounded from below and above, which may apply to certain external characteristics
of the physical system (e.g. occurring capacities, inductance, energy characteristics,
etc.) as well as certain characteristics of physical phenomena (e.g. maximum
voltage, permissible frequency, etc.). Depending on the degree to which individual
characteristic quantities influence the above mentioned characteristics, we may
distinguish sensitive and insensitive characteristic quantities. Their positions may
be marked on the characteristics. However, we may also speak about generalized
characteristics, such as efficiency characteristics and characteristics of the safety of
the EPS system development. It is also possible to speak about evaluation
characteristics or ageing as well as characteristics of the electrical power systems
development [23, 26-33].
Moreover, it is necessary to add that the EPS system, seen from this
perspective, should only be treated as a technical object, and it should be used to
study only working processes, i.e. material and energy flows, i.e. power and
electrical energy transfer. One may also study it as an object of management and
distinguish information-decision flows in it - e.g. functioning of the Energy
Regulatory Office, etc. as a regulation system for the EPS system. Finally, it can be
seen as a control system, in which it is possible to distinguish processes of energy
flow and transfer services flow as well as processes connected with the flow of
financial means.
It was assumed that the EPS system comprises control processes, i.e. takes into
account processes connected with physical flow of energy and services (realization
system) and oppositely directed processes connected with the flow of information
and financial stream (control system), which allows to distinguish in the EPS
94
development model energy and material processes as well as oppositely directed

information and financial processes. The development of the EPS system presented
in this way comes down to examining regularities in the development of the EPS
system treated as a technical-economic system, where: u = {u1 , u2 ,..., un } - a set of
quantities influencing the system e.g. a set of input quantities, y = { y1 , y2 ,..., yn } - a set
of external quantities of the system e.g. output quantities.
Thus, based on the theory of control and systems, the EPS system is composed
of two sub-systems: a control system and the object of control, and in the object of
control there occur working (technological) processes and in the control system -
processes responsible for the organization of operation, i.e. processes that decide
about the degree to which working (financial) processes are used.
When studying regularities in the EPS system development over years, i.e.
taking historical data into account, it is significant to examine individual
characteristic quantities as regards sensitivity to structural changes and parametric
changes. It may turn out that small change in the value of the quantity Ki(θ) may
result in significant improvement in the quality of the EPS system development,
and, in another case, even a significant change of these quantities may not result in
any changes at all, and even result in the deterioration of the quality of the system.
Therefore, it is important to determine, which artificial genes, i.e. which values
Ki(θ) are able to improve the quality of the system, and, to what extent. Among
problems connected with the EPS system development, there is a problem
connected with control of development using specified characteristic quantities, in
this case using specified artificial genes. Mathematical description of the EPS
system development control, as a technical-economic system, in accordance with
information about technical, economic changes and about the change of the EPS
system users requirements that appear in specified time is important. Hence, the
following model of the EPS system development may be adopted in order to jointly
present working and control processes:
dK
= J [ K (θ ), s (θ ), r (θ )] (11)
dθ
where: s(θ) – control vector, r(θ) – vector containing technical and economic
information, as well as information concerning users requirements (interference
vector).
Examination of the EPS system development state changes in the state space
expressed using (11) depends on the current value of the development state
described by vector K(θ), the control vector (usually external input functions) s(θ)
and the vector of technical, economic and functional information r(θ), which can be
jointly written in the following form:
dK ( K , θ )
= A( K , θ ) ⋅ K (θ ) + D( K , θ ) ⋅ s(θ ) + r (θ ), (12)
dθ
95
where: A(K,θ) – development process matrix, D(K,θ) – control matrix, K(θ) – state
vector.
If the process is of n-th rank, then matrix A is a square matrix n x n with
constant coefficients, and the process is linear and non-stationary process.
However, when the process and control matrices depend on time θ, the process is
linear and the parameters are random.
It is worth noticing here that there occurs a special case for short time t=const
(dependence 15), i.e. A(K,θ) is a square matrix n x n with constant coefficients,
which ensure linearity (in relation to t - which results from the formula). It may be
added that in relation to t, the process is a stationary process, with the parameters
independent of short time t for a given stage of development θi, but the process is a
non-stationary process in long time θ (the matrix of process A and control matrix
B depend on long time θ) or a process with random parameters if the above
mentioned matrices depend on random values of process development parameters.
If matrices A and B do not depend on time θ, the process is stationary in relation to
long time θ.
Examination of regularities in the EPS system development is not about solving
equation (15) with the set initial conditions but it is about defining the model on the
basis of the process of system identification, and finding artificial genetic code
based directly on the model of the system, as well as studying sensitivity of
changes in the model quality depending on changes of individual parameters and
the EPS system structure.
In case of the development of the EPS system, understood as an integrated
automated system (a system that develops in the same way as unmanned factory)
work quality may be determined e.g. by the degree of control error (deviation)
during the whole period of the EPS system operation, i.e. in long time θ. However,
due to random nature of interference, analytical determination of the real course of
the control error is not possible. Thus, it is convenient to perform the evaluation of
the system development quality based on the features and parameters of processes
that occur as a result of certain typical input functions, such as unit step, sinusoidal
function, or other typical input functions. The quality (goodness) of control
systems is evaluated using appropriately adjusted indexes of quality expressing
technological, economic and other requirements the system must meet (in the
theory of control and systems they include, i.a. the following regulation quality
criteria: stability margin, distribution of the characteristic equation roots, time,
frequency and integral criteria).
4. Models of the DEP system development
It is convenient to bring down the examination of regularities in the electrical

power system (EPS system or EPS) development to studying structural changes
and parametric changes of the system in long time θ. For this purpose, a problem of
96
the EPS development may be bound with movement of roots on the complex
variable plane s using the method of Evans root loci [35].
Parametric changes do not undermine the state of safe development of the
system and the EPS system model. However, structural changes introduce an
unsafe state of the development, both as regards the EPS system and model, which
may result in a sudden change of characteristics, in the number and value of the
state variable quantities as well as system's inputs and outputs.
Examination of stability of linear continuous and pulse systems comes down to
examining the positions of characteristic equation roots using appropriate criteria
of stability [7, 23, 35-36], which showed that some of the models obtained as a
result of the identification process using data for years 1946-2007 [24], both arx
and corresponding models in the state space, were unstable. These models as
models of the EPS system development and models of the EPS model of
development (metamodels) were discussed in detail in works [26-33]. Therefore, in
this work, the author will only discuss certain tendencies resulting from these
models, obtained based on the examination of the behaviour of the course of Evans
root loci.
Due to the fact that in the obtained models of the EPS development (table I)
some proper values (eigenvalues) did not have negative real parts, these models are
unstable. Usually, stability of closed models, and not open models is examined.
Therefore, the stability of the EPS development as a closed system was assessed in
the work, with denominator's roots being assumed poles and roots of the
transmittance enumerator being assumed zeros. Then, the problem of the EPS
system development was bound with the movement of roots on the complex
variable plane s using Evans root loci method, and making amplification factor
variable. Movement of roots along the existing root locus was obtained when the
parameters of the system were changed (k as a cumulated amplification parameter)
and appearance or disappearance of root loci in case of structural changes was
observed.
Unstable development may be caused by the physical nature of the EPS
development itself (some physical processes in the system may have a tendency
towards unstable development, e.g. imposed emission restrictions, construction
errors of the systems and elements of the EPS as a technical system, technological
errors, especially those resulting from cooperation of new technologies connected
with renewable power engineering with the existing elements of the EPS system,
etc.
97
Table I. Models of the EPS system development

FOR THE OUTPUT ELECTRICAL POWER CONSUMPTION AND 14 INPUTS
(YEARS 1946-2007)
arx111 A(q) = 1 + 0.1786q^-1, B1(q) = 2.162q^-1, B2(q) = 5.626q^-1, B3(q )=
90,1% 223.9q^-1, B4(q) = -2.289q^-1,bB5(q) = 1.255q^-1
u1 B6(q) = 0.2639q^-1, B7(q) = -10.66q^-1, B8(q) = -0.003189q^-1, B9(q)
y21 = 1.261q^-1, B10(q) = -2.09q^-1, B11(q) = -1.822q^-1, B12(q) = -
46-75 4.284q^-1, B13(q) = 27.43q^-1B14(q) = -18.55q^-1, ans 5 = [− 0,1786 ]
B1 = [2 .162 5 .626 223 .8853 − 2 .2886 1 .2551 0 .2639 − 10 .6557 − 0 .0032 1 .2615 − 2 .0896 − 1 .8219 − 4 .2842 27
A1 = [− 0.1786] C1 = 1 [ ], D1 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0] , 1
CE1 = 2.162 ⋅
( s + 0.1786)( s − 2.1620)
arx113 A(q) = 1 - 1.554q^-1, B1(q) = 0.0157q^-3, B2(q) = -1.113q^-3, B3(q) = -

98,7% 106.2q^-3, B4(q) = -78.44q^-3, B5(q) = -0.4607q^-3, B6(q) = 0.05317q^-
u5 3, B7(q) = 0.3395q^-3, B8(q) = 0.05702q^-3, B9(q) = -0.3424q^-3, B10(q)
y25 = -0.4439q^-3, B11(q) = -0.3456q^-3, B12(q) = -4.166q^-3, B13(q) =
50-79 6.395q^-3, B14(q) = -0.923q^-3, ans 5 = [1,5545 0 0]
⎡ 0 0 0 0 0 0 0 0 0 0 0 0
B5 = ⎢⎢ 0 0 0 0 0 0 0 0 0 0 0 0
⎣⎢0,9157 − 1,1125 − 106,2195 − 78,4388 − 0,4607 0,0532 0,3395 0,0570 − 0,3424 − 0,4439 − 0,3456 − 4,1656 6
⎡1,3345 1 0⎤ C5 = [1 0 0] , D5 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0]
A5 = ⎢⎢ 0 0 1⎥⎥
⎣⎢ 0 0 0⎥⎦
1
C E 5 = 0,0157 ⋅
s 2 ( s + 1,5545)
A(q) = 1 - 1.766q^-1, B1(q) = 0.03545q^-2, B2(q) = 0.06842q^-2, B3(q) =
arx112 98,6%
83.57q^-2, B4(q) = -1.584q^-2, B5(q) = 0.1385q^-2,
u10
B6(q) = -0.09277q^-2, B7(q) = -2.464q^-2, B8(q) = 0.02396q^-2, B9(q) =
y210
-0.2675q^-2, B10(q) = -2.324q^-2, B11(q) = -0.7672q^-2, B12(q) =
55-84
2.762q^-2, B13(q) = 17.04q^-2 , B14(q) = -4.71q^-2
⎡ 0 0 0 0 0 0 0 0 0 0 0 0
B10 = ⎢
⎣0,0354 0,0684 83,5675 − 1,5837 0,1385 − 0,0928 − 2,4637 0,0240 − 0,2675 − 2,3241 − 0,7672 2,7618 17
⎡1,7656 1⎤ C10 = [1 0] D10 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0] 1

A10 = ⎢ , , C E1 = 0,0354 ⋅
0⎥⎦
,
⎣ 0 s( s + 1,7656)
ans 1 = [1, 7656 0]
A(q) = 1 + 0.1786q^-1B1(q) = 2.162q^-1B2(q) = 5.626q^-1B3(q) =
arx111 223.9q^-1B4(q) = -2.289q^-1B5(q) = 1.255q^-1, B6(q) = 0.2639q^-1,
90,067% u15 B7(q) = -10.66q^-1, B8(q) = -0.003189q^-1, B9(q) = 1.261q^-1, B10(q) =
y215 -2.09q^-1, B11(q) = -1.822q^-1, B12(q) = -4.284q^-1, B13(q) = 27.43q^-
61-90 1B14(q) = -18.55q^-1
A15 = [− 0,1786]
B15 = [2,162 5,626 223 ,8853 − 2, 2886 1,2551 0, 2639 − 10 ,6557 − 0,0032 1, 2615 − 2,0896 − 1,8219 − 4,2842 27 , 43
C15 = [1] , D15 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0] ,
98
C E15 = 2,1620 ⋅
1 ans 15 = [− 0 .1786 ]
( s − 0,1786)
A(q) = 1 - 1.554q^-1, B1(q) = 0.0157q^-3, B2(q) = -1.113q^-3, B3(q) = -

arx113
106.2q^-3, B4(q) = -78.44q^-3,
98,7166%u20
B5(q) = -0.4607q^-3, B6(q) = 0.05317q^-3, B7(q) = 0.3395q^-3, B8(q) =
y220
0.05702q^-3, B9(q) = -0.3424q^-3, B10(q)=0.4439q^-3, B11(q) = -0.3456
66-95
^-3, B12(q) = -4.166q^-3, B13(q) = 6.395q^-3, B14(q) = -0.923q^-3
⎡ 0 0 0 0 0 0 0 0 0 0 0 0
B 20 = ⎢⎢ 0 0 0 0 0 0 0 0 0 0 0 0
⎢⎣0,0157 − 1,1125 − 106,2195 − 78,4388 − 0,4607 0,0532 0,3395 0,057 − 0,3424 − 0,4439 − 0,3456 − 4,1656
⎡1,5545 1 0 ⎤ C20 = [1 0 0] D20 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0] 1
C E 20 = 0,0157 ⋅ 2
A 20 = ⎢⎢ 0 0 1 ⎥⎥ s ( s + 1,5545)
⎢⎣ 0 0 0 ⎥⎦
ans 20 = [1,5545 0 0]
A(q) = 1 + 0.1786q^-1, B1(q) = 2.162q^-1, B2(q) = 5.626q^-1, B3(q) =
arx111 223.9q^-1, B4(q) = -2.289q^-1,
90.1% B5(q) = 1.255q^-1, B6(q) = 0.2639q^-1, B7(q) = -10.66q^-1, B8(q) = -
u25 0.003189q^-1, B9(q) = 1.261q^-1,
y225 B10(q) = -2.09q^-1, B11(q) = -1.822q^-1, B12(q) = -4.284q^-1, B13(q) =
71-00 27.43q^-1B14(q) = -18.55q^-1
B 25 = [2,162 5,626 223,8853 − 2,2886 1,2551 0,2639 − 10,6557 − 0,0032 1,2615 − 2,0896 − 1,8219 − 4,2842
A25 = [− 0,1786] , C25 = [1] D25 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0] 1
C E 25 = 2,1620 ⋅
( s − 0,1786)
ans 25 = [− 0 .1786 ]
arx111 A(q) = 1 + 0.1786q^-1, B1(q) = 2.162q^-1, B2(q) = 5.626q^-1, B3(q) =
90,067% 223.9q^-1, B4(q) = -2.289q^-1, B5(q) = 1.255q^-1, B6(q) = 0.2639q^-1,
u30 B7(q) = -10.66q^-1, B8(q) = -0.003189q^-1, B9(q) = 1.261q^-1, B10(q) =
y230 -2.09q^-1, B11(q) = -1.822q^-1, B12(q) = -4.284q^-1, B13(q) = 27.43q^-
76-05 1B14(q) = -18.55q^-1
B 30 = [2,162 5,626 223,8853 − 2,2886 1,2551 0,2639 − 10 ,6557 − 0,0032 1,2615 − 2,0896 − 1,8219 − 4,2842 27
A30 = [− 0,1786], C30 = [1], D30 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0], 1
C E 30 = 2,1620 ⋅ ,
( s − 0,1786)
ans 30 = [− 0 . 1786 ]
5. Summary and directions of further research
The work shows that it is possible to obtain a model of development of the

domestic electrical power system using stepping identification, which is connected
with the preparation of appropriate numeric input and output data of the system,
initial processing of data, selection of the arx model and its structure
(determination of na, nb, nk) for each stepping model, determination of parameters
evaluation (selection of the appropriate estimation algorithm), performing the
verification of the model e.g. by comparing the signal obtained from the model
with the real signal and interpretation of the model obtained as a result of
identification.
99
As a result of identification, 33 arx models were obtained in the stepping

system, which were then transformed into state variable models. It is important to
examine the rank of matrices in the models of state variables and changes of their
elements while studying regularities and tendencies in the development. The
second part entitled "Part 2. Analysis and interpretation of development" is a
continuation of this work.
References
[1] Bolkowska B.: Planowanie systemu elektroenergetycznego w ujęciu teorii sterowania.

Archiwum Energetyki nr 4/1974.
[2] Brown M., Do Coutto Filho M., Cesar Stacchini de Souza J.: Forecasting-Aided State
Estimation – Part I: Panorama, IEEE Transactions on Power Systems, Vol. 24, No. 4,
Nov. 2009, pp. 1667-1677.
[3] BrownM., Do Coutto Filho M., Cesar Stacchini de Souza J., Sergio Freund R.,
Forecasting-Aided State Estimation – Part II: Implementation, IEEE Transactions on
Power Systems, Vol. 24, No. 4, Nov. 2009, pp. 1678-1685.
[4] Cegielski M, Kremens Z., Soberajski M.: Systemy elektroenergetyczne –
teraźniejszość i przyszłość. Archiwum Energetyki nr. 1-2/1997.
[5] Crandall J., Cummings M. L.: Identifying Predictive Metrics for Supervisory Control
of Multiple Robots. IEEE Transaction on Robotics. Vol. 23, Issue 5, Oct. 2007, pp.
942-951.
[6] Kacejko P.: Analiza systemu elektroenergtycznego w ujęciu obiektowym. PN PW.
Elektryka z. 104/1998.
[7] Kaczorek T., Dzieliński A., Dąbrowski W., Łopatka R.: Podstawy teorii sterowania.
WNT, Warszawa 2005.
[8] Kłos A.: Planowanie rozwoju systemu elektroenergetycznego. Poradnik Inżyniera
Elektryka. Tom IV. Elektroenergetyka. Warszawa 1975.
[9] Krawiec F.: Ewolucja planowania rozwoju elektroenergetyki w ramach deregulacji.
Zeszyty Naukowe, Wyd. Adam Marszałek. Toruń Vol. 6, 1998.
[10] Kremens Z., Sobierajski M.: Analiza układów elektroenergetycznych. WNT.
Warszawa 1996.
[11] Kulczycki J.: Optymalizacja struktur sieci elektroenergetycznych. Warszawa. WNT
1990, s. 191.
[12] Kulikowski R.: Sterowanie w wielkich systemach. WNT. Warszawa 1974.
[13] Kwaśnicki W.: Ewolucyjny model rozwoju przemysłu – perspektywy badawcze
i dydaktyczne. Ekonomista 4, 2000.
[14] Machowski J., Bialik J.: Power System Dynamice and Stability. J. Wiley 1996.
[15] Malko J.: Planowanie systemów elektroenergetycznych. PWN. Warszawa 1976.
[16] Malko J., Otręba L., Skorupski W.: Identyfikacja modelu autoregresji-średniej
ruchomej krótkoterminowej predykcji procesu zapotrzebowania mocy. Archiwum
Energetyki, Nr 1-2 1990, s. 101-108.
[17] Mulawka J.: Systemy ekspertowe. WNT. Warszawa 1996.
[18] Niebrzydowski J.: Problemy rozwoju systemów elektroenergetycznych. Wyd. PB.
Białystok 1989.
[19] Paska J.: Niezawodność systemów elektroenergetycznych. OW PW, Warszawa 2005
100
[20] Rebizant W.: Metody inteligentne w automatyce zabezpieczeniowej. Monografie Nr

29 (93). OW PWr, Wrocław 2004.
[21] Sienkiewicz P.: Cybernetyczna teoria systemów rozwijających się. [w:] Sztuczna
Inteligencja i Rozwój Systemów. Tom 1. ZG PTC – AP. Warszawa – Siedlce 1988.
[22] Sobierajski M., Wikosz K.: Sieci elektroenergetyczne a rynki energii elektrycznej.
Problemy i Perspektywy. Prace Naukowe Instytutu Energoelektryki Politechniki
Wrocławskiej. OW PWr. Wroclaw 2000.
[23] Staniszewski R., Sterowanie procesem eksploatacji. WNT. Warszawa 1990.
[24] Statystyka Elektroenergetyki Polskiej. Wydawnictwo Zjednoczenia Energetyków,
Centrum Informatyki Energetyki, ARE S.A., Warszawa 1952-2004.
[25] Tadeusiewicz R.: Badanie właściwości układów samodzielnych współpracujących ze
stochastycznie zmiennym środowiskiem. Postępy Cybernetyki. 4/1976.
[26] Tchórzewski J.: A Model of Electric Power System from the Point of View of its
Growght. Development of the internal organization and control level on the basis of
numerical data for the years 1999-2008. IEEE Modern Electric Power System,
MEPS’2010, paper 08.3, Wroclaw 2010.
[27] Tchórzewski J.: Identification of Electric Power System from the Point of View of
Development. Polish Journal of Environmental Studies. Vol. 18, No. 4B/2009.
[28] Tchórzewski J.: The Development Process of Electrical Power System as Unmanned
Factory. Electrical Engineering, Poznan University of Technology. Academic
Journals, No 59/2009.
[29] Tchórzewski J.: Development models of electric power system as a in MATLAB
environment using to System Identification Toolbox. Materiały XIV Międzynarodowej
Konferencji Naukowej pt. „Aktualne Problemy w Elektroenergetyce”. PG, Jurata 2009.
[30] Tchórzewski J.: Systemowe wspomaganie procesu badania prawidłowości rozwoju
systemu sterowania na przykładzie elektroenergetycznej sieci przesyłowej. Rozprawa
nr. 58. A.P. Siedlce 1999.
[31] Tchórzewski J.: Badanie prawidłowości rozwoju systemu elektroenergetycznego. Cz.
I. Wybrane modele rozwoju. Computer Applications in Electrical Engineering. IEEE
PP, EEC PAN, IEEE Poland Section, PP, Poznań 2011, pp. 71-72.
[32] Tchórzewski J.: Badanie prawidłowości rozwoju systemu elektroenergetycznego. Part
2. Interpretacja rozwoju z wykorzystaniem linii pierwiastkowych Evansa. Computer
Applications in Electrical Engineering. IEEP PP, EEC PAN, IEEE Poland Section,
PP, Poznań 2011, pp.73-74.
[33] Tchórzewski J.: Electric Power System from Point of View of Model Development.
[in:] Computer Applications in Electrical Engineering [ed.] Nawrowski R., POLI-
GRAF-JAK, Vol. 8, Ponan 2010.
[34] Trzaska Z.: A Modelowanie i symulacja układów dynamicznych. Wyd. PW.
Warszawa 1993.
[35] Wajs K.: Linie pierwiastkowe Evansa w automatyce. PWN. Warszawa 1973.
[36] Zajczyk R.: Modele matematyczne systemu elektroenergetycznego do badania
elektromechanicznych stanów nieustalonych i procesów regulacyjnych. Wyd. PG.
Gdańsk 2003.
[37] Zimmer A., Englot A.: Identyfikacja obiektów i sygnałów. Teoria i praktyka dla
użytkowników MATLABA, Politechnika Krakowska. Kraków 2005.
101
Searching of regularities in the development

of electrical power system. Part II – Analysis and interpretation
of changes
Jerzy Tchórzewski
Siedlce University of Natural Sciences and Humanities
08-110 Siedlce, ul. 3-Maja 54, e-mail: jtchorzewski@interia.pl
The paper contains selected results of research on searching of regularities of

development of the electrical power system. Analysis and interpretation of development
were also presented on the example of the output of electrical power consumption type (y2)
and 14 inputs (years 1946-2007). This direction of research also includes seeking the
interpretation of the EPS development using Evans root line.
1. Introduction
In the literature on the subject, attention is increasingly often paid to the need of
seeking future states of the Electrical Power System (EPS) development by modelling
or performing identification of the system in order to develop its model [1-3, 5, 8-11,
13]. This direction of research also includes seeking the interpretation of the EPS
development using Evans root line. Based on the analysis of Evans root line used in
automatics [8-12], movement of roots along the existing paths is connected with
parametric changes, and appearance as well as disappearance of the existing root line is
the evidence of the occurrence of structural changes. Therefore, parametric changes
may be observed by following the changes of the elements of the state variable model
matrices, as well as structural changes - by observing the number of zeros and poles of
transmittance or dimensions of matrices A, B, C and D respectively.
This work uses both methods. 8 out of 33 models were used to present the research
results in order to clearly show a method of analysis of changes using an analytical
example [8-11]. As a result, an attempt was also made to formulate, based on the
obtained results, regularities and tendencies in the EPS system development in long
time θ [6]. Due to the fact that in the state variable model, matrix A is responsible for
internal organization of the EPS system, and matrix B - for the level of control,
elements of both matrices mentioned above have been analyzed in detail. [4, 6, 8-13].
2. Selected evans root line
Research experiment selected for interpretation is connected with the subsystem

of electrical power consumption, assumed as output variable y2 depending on 14
input variables [7-11]:
102
J. Tchórzewski / Searching regularity of development of …
u1 - employment in power plants (total) [persons],

u2 - power installed in power plants [MW],
u3 - the number of turbine sets [pcs],
u4 - the number of power boilers (total) [pcs],
u5 - the number of transformers in professional power industry [pcs],
u6 - the number of substations [pcs],
u7 - the number of switches [pcs],
u8 - length of overhead power lines (all voltages in total) [km],
u9 - length of cable lines (with service lines (connectors)) [km],
u10 - consumption of hard coal (total) [thousand tons],
u11 - consumption of brown coal (total) [thousand tons],
u12 - consumption of gas fuels (total) [thousand m3],
u13 - consumption of other raw materials in professional power plants, including
liquid fuels (total) [TJ],
u14 - import of electrical power (total) [GWh].
Models selected for the analysis related to the following periods: 1946-1975,
1950-1979, 1955-1984, 1961-1990, 1966-1995, 1971-2000, 1976-2005 and 1978-
2007. They were labelled: M1, M5, M10, M15, M20, M25, M30 and M33
respectively. Basic characteristics th, courses of Evans root loci and selected
elements of models were presented in table 1 [8].
3. Analysis of the electrical power system development
It can be inferred from the analysis of the course of Evans root line that there
occurred two kinds of models, one kind with the course of one Evans root line
(models: M1, M15, M25, M30) and the other kind with the course of 5 Evens root
line (models: M5, M10, M20, M33).
The first kind of model indicates that the EPS was a very stable system
(characteristic runs from the pole located in point - 0.1786 on the real numbers axis
and approaches - ∞ along it. Such a situation occurred in Poland in the years: 1946-
1979 (model M1), 1961-1990 (model M15), 1971-2000 (model M25) and 1976-
2005 (model M30).
However, the second type of the obtained models were unstable because 4 root
line were located on the right semiplane of the complex variable s, and only one
characteristic runs along the axis of real numbers in the left semiplane from the
pole located at the zero point of the coordinate system to - ∞.
It may be noticed that two kinds of characteristic equations occur, which have
the following form:
CI ( s) = s + 0.1786, (1)
and
CII (s) = s 3 + 1.5545⋅ s. (2)
103
Table 1. Selected characteristics of EPS for the output of electrical power consumption
type (y2) and 14 inputs (years 1946-2007). Source: Author's own compilation
I type of Evans root loci plots
M1
M15
M25
M30
Selected elements of models

M1 CE1 = 2.162⋅
1
,
M15 s + 0.1786
M25 A1 = [− 0.1786], C1 = [1], D1 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0]
M30 B1 = [2.162 5.625 223.8853 − 2.2886 1.2551 0,2639 − 10.6557 − 0.0032 1.2615 − 2.0896 − 1.8219 − 4.2842 27.4339 − 18.5515]
Thus, parametric changes took place during the analyzed periods, and structural
changes occurred between the periods. We may jointly call them parametric and
structural changes resulting from equations (1) and (2).
Writing equation (1) in the form adequate to equation (2) the following is
obtained:
C II ( s ) = a3 ⋅ s 3 + 0,1786 ⋅ s + a0 . (3)
From the comparison of equations (2) and (3) the following regularities may be
obtained:
- a structural change described by the coefficient a3 =1, a2=0 occurred,
- parametric changes occurred described by the changes of values: a1 from value
1 to 1.5545 and a0 from value 0.1786 to 0.
Other changes of the values of elements of both matrix A and B, which result in
certain regularities in development were presented using selected examples in Fig. 1
(for elements of matrix A) and in Fig. 2 (for elements of matrix B).
104
Table 2. Selected characteristics of the EPS for the output of electrical power consumption
type (y2) and 14 inputs (years 1946-2007). Source: Author's own compilation
II type of Evans root loci plots
M5
M10
M20
M33
1 ⎡1,3345 1 0⎤
C E 5 = 0,0157 ⋅ ,
s 2 ( s + 1,5545) A5 = ⎢⎢ 0 0 1⎥⎥
⎢⎣ 0 0 0⎥⎦
M5 C5 = [1 0 0] D5 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0]
⎡ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ⎤
B5 = ⎢⎢ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ⎥⎥
⎢⎣0,9157 −1,1125 −106,2195 −78,4388 −0,4607 0,0532 0,3395 0,0570 −0,3424 −0,4439 −0,3456 −4,1656 6,3946 −0,923⎥⎦
1 ⎡1,7656 1⎤ C10 = [1 0]
CE1 = 0,0354 ⋅ A10 = ⎢
s( s + 1,7656) ⎣ 0 0⎥⎦
M10 D10 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0]
⎡ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ⎤
B10= ⎢ ⎥
⎣0,0354 0,0684 83,5675 −1,5837 0,1385 − 0,0928 − 2,4637 0,0240 − 0,2675 − 2,3241 − 0,7672 2,7618 17,0352 − 4,7102⎦
1 ⎡1,7656 1⎤
CE1 = 0,0354 ⋅ A10 = ⎢
s( s + 1,7656) ⎣ 0 0⎥⎦
M20 C10 = [1 0] D10 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0]
⎡ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ⎤
B10= ⎢ ⎥
⎣0,0354 0,0684 83,5675 −1,5837 0,1385 − 0,0928 − 2,4637 0,0240 − 0,2675 − 2,3241 − 0,7672 2,7618 17,0352 − 4,7102⎦
1 ⎡1,7656 1⎤
C E 33 = 0,0354 ⋅ , A33 = ⎢
s ( s + 1,7656) ⎣ 0 0⎥⎦
M33 C33 = [1 0] D33 = [0 0 0 0 0 0 0 0 0 0 0 0 0 0]
⎡ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ⎤
B33= ⎢ ⎥
⎣0,0354 0,0684 83,5675 −1,5837 0,1385 − 0,0928 − 2,4637 0,0240 −0,2675 − 2,3241 −0,7672 2,7618 17,0352 − 4,7102
⎦
They include the following, i.a. regularities in structural development:

- changes in the rank of matrices A and B (structural changes) from model:
105
1) M1 (A [1 x 1], B [1 x 14]) into model M5 (A [3 x 3], B [3 x 14]),

4) M15 ( A [1 x 1], B [1 x 14]) into model M20 (A [ 3 x 3], B [3 x 14]),
5) M20 (A [ 3 x 3], B [3 x 14]) into model M25 (A [1 x 1], B [1 x 14]),
6) M30 (A [1 x 1], B [1 x 14]) into model M33 (A [3 x 3], B [3 x 14])
respectively,
6 structural changes took place - in all discussed changes but one, namely the
transition from model M25 into model M30, when a structural change did not
occur.
- changes in the value (parametric changes) e.g.:
1) of element a11 of matrix A from the value:
a. -0.1786 in M1 into 1.3345 in M5,
b. 1.3345 in M5 into 1.7656 in M10,
c. 1.7656 in M10 into -0.1786 in M15,
d. -0.1786 in M15 into 1.5545 in M20,
e. 1.5545 in M20 into -0.1786 in M25,
f. 0.1786 in M30 into 1.7656 in M33.
and parametric changes characterized by element a11 did not occur during
the change from model M25 into M30 (the values did not change),
2) of element b11 of matrix B from the value 2.162 in M1 into 0.0000 in M5,
and parametric changes did not occur afterwards (the value of element b11
matrix B was invariable and equal to 0.0000).
Zmiany wartości elementów macierzy A
1,5
wartośc elementu macierzy A
0,5
0
M1 M5 M10 M15 M20 M25 M30 M33
-0,5
model
a11 a12 a23
Fig. 1. Changes in the value of matrix A elements. Symbols: M1, M5,..., M33 – values of matrix A
elements for subsequent models. Source: Author's own compilation
Interpretation of state variables in individual models undergoing changes as

well as interpretation of elements of matrices A and B, in which there occur
changes in the structure and value is interesting.
106
Due to the fact that state variable x1 is connected with output variable y2
representing consumption of electrical power in model M1, thus, assuming that the
quantity that binds them, c11 is dimensionless quantity, we may assume electrical
energy consumption [kWh] as state variable x1, and consequently, element a11 of
matrix A may be expressed in units [1/h] and may be interpreted as a period of
changes as regards electrical power consumption.
In turn, due to the fact that input variable u1 represents employment in power
plants [persons], element b11 of matrix B may be expressed in units [kW/person]
and may be interpreted as electrical power consumption for one person producing
electrical power. As it can inferred from the course of changes in the values of
element b11, cyclic changes in the value of this coefficient took place in the
domestic electrical power system.
Zmiany wartości elementów macierzy B
2,5
1,5
0,5
0
M1 M5 M10 M15 M20 M25 M30 M33
b11 b12 b13
Fig. 2. Changes in the value of matrix B elements. Symbols: M1, M5,..., M33 – values of matrix B
elements for subsequent models. Source: Author's own compilation
Models: M1, M15, M25 i M30 was described in state space as follow:
•
x1(K,θ ) = −0,1786⋅ x1 + 2,1620⋅ u1 + 5,6260⋅ u2 + 223,8850⋅ u3 − 2,2886⋅ u4 +1,2551⋅ u5 +
+ 0,2639⋅ u6 −10,6557⋅ u7 − 0,0032⋅ u8 +1,2615⋅ u9 − 2,0896⋅ u10 −1,8219⋅ u11 + (4)
− 4,2842⋅ u12 + 27,4339⋅ u13 −18,5515⋅ u14,
y21(K,θ ) = y215(K,θ ) = y225(K,θ ) = y230(K,θ ) = x1,
for which the block diagram of the state variables have the form as in Fig. 3.
107
u1 2,1620
u2 5,6260
223.8853
-2,2886
1,2551 -0,1786
•
1.5545
x1 x1 y2
-10,6557 ∫ 1,000
-0,0032
1,2615
-2,0896
-1,8219
-4,2842
27,4339
u1 -18,5515
Fig. 3. The flowchart of the state variables for M1 model (years 1946-1975), where: y21 – electrical
power consumption, u1 – u14 – input variable
and subsequent models hale the but following equations in state space:
M5:
•
x1(K,θ ) = 1,3345⋅ x1 + x2 ,
•
x2 (K,θ ) = x3 ,
•
x3 (K,θ ) = 0,9157⋅ u1 −1,1125⋅ u2 −106,2195⋅u3−78,4388⋅ u4 − 0,4607⋅ u5 + 0,0532⋅ u6 + (5)
+ 0,3395⋅ u7 + 0,0570⋅ u8 − 0,3424⋅ u9 − 0,4439⋅ u10 − 0,3456⋅ u11 − 4,1656⋅ u12 +
+ 6,3946⋅ u13 − 0,9230⋅ u14,
y25(K,θ ) = x1,
108
M10 and M33:

•
x1(K,θ ) = 1,7656⋅ x1 + x2 ,
•
x2 (K,θ ) = 0,0354⋅ u1 + 0,0684⋅ u2 + 83,5675⋅u3−1,5837⋅ u4 + 0,1385⋅ u5 − 0,0928⋅ u6 + (6)
− 2,4637⋅ u7 + 0,0240⋅ u8 − 0,2675⋅ u9 − 2,3241⋅ u10 − 0,7672⋅ u11 + 2,7618⋅ u12 +
+17,0352⋅ u13 − 4,7102⋅ u14,
y210(K,θ ) = y233(K,θ ) ⋅ x1,

M20:
•
x1(K,θ ) = 1,5545⋅ x1 + x2 ,
•
x2 (K,θ ) = x3
•
x3 (K,θ ) = 0,0157⋅ u1 −1,1125⋅ u2 −106,2195⋅u3−78,4388⋅ u4 − 0,4607⋅ u5 + 0,0532⋅ u6 + (7)
+ 0,3395⋅ u7 + 0,0570⋅ u8 − 0,3424⋅ u9 − 0,4439⋅ u10 − 0,3456⋅ u11 − 4.1656⋅ u12 +
+ 6,3946⋅ u13 − 0,9230⋅ u14,
y220(K,θ ) = x1.
have the form as in Fig. 4.
u1
0,0157
u2
-1,1125
-106,21
-0,4607
• •
0,0532
x3 x2 = x3 X2
•
x1 y21
0,3395
∫ ∫ x1 ∫ 1,000
0,957
-0,3424
-0,4439
-0,3456
-4,1656
6,3946
u14
-0,923
Fig. 4. The flowchart of the state variables for M33 model (years 1946-1975), where: y21 – electrical
power consumption, u1 – u14 – input variable
109
4. Conclusion
The elemental analysis of the Evans root line show that the occurrence two
kinds of the models , with the course of a single line elemental Evans (models: M1,
M15, M25, M30) and with the course of five elemental lines (models: M5, M10,
M20, M33).
The first type of model show that the EPS was a very stable system
(characteristics of runs from the pole located At the point - 0,1786 on the axis of
real numbers and along it tends to - ∞. Such a situation occurred in Poland in the
years: 1946-1979 (model M1), 1961-1990 (model M15), 1971-2000 (model M25)
and 1976-2005 (model M30).
However, in the case of the second type of model was obtained unstable
models, since four elemental lines are located in the right half-plane of a complex
variable s, and only one characteristic runs along the axis of real numbers in the left
half-plane from the pole located of the beginning of the coordinate system to - ∞.
Continue to investigate relationships between the received models of
development, including primarily changes in model state variables and state
variables flowcharts.
References
[1] Bolkowska B.: Planowanie systemu elektroenergetycznego w ujęciu teorii sterowania,

Archiwum Energetyki nr 4/1976.
[2] Cegielski M., Kremens Z., Soberajski M.: Systemy elektroenergetyczne – teraźniejszość
i przyszłość. Archiwum Energetyki nr. 1-2/1997.
[3] Brown M., Filho D. C., de Souza J. C. S..: Forecasting-Aided State Estimation – Part I:
Panorama, IEEE Transactions on Power Systems, Vol. 24, No. 4, Nov. 2009, pp. 1667-
1677.
[4] Brown M., Filho D. C., de Souza J. C. S., Freund R. S.: Forecasting-Aided State
Estimation – Part II: Implementation, IEEE Transactions on Power Systems, Vol. 24,
No. 4, Nov. 2009, pp. 1678-1685.
[5] Kaczorek T., Dzieliński A., Dąbrowski W., Łopatka R.: Podstawy teorii sterowania.
WNT. Warszawa 2005.
[6] Sobierajski M., Wikosz K.: Sieci elektroenergetyczne a rynki energii elektrycznej.
Problemy i Perspektywy. Prace Naukowe Instytutu Energoelektryki Politechniki
Wrocławskiej. OW PWr. Wroclaw 2000.
[7] Staniszewski R., Sterowanie procesem eksploatacji. WNT. Warszawa 1990.
[8] Statystyka Elektroenergetyki Polskiej. Wydawnictwo Zjednoczenia Energetyków,
Centrum Informatyki Energetyki. ARE S.A.. Warszawa 1952-2004.
[9] Tchórzewski J.: Badanie prawidłowości rozwoju systemu elektroenergetycznego.
Wybrane modele rozwoju. Computer Applications in Electrical Engineering. IEEE PP,
EEC PAN, IEEE Poland Section, PP. Poznań 2011.
[10] Tchórzewski J.: The Development Process of Electrical Power System as Unmanned
Factory. Electrical Engineering. Issue 59. Academic Journals 2009.
110
[11] Tchórzewski J.: Electric Power System from the Point of View of Model Development.
Part 1. Some Results for the Period of 1946-2007. Computer Applications in Electrical
Engineering. IEEE PP, EEC PAN, IEEE Poland Section, PP, Poznań 2010, pp. 111-112.
[12] Tchórzewski J.: Electric Power System from the Point of View of Model Development.
Part 2. Coding of Development System. Computer Applications in Electrical Engineering.
IEEP PP, EEC PAN, IEEE Poland Section, PP, Poznań 2009, pp.113-114.
[13] Wajs K.: Linie pierwiastkowe w automatyce. WNT. Warszawa 1973.
[14] Zajczyk R.: Modele matematyczne systemu elektroenergetycznego do badania
elektromechanicznych stanów nieustalonych i procesów regulacyjnych. Wyd. PG.
Gdańsk 2003.
Electrical energy consumption by the household - type consumers

Ryszard Frąckowiak, Tomasz Gałan
60 – 965 Poznań, Piotrowo 3A, e-mail: tomasz.galan@doctorate.put.poznan.pl
In the paper, selected results of analysis of the electrical energy consumption by the
household-type consumer group as well as results for energy defined regarding the standard
profiles for such a group are presented. In the analysis, a developed tailored IT tool has
been used
1. Introduction
Forecast of the average electricity consumption by the individual consumers

groups is an important topic regarding the electricity market. Good knowledge of
the factors shaping the load graphs supports the electricity demand forecasting.
In technical practice, there are some forecasting methods for different time scale
where the input parameter is the annual amount of electrical energy demand [1].
Regarding the measuring data on the previous year, the distribution companies’
operators develop so called average electricity load graphs - standard profiles – for
individual consumer groups showing some special features (voltage level, annual
electricity consumption, use of a common tariff, way of space heating, etc.) [2, 3, 4].
In the paper, the results of research work on the electrical energy consumption
in selected consumer group settled according to the G tariff are presented. The
research was based on the results of the individual consumers’ measurements and
on the electrical load graphs for profiled consumers developed by Polish
Association of the Electricity Transmission and Distribution (PTPiREE).
The analysis was aided by the developed purpose-oriented Excel file [5].
2. Selected analysis results
Characteristics of individual consumers under survey regarding consumer’s

location, electrical heating method applied and other features are shown in Table 1.
In Table 2, the characteristics of the standard profiles developed by PTPiREE for
household consumers are presented.
In Table 3, annual electrical energy consumption (Er) by consumers under
consideration in successive years of the electrical load recording is shown. Annual
energy consumption found for four successive years when the profile graphs were
valid are shown in Table 4.
112
R. Frąckowiak, T. Gałan / Electrical energy consumption by the household...
Table 1. Characteristics of surveyed consumers by special features
Consumer
Consumer features Average for the group
type
Number of
Tariff group
Level of Number of
Number of Household Persons in Group
M– Electrical Electrical implemen consumers
phases in the Space the code
unicipal water space -tation of in the group
household surface household
W – rural heating heating Electrical
power supply m2
appliances
G11
NO NO 1 Low 50-75 up to 5 40
M1
M
G11
YES NO 1 Average 50-75 up to 5 24
M2
G11
G11
NO NO 1 Average 75-100 5 to 10 25
W1
W
G11
YES NO 3 High 75-100 5 to 10 15
W2
G12
NO NO 1 Average 25-50 up to 5 42
M1
G12
G12 M YES NO 1 Average 50-75 up to 5 55
M2
G12
YES YES 3 High 75-100 5 to 10 49
M3
Table 2. Standard profile types (for households) developed by PTPiREE
Approx. Number of
Classification of profile of Electricity consumer Consumers classified into the
Profile Profile by survey year
name
Tariff Residential consumer meeting the
2007 2008 2009 2010
group requirements
Possesses a Single-tariff electrical

A Profile G11 ~700 ~700 ~600 n.d.
energy meter
Possesses a double-tariff electrical
B Profile G12 energy meter and no electrical space ~700 ~500 ~500 n.d.
heating .
C Profile G12 energy meter and electrical space ~80 ~60 ~60 n.d.
heating other than dynamic

D Profile G12 energy meter and dynamic electrical ~10 ~10 ~10 n.d.
space heating
n.d. – no data
113
Table 3. Annual electrical energy consumption in surveyed consumer groups
Er
kWh
Group Code
2003 2004 2005 2006
G11 M1 2536 2615 2581 2603
G11 W1 2715 2684 2785 2740
G11 M2 5376 5452 4959 4789
G11 W2 6390 6257 6409 6153
G12 M1 5259 4986 5015 4862
G12 M2 6841 6590 6677 6421
G12 M3 10503 11843 10282 10503
Table 4. Annual electrical energy consumption per individual profile groups
Er
Profile Name kWh
2007 2008 2009 2010
A Profile 3904 5075 3449 1969
B Profile 4724 5449 4498 3941
C Profile 7835 6708 5807 4596
Total average annual electrical energy consumption for all of surveyed

consumer groups settled according to the G11 tariff oscillates between 3672kWh
and 3838 kWh, and average is 3769 kWh per 1 consumer in a year. However,
electrical energy consumption for individual consumer groups differs significantly.
Energy consumption by the consumers living in rural areas is a bit greater. Also,
values of consumed energy determined according to the profiles developed for
particular years are different. For instance, average annual consumption of
electrical energy found referring to the A profile for 2008 is much higher than the
values reported above whilst, for A profile 2010, the value is much lower.
114
0,8
A profile 2007
0,7 A profile 2008
A profile 2009
0,6 A profile 2010
0,5
Pśr h / consumer
0,4
0,3
0,2
0,1
0,0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour of day, h
Fig. 1. Average annual load variation for standard A profile
In the shape of the load graph for a standard A profile, any explicit trend in
variation is not found. Shape and values on the graph depend on contribution of the
particular consumer group taken into account when developing the particular
profile.
20
18
16
Power consumption
14
12
10
4 Profile A 2010
2 Profile A 2008
0
14 15 16 17 18 19 20 21 22
Sunset hour
Fig. 2. Average annual power consumption variation versus sunset hour for A Profile consumer
in 2008 and 2010
115
In Fig. 2, relationship between the average annual energy consumption and the
monthly averaged sunset hour for two selected years when the A profile was in
force are presented for 2008 and for 2010.
Regarding strength of relationship between magnitudes reported in Fig.2 for
2010, it can be defined using linear function in the form:
E śrm = 12,27 - 0,367 ⋅ Tzach śrm (1)
where: R 2 ≈ 0,98 .
It indicates that, in the A profile for 2010, there is a high (deciding on the graph
form) contribution of consumers with relatively high load resulting from lighting,
i.e. high contribution of consumers belonging to the G11M1 and G11W1 groups
who do not use electricity for heating. Such a conclusion is supported by small
power consumption as compared to the consumption in other years for which the A
profile has been developed (Fig. 1). For other years in A profile, the relationship
under consideration is weaker. Especially, it is less evident in 2008 (Fig. 2) where
the match coefficient is the lowest and is R2 ≈ 0,51.
These conclusions are confirmed by the test using the correlation coefficient
Persona (rxy) – (Fig. 3) concerning relationship between the energy consumption
and sunset hour and temperature.
High values of rxy coefficient in particular hours (late afternoon, early morning)
indicate that electrical power is used for lighting purposes – A Profile 2010 (Fig.
3b). For consumers using electrical energy for heating the space, rxy becomes high
all day long (Fig. 3a).
Graphs for daily variation of correlation coefficient for consumed power,
temperature and sunset hour shown in Fig. 3 runs like each other. It results from
strong dependence between the sunset hour and air temperature.
The way of the space heating is a relevant parameter affecting the values of
electricity consumed by the household [4].
In Fig. 4, the relationship between the monthly averaged energy consumption
and the monthly averaged air temperature ( ϑ śrm) by the consumers of the G12 M3
group using electricity to space heating and C profile are presented.
Strong dependence presented in Fig.4 can be expressed with linear function in
the form:
- for G12 M3: Eśrm = 45,05 − 1,75 ⋅ ϑśrm (2)
- for C Profile 2008: Eśrm = 29,4 − 1,17 ⋅ ϑśrm (3)
In both consumer groups shown in Fig.4, the determination coefficient is very
high (R2 > 0,95).
For economical reasons, significant part of electrical energy consumption by the
G12 group consumers falls into the time period with lower energy price.
116
In Fig. 5, variation in value of correlation coefficient between the power

consumption in day hours and the daily averaged temperature for the C profile
consumer from 2008 is presented for two air temperature ranges.
Hour of the day
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
0,0
-0,1 Sunset
Correlation coefficient value
Temperature
-0,2
-0,3
-0,4
-0,5
-0,6
-0,7
-0,8
-0,9
-1,0
Hour of the day

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
0,0
Sunset
-0,1
Temperature
-0,2
-0,3
-0,4
-0,5
-0,6
-0,7
-0,8
-0,9
Fig. 3. Values of Correlation coefficient between energy consumption, temperature and sunset
hour for A Profile in (a) 2008, (b) 2010
117
60
50 C profile 2008
G12 M3
Energy consumption
40
30
20
10
0
-5 0 5 10 15 20 25
Temperature
Fig. 4. Monthly averaged energy consumption versus monthly averaged air temperature for G12
M3 and C Profile 2008
Hour of the day

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
0
-0,1
-0,2
Temperature >10°C
-0,3
Temperature <10°C
-0,4
-0,5
-0,6
-0,7
-0,8
-0,9
-1
Fig. 5. Correlation coefficient between consumed power and air temperature for C profile
consumer
High nominal power consumption of heating appliances as well as the period of

their operation considerably affect the power consumption. It can be especially
observed when the temperature falls below 10°C.
118
3. Final remarks
Specific features of an individual consumer as well as those of the consumer

group considerably affect the shape and values of the electrical load graphs. The
factors that, among others, have evident influence on the value of the consumed
power value are the sunset hour and the air temperature. Any of the factors affects
this value in different way. In the G11 M1 and G11 W1 group, the consumed
energy value is mostly influenced by the sunset hour, whilst in the G12 M3 group –
by the temperature. It is also reflected in the developed standard profiles.
The power consumption values depend strictly on the tariff type, localization
and the way the electricity is used for heating purposes. For the G12 group living in
rural areas, the electrical energy consumed by the consumers using the electrical
heating of space and water (G12 M3) is approximately from 2 to 2,5 times greater
than that consumed by the consumers using no electrical heating (G12 M1).
The A profile that includes different groups of consumers shows special
features. Therefore, proper classification of consumers for the profile development
purposes is crucial.
References
[1] K. Majka, Bilansowanie zapotrzebowania godzinnego energii elektrycznej przez

odbiorców z wykorzystaniem profili obciążeń, Energetyka, Czerwiec 2006, s. 410-415.
[2] Internet: www.operator.enea.pl – Instrukcja Ruchu i Eksploatacji Sieci Dystrybucyjnej
(IRiESD).
[3] R. Frąckowiak, T. Gałan, Wpływ wybranych czynników charakteryzujących odbiorcę
i warunki zewnętrzne na krzywe obciążenia odbiorców grupy G11, Materiały
konferencyjne – Aktualne Problemy w Elektroenergetyce 2007, Jurata 2007.
[4] R. Frąckowiak, T. Gałan, Krzywe obciążenia elektrycznego odbiorców typu
gospodarstwa domowe rozliczane według taryfy G12, Przegląd Elektrotechniczny,
Czerwiec 2009.
[5] Frąckowiak R., Gałan T.: Komputerowe wspomaganie analizy przebiegów zmienności
obciążeń odbiorców zasilanych z sieci niskiego napięcia, Materiały XII Konferencji
Naukowej – Zastosowanie Komputerów w Elektrotechnice, ZKwE’08, Poznań 2008,
s. 95-96.
119
Numerical modeling of dynamic heating processes

Marcin Wesołowski, Ryszard Niedbała, Daniel Kucharski
Warsaw University of Technology
00-661 Warszawa, ul. Koszykowa 75, e-ail: Marcin.Wesolowski@ien.pw.edu.pl
Jacek Hauser
61-965 Poznań, ul. Piotrowo 3a, e-mail: Jacek.Hauser@put.poznan.pl
The article deals with the numerical modeling and simulation of dynamic heating
processes. The mathematical models, based on classical Fourier’s law and Vernotte
equation were described. Both models can be used for analysis of the electro – thermal
processes. The numerical model of exemplary heating process was created and
implemented in the Mathcad environment. The calculation results of two different models
of heat transfer were compared. The modeling accuracy and the range of applicability of the
described models were determined.
1. Introduction
The specific quantities of electrical heating devices enables, in many cases, for
achieving of high volumetric power values within the external surfaces. This fact
enables for realization of huge number of electro-thermal technologies
characterized by high speed of heating, as volumetric as surface. From a range of
such applications, the following issues can be listed:
- analysis and synthesis of phenomenon during shock wave propagation;
- nanotechnology;
- the surface processes of high - energetic laser impulses;
- the medical applications.
All applications above require usage of advanced methematical describtion of
heat transfer phenomenon within heated elements. Unlike a typical description of
heat transfer, based on the Fourier equation, such problems require the Cattaneo –
Vernotte approach. In most calculating systems for modeling and simulation, the
Cattaneo – Vernotte equations are not implemented. The fact can result a very large
simulation errors, especially in modern electrothermal techniques. In the article the
mathematical and numerical model for the simulation of dynamic heating processes
was described. The model and the classical approach were used for multi-variant
calculations. The comparison between calculation results in both cases allow
evaluation of accuracy and determination the limits of utility of both heat transfer
models.
120
M. Wesołowski, R. Niedbała, D. Kucharski, J. Hauser / Numerical modeling of …
2. Heat transfer in solids
The basic problems connected with the electrical heating enable application of
classical differential equations of heat transfer. The Fouriers law constitute that
heat flux density is proportional to the temperature gradient in every point of space
and every time step (1) [3; 4]. The classical equation of transient heat transfer can
be presented as (2) for space without heat sources [6]. The parabolic equation
contains some simplifications resulting from establishment of infinite speed of heat
waves propagation. Such models can be used in most techniques of electro –
thermal energy conversions. But in general, the assumption of finite heat
propagation speed is true [1; 5; 6]. Basing on this assumption, Cattaneo and
Vernotte formulated a modified transient heat transfer equation. The equation for
the case of 1D heat transfer can be written as (3).
q ( x, τ ) = − λ ∇t ( x , τ ) (1)
where: q- heat flux density, λ - thermal conductivity, t- temperature, τ - time.
∂t
= a∇ 2 t (2)
∂τ
where: a- thermal diffusivity defined as a ratio of thermal conductivity and a
(
product of density and the specific heat a = λ / c p ⋅ ρ .( ))
∂t( x ,τ )
q( x , t + τ q ) = −λ (3)
∂x
Where: x- spatial coordinate, τq- the relaxation time resulting from the velocity
of heat propagation.
τq = a / C2 (4)
After the development of equation (3) into the Taylor’s series for heat flux
density, it was obtained:
∂q ( x, τ ) ∂T ( x, τ )
q ( x, t ) + τ q = −λ (5)
∂τ ∂x
In general, the equation of heat transfer in micro scale in layers of heated body,
requires the usage of the model taking into account delay of temperature gradients
from heat fluxes [3; 4; 6]. In the cases where the normalization time can be
skipped, the equation (5) can be written as classical Cattaneo – Vernotte equation
(6) [6].
⎡ ∂t ( x ,τ ) ∂ 2t ( x ,τ ) ⎤ ∂p ( x ,τ )
cp ⎢ +τq 2
⎥ = λ∇ 2t (x ,τ ) + pV ( x ,τ ) + τ q V (6)
⎢⎣ ∂τ ∂τ ⎥⎦ ∂τ
where: pV - volumetric power density.
In regions without heat sources, the equation (7) is valid [6].
121
⎡ ∂T ( x, τ ) ∂ 2T ( x , τ ) ⎤
cp ⎢ +τq [
⎥ = ∇ λ ∇T ( x, τ ) ] (7)
⎢⎣ ∂τ ∂τ 2 ⎥⎦
The equations above were used to modeling of heating processes to compare the
simulation results with classical model of heat transfer phenomenon. During the
calculations, the axially symmetric models were used. So that, in the axis of
symmetry the zero normal heat flux was assumed. The equations were
supplemented with required third type boundary conditions in external surfaces.
During the calculations of dynamic heating processes, the commercial
calculating systems were not used. In most popular programs there is a less of
models and calculating procedures enables to account the effects of delay of
temperature gradients from heat fluxes [6]. Therefore, the authors algorithm was
created during the calculations. The numerical algorithm was based on finite
difference method and was implemented in the Mathcad system. Two meshes were
used for geometrical coordinates ( Δx ) and the time steps ( Δτ ) . The equation of
heat transfer, after derivatives approximation with differential quotients can be
written as (8).
∂ ⎛ ∂T ⎞ Ti +1,τ − Ti ,τ T − Ti ,τ
⎜λ ⎟ = ψ i +1 + i −1,τ ψ i −1 (8)
∂x ⎝ ∂x ⎠ i ,τ Wi +1,τ −1 Wi −1,τ −1
where: for regular mesh: ψ i +1 = ψ i −1 = 1 / Δx . The denominator of equation above
contains the thermal resistances.
3. Calculations
Calculation work was begin on the analysis of general induction heating cases,
also typical for the most of the classic cases. The case of cylindrical geometry
surface heated charge measuring 30 mm in outer diameter and 300 mm high have
been tested. The realization of surface heating consisted in forcing the operating
frequency of the current in the inductor at the level of 150 kHz. With this value the
skin depth was about 2 mm (9). The entire power secreted in heated material was
about 3 kW. Results of the heating process simulation with the usage of the basic
(1) and hyperbolic (3) heat conducting equation were achieved for a 20 second
heating time. Relative differences between results received with the usage of
various heat propagation were determined to illustrate the differences in results.
There is data obtained shown in the form of relative temperature differences vs.
time curves in the figure 1.
2⋅ρ
δg = (9)
ω⋅μ
where: ρ - resistivity, ω - pulse, μ - magnetic permeability.
122
Fig. 1. Describtions of the relative temperature differences between the results achieved on the basis
of two different numerical models. Δ1 – Δ5 – differences between results for corresponding
temperatures
Results presented help to pronounce relatively small differences obtained with

the usage of the heat propagation models described. Therefore, it is possible to find
the lack of functionality of utility of hyperbolic heat transfer equation in classic
applications, with relatively long heat time.
Subsequent calculations were designed to demonstrate the impact of heating
power values on the dynamic characteristics. Moreover, the designation of
deviations between results obtained on the basis of the analyzed equations of
heatpropagation was performed. Calculations were made for the case of surface
heating of the non-magnetic steel charge. The charge in the form of cylinder had a
diameter of 3 and height of 30 mm. Depending on the heating power dissipated by
the surface of charge, different heating characteristics and temperature values were
obtained. The values of heating power were changed in the range of 100- 200000
W, which resulted in charge analyzed heating up to temperature limits from 200 to
2000°C within 0.01-2s.
In the Figure 2 characteristics of the time-temperature coordinates relative to the
three cases of power amounting to 200, 2000 and 20000 Watts is presented. The
characteristics were prepared as dependences of relative temperatures, defined as
percentage deviations from maximal values , from the relative heating times. It was
observed that with increasing power values, the heating times were reduced and the
maximal temperatures have reached a higher values.
123
Fig. 2. Relative temperatures characteristics during the heating time for different power values.
tf- temperatures from the Fourier equation, tv- temperatures from Vernotte equation. 1- the power
value of 200 W; 2- the power value of 2000 W; 3- the power value of 20 000 W
Basing on the calculation results, it was shown that deviations characteristics,

resulting from differences of algorithms used is quite similar for every power
values. Maximal percentage differences between results were observed in the time
of 25% of analysed range. The deviations, despite similar percentage values, are
quite different in respect to the absolute values. With increasing heating power
values, the maximal divergences between results reached high values. The
characteristics were shown in Fig. 3.
Fig. 3. Characteristics of absolute temperature divergences between calculating results from two
models of heat transfer. 1- the power value of 200 W; 2- the power value of 2000 W; 3- the power
value of 20 000 W
124
It was demonstrated that in cases of relative small heating power values and
process times larger than 10 milliseconds, the differences between results from
classical (1) and Vernotte (3) equations has reached the values of 10 K. Maximal
divergences has been reached for the case of the largest analyzed heating power
(the power density was 95x106 W/m3). In this case the calculating results
divergences were on the level of 90 K. In heating systems characterized by small
dimensions and short heating times, the utility of close to reality Vernotte equation
is very necessary. The dynamic of processes requires a great precision by dosing a
heating power.
Increasingly, the processes realized in induction heating systems are used in
semiconductor industry and wide – known nanotechnology industry. As
manufacturing of modern semiconducting materials and elements as production of
nanostructures requires a high accuracy of operating characteristics of heating
devices. In this example, the heat transfer phenomenon during heating of the most
popular semiconducting materials was discussed. The basic subject of calculations
was examination of temperature characteristics, obtained from classical Fourier (1)
and Vernotte (3) heat transfer equation. The model of cylindrical workpiece was
used. The power value was determined to guarantee warming of the workpiece to
the temperature of 1900°C in the time of 400 ms. The heating characteristics of Si,
Ge, GaAs and GaP were shown in figure 4. Additionally, in this figure, the results
from Fourier’s model ( the same for all analyzed materials) were shown.
Fig. 4. Temperatures characteristics in the centre of the workpieces geometry for different
semiconducting materials. 1- results from Fourier equation; 2- results for Si; 3- results for Ge; 4-
results for GaAs; 5- results for GaP
125
Taking into consideration the relaxation time, only GaP was heated quite
differently from classical model of heat transfer. The characteristics of deviations
from classical approach for GaP were shown in Fig. 5.
Fig. 5. Characteristics of temperature deviations of the hyperbolic heat transfer equation from
the Fourier equation for GaP. Δ1 – Δ4 – deviations of subsequent layers deep into the workpiece
Maximal divergences between results for two models exceeds in this case the values
of 130 K. This fact indicates to the inadequacy of classical model to the practical issues.
However, for other semiconducting materials (Si, Ge, GaAs), the calculating errors
resulting from rejection of relaxation times did not exceed the value of 5 K.
In many modern technologies, giant energetic impulses with short remaining
times are being used. The general case of heating propagation in surface heated
material was tested by interacting with a power impulse with remaining time 100
μs. Because of the occurrence of the impulse with nominal power 100 000 W, outer
layers of the material were heated up to 900°C. Results of the simulation for the
case of Fourier and Vernotte equation utility were shown in the Figure 6.
Fig. 6. Temperature curve in time function achieved as a result of simulation of the system reaction
on a power impulse with remaining time 100 μs. A- results for Fourier equation; B- results for
Vernotte equation in the first node. 1-5 – temperature curves in next nodes investigating the charge
material
126
Both temperatures in successive layers of heated body and dynamic of heating

and cooling process were quite different. Utility of classical heat transfer equation
has led to obtain a higher temperature values with shorter latency times in relation
to the Vernotte model. The temperature divergences between the models in time
domain were shown in Figure 7.
Fig. 7. Characteristics of temperature deviations between parabolic and hyperbolic heat equations as
the functions of time. 1-5 - curves in next nodes investigating the charge material (as in the Fig. 6)
4. Conclusions
The subject of this paper consisted of the analysis of widely understood fast heat
modeling accuracy with the usage of two different heat propagation models. Basic
factors affecting the divergence between results achieved were tested. Apart from
parameters and charge geometry, heat power value has a fundamental impact on
the heating speed. Calculations, carried out while accomplishing the project,
proved the necessity of utilizing the Vernotte equation in issues concerning short
heating processes with heating time not exceeding 1 second. A numeric procedure
allowing precise analysis of this kind of systems, used particularly for highly
specialized heating devices calculations was drew out.
References
[1] Auriault J.-L., Lewandowska J., Royer P., On non-Fickian Hyperbolic Diffusion,
Studia Geotechnica et Mechanica, Vol. XXX, No 1-2, 2008.
[2] Galunin S., Zlobina M., Blinov Yu., Nikanorov A., Zedler T., Nacke B., Electrothermal
Modeling and Numerical Optimization of Induction System for Disk Heating,
International Scientific Colloquium Modeling for Material Processing, Riga, 2006.
[3] Hauser J., Elektrotechnika. Podstawy Elektrotermii i Techniki Świetlnej, Wydawnictwo
PP, Poznań, 2006.
[4] Koval’kov V. P., Heat Conduction Equations Allowing for Finite Phonon Velocity,
Journal of Engineering Physics and Thermophysics, vol. 70, No 1, 1997.
127
[5] Majchrzak E., Mochnacki B., Szopa R., Numerical models of thermal processes in
domain of metal superficial layer subjected to an external heat flux. ECCM 2010.
[6] Tavakoli M.H., Karbaschi H., Samavat F., Computational Modeling of Induction
Heating Process, Process, in Electromagnetics Research Letters, vol. 11, 2009.
An analysis and verification of a dynamical model

of a transverse flux motor
Marian Łukaniszyn, Marcin Kowol, Janusz Kołodziej
Opole University of Technology
45-272 Opole, ul. Sosnkowskiego 31, e-mail: m.lukaniszyn; m.kowol;
ja.kolodziej@po.opole.pl
The paper presents dynamics simulation results for the three-module reluctance motor
(Transverse Flux Motor -TFM) with an outer rotor. Calculations of the most important integral
parameter used in the circuit model, are performed using the Flux3D package based on the finite
element method. A mathematical model for a dynamics analysis of TFM is built in the
Matlab/Simulink environment. The model has a hierarchical structure with its fundamental part
being a single phase-belt of the motor. The basis for the construction of the laboratory setup is a
DS1104 R&D Controller Board with a CP1104 Connector Panel. The experimental validation is
divided into two parts. In the first step, the mathematical model of TFM under steady-state
operation is examined. In the second step, experimental validation of the results obtained from
the dynamical circuit models is performed. Results of the simulations, confirmed by the
experimental validation presented in this work, allow to determine the most important
characteristics of TFM in both dynamic and steady-state conditions.
1. Introduction
Transverse flux motors (TFMs) have recently attracted remarkable interest both
from the academia and various industrial environments [1, 3, 5, 6, 9, 10, 11]. The
low-speed motor is characterized by a high ratio of the electromagnetic torque to
its volume [2, 4, 11], leading immediately to various high-torque transmission-free
applications, to mention electric wind generators [1], electric (and hybrid) drives
[11] and in-wheel drives [3, 9].
Our previous papers on TFM’s have been devoted to the problem of reduction
of accompanying torque pulsations, which have been plaguing not only TFM’s [2,
4, 5, 7]. Effective construction optimization tools have been offered in order to
reduce the torque pulsations while maintaining a high level of the average
electromagnetic torque [5, 7].
In this paper, we extend our TFM modeling and simulation interest to include
dynamical states of the motor, which is justified mainly by the start-up operating
conditions, the switches in the supply bridge and possible changes in the load.
2. Prototype of TFM
The main object of the research is switched reluctance motor with axial flux
(TFM). The prototype motor structure is shown schematically in Fig. 1. The
129
M. Łukaniszyn, M. Kowol, J. Kołodziej / An analysis and verification of a dynamical…
considered motor consists of three modules in which the rotors are shifted by 10
mechanical degrees with respect to each other. The stator modules are placed
symmetrically on an acid-resistant steel shaft. Each module has twelve teeth and
includes one phase belt that forms a solenoid. The rotor teeth are made of solid
iron. The modules are insulated by nonmagnetic inserts. The outer layer and rotor
layer are made of a nonmagnetic material (aluminium). The main specifications for
the motor are given in Table 1.
Table 1. Specifications for TFM

Supply voltage Un=24V
Rated current In=12A
Rotational speed 0÷300 obr/min
Winding Three-phase
Number of turns 130
External diameter of rotor 158 mm
External diameter of stator 103.5 mm
Air gap δ=0.5 mm
Fig. 1. TFM structure
A power converter (Fig. 2) is required to activate and commutate the SRM

phases, and the classic asymmetric half-bridge inverter is in general used, requiring
two switching devices and two power diodes per phase [4, 5]. There are two
control procedures possible – Hard-Chopping and Soft-Chopping. The simplest
motor control consists in sequentially switching on and off for the phase current
(phases A, B, C, A etc.).
130
Fig. 2. The half-bridge (“H”) TFM power converter
3. Circuit and field-circuit models
The modeling process for SRMs with axial flux is different from the techniques
used in conventional cylindrical motors. The main reason for that is high
nonlinearity of the reluctance motor. There are many interesting solutions for this
type of problems in the bibliography [5, 10, 11, 12].
The dynamics of the motor can be described by two below equations presented.
The first equation describes the electromagnetic phenomena
dψ n (t )
u n (t ) = R n i n (t ) + (1)
dt
where u n (t ) is the n-th phase supply voltage, R is the phase winding resistance,
in(t) is the n-th phase current and ψ n (t ) is the flux in the n-th phase winding.
The second equation involves the nonlinear mechanical phenomena
dω (t )
Te (t ) = Tl (t ) + kω ω (t ) + J
dt
(2)
3
Te = ∑ Te n (in , Θ )
n =1
where Ten(t) is the n-th phase electromagnetic torque, Tl(t) is the load torque, Te(t) is
the total electromagnetic torque, k ω is the friction damping coefficient, J is the
dΘ(t )
moment of inertia of the rotor , ω (t ) = is the angular rotor speed and Θ(t )
dt
is the rotor position.
131
Solution to these equations is easy under the linear model assumption for (2), at
the cost, however, of considerable errors. The second way of solving these
equations is to take into account the nonlinear flux and torque characters. The
magnetostatic calculations are one of the methods for including the nonlinearity.
The nonlinear flux and torque characteristics ψ = f1 (i, Θ) and T = f 2 (i, Θ) ,
respectively, depending on current and rotor position, have been obtained from the MES
field calculations verified by measurements on the prototype motor ( see Fig. 3) [5].
The above mentioned software enables to solve the TFM transient-state
problem after coupling an external power circuit with a control system. The
nonlinear mathematical model (1) and (2) coupled with magnetostatic FEM is
highly competitive with a possible use of the complicated, time-consuming FEM
field-circuit model.
Fig. 3. Single phase subsystem
Fig. 4. The structure of a simulation system in Matlab/Simulink
The structure of a simulation system in Matlab/Simulink, presented in Fig. 4, is

hierarchical. It consists of several units, the most important of which is a model of
a single phase. All the belts are supplied from a power electronic supply system
implemented in the Plecs environment (classical “H” half-bridge). The main
supervisory system enables to introduce mechanical and electrical parameters
related to the nature of simulations run, as well as output data acquisition.
132
4. Test bench
A laboratory stand for the TFM simulation test bench includes the DS1104 fast-
prototyping card of dSPACE, cooperating with the PCI bus, the CP1104
input/output panel and auxiliary units like RTI and ControlDesk modules. The first
module enables to create applications in Simulink from the graphics level while the
second one provides on-line visualization of variables of interest. An easy access to
the tools enabling to modify selected parameters during the simulation run is of
importance as it is not necessary to change and re-generate a source code for the
DSP processor.
A view of the laboratory stand and a block diagram of the simulation test bench
for the TFM motor under test are shown in Figs. 5 and 6, respectively.
Fig. 5. The laboratory set of TFM
The test TFM is supplied from batteries. The torque is measured with the E300-
RWT1-02 tensometric torque sensor. On-line measurements of angular position of
the rotor with respect to the stator are performed by means of a rotational-pulse
transducer with the resolution of 1000 pulse/rev. The TFM is loaded with an
133
electric dynamometer. For supervisory control, the above mentioned DS1104 card
is applied, together with the CP1104 panel. The card also enables to measure both
analog and digital signals. However, due to the lack of a separation system,
measurements of currents and voltages in particular belts of the motor are
performed by means of a 12-channel power analyzer.
Fig. 6. The Block diagram of TFM test bench
5. Calculations and experimental validation
A due "closeness" of mechanical characteristics for the test TFM and its
simulation model is selected as a criterion for the correctness of the model. A
comparison of mechanical characteristics n=f(Tl), where n is the rotational speed,
for various switch-on (αon) and switch-off (αoff) angles shows a good accuracy of
the considered circuit model (Fig. 7).
In the next part of this work, TFM simulation and experimental verification
results will be presented in case of the motor start-up under load. To this end, Fig.
8 presents plots of currents and voltages drop in each particular phase belt in case
of 1) measured ones in the test TFM (with the NORMA 6200 power analyzer), 2)
obtained from the circuit model (Matlab/Simulink) and 3) obtained from the field-
circuit model (Flux3d). Comparison of source current and electromagnetic torque
for the measurements, circuit and field-circuit models at the supply voltage
Usup=12V, load torque Tl=0,7 N·m, αon=18°, αoff=28°are shown additionally in
Fig. 9.
134
a)
350
Meas ured
300 Sim ulated
250
200
n [rpm]
150
100
50
0
0 0.2 0.4 0.6 0.8 1
T l [N⋅ m]
b)
200
Meas ured
180
Sim ulated
160
140
120
n [rpm]
100
80
60
40
20
0
0 0.2 0.4 0.6 0.8 1
T l [N⋅ m]
Fig. 7. Mechanical characteristics: a) αon=17°, αoff=27°, Usup=24V b) αon=17°, αoff=27°, Usup=12V
Quite slight discrepancies of some 5% to 15% between the measurements and

simulations for the finally accepted circuit model result mainly from the
simplifying assumptions made. In fact, magnetic couplings between the adjacent
modules have been omitted in the model. Also, the influence of temperature
changes on resistance of windings for each particular phase belt as well as possible
induction of eddy currents in solid parts of the rotor and stator have not been
accounted for.
135
a) b)
Fig. 8. Comparison of phase currents a) and voltage drop b) for measurements, circuit and field-
circuit models, at the supply voltage Usup=12V, load torque Tl=0,7 N·m, αon=18°, αoff=28°
a) b)
Fig. 9. Comparison of source current a) and electromagnetic torque b) for the measurements, circuit
and field-circuit models at the supply voltage Usup=12V, load torque Tl=0,7 N·m, αon=18°, αoff=28°
136
Fig. 10 presents plots of phase belt currents under step change in load. After a
no-load start-up, at time t1=2s the motor is step-wise loaded with the moment Tl=1
N·m. At time t2=3s, the load is coming back to Tl=0 N·m. Rotational speed at the
step change of load torque for various supply voltages and αon=17˚, αoff=27˚ are
shown additionally in Fig. 11.
Fig. 10. Currents of phases A, B, C for a step change of load torque at supply voltage
Usup= 12V and αon=17˚, αoff=27˚
Fig. 11. Rotational speed at the step change of load torque for various supply voltages
and αon=17˚, αoff=27˚
137
6. Conclusions
The developed nonlinear, dynamic circuit model of transverse flux motor has
been shown in simulations and experimental verification to have a good accuracy.
The model enables a fast analysis of the motor operation and it is competitive with
respect to a field-circuit model. In addition to the construction optimization
methodology for the TFM offered earlier [6, 7], the nonlinear dynamic model
presented here, supports analysis and design tools for the TFM.
Further research works in the field will be focused on an analysis of influence of
control algorithms on the electromechanical parameters of TFM as well as control
optimization in order to reduce torque pulsations and to increase the efficiency of TFM.
References
[1] Arshad W. M., Thelin P., Bäckström T., Sadarangani C.: Use of Transverse-Flux
Machines in a Free-Piston Generator, IEEE Transactions on Industry Applications,
July/August 2004, vol. 40, No. 4, pp. 1092÷1100.
[2] Babazadeh A., Parspour N., and Hanifi A.: Transverse flux machine for direct drive
robots: Modelling and analysis, IEEE Conference on Robotics, Automation and
Mechatronics, Singapore, vol. 1, 2004, pp. 376- 380.
[3] Goryca Z., Łukaniszyn M., Jagieła M., Wróbel R.: Analiza momentu
elektromagnetycznego w silniku reluktancyjnym, XXXIX SME, Cedzyna/Kielce
2002, pp. 333÷339.
[4] Kastinger G.: Design of a novel transverse flux machine, ICEM 2002, Brugge,
August 2002.
[5] Kołodziej J.: Analiza dynamicznych i ustalonych stanów pracy silnika
reluktancyjnego ze strumieniem poprzecznym, Praca Doktorska, Opole, 2011.
[6] Kowol M., Łukaniszyn M., Latawiec K.: Modeling and construction optimization of
a modular TFM with an outer rotor. Electrical Engg., Vol. 92, No 3, September
2010, pp.111-118.
[7] Kowol M.: Analiza pracy przełączalnego silnika reluktancyjnego z wirnikiem
zewnętrznym do napędu lekkich pojazdów, Rozprawa doktorska, Opole, 2007.
[8] Łukaniszyn M., Kowol M.: Analiza pracy modułowego silnika reluktancyjnego
z wirnikiem zewnętrznym, Śląskie Wiadomości Elektryczne, 4’2005, s. 4-7.
[9] Łukaniszyn M., Kowol M.: Wpływ zmian konstrukcyjnych na parametry
elektromechaniczne silnika reluktancyjnego z wirnikiem zewnętrznym, Przegląd
Elektrotechniczny, Vol. LXXXII, No. 11, Sigma-NOT, 2006, pp. 43-45.
[10] Peethamparam A.: Design of Transverse Flux Machines using Analytical Calculations
& Finite Element Analysis, March 2001 PhD Thesis, Royal Institute of Technology,
Stockholm.
[11] Ritchie E., Tutelea L.: An overview of electric vehicle in-wheel drive systems,
XXXIX Międzynarodowe Sympozjum Maszyn Elektrycznych, 9-11 June 2003,
Gdańsk-Jurata, pp. 1-21.
[12] Tomczewski K., Wróbel K.: Jednoczesna optymalizacja kształtu obwodu
magnetycznego i parametrów zasilania przełączalnego silnika reluktancyjnego, Przegląd
Elektrotechniczny, 03/2009, s. 107-110.
138
Determining the optimal working point

and the loading characteristic of synchronous machine
for fan and pump drive system
Lech Nowak, Krzysztof Kowalski, Dorota Stachowiak, Jacek Mikołajewicz,
Wojciech Pietrowski, Kazimierz Radziuk
60 - 965 Poznań, ul. Piotrowo 3A, e-mail: lech.nowak@put.poznan.pl
In the paper the computer software for determining the optimal working point of electric
motors for fan (and pump) drive systems is presented. The main use of this software is the
preliminary designing of permanent magnet synchronous motor for such type of drives. The
second task of the software is determining the load characteristic of the fan system (pump
system), i.e. relationship between loading torque and angular speed. The obtained
characteristics can be used in the designing of control system for fan or pump drives.
1. Introduction
Electric drives for ventilation and drainage systems used in mining industry
represent significant part of all electric drives. Thousands of electric motors with
power ranging from just a few kilowatts to several megawatts are used in the
ventilation systems. The second most power consuming systems are the drainage
systems in which mainly the centrifugal pumps are employed pumps. Overall
power usage of these systems in the mining industry in Poland can be counted in
gigawatts. Proper design of the drives and the control systems is crucial to
minimize energy consumption in this sector.
Asynchronous drives are usually used in these two systems, with the exception
of the most powerful fan drive applications, where synchronous motors are
employed. It can be observed, however, that the nowadays the tendency to use
synchronous motors for the less power demanding applications is becoming more
popular. Permanent magnet synchronous motors using high-energy magnets made
with neodymium-iron-boron (ND-Fe-B) alloy, are particularly effective in these
applications. The constant angular speed ensures constant air or liquid flow rate in
any given period of time. Such drives have also higher power coefficient and have
grater energy efficiency due to lack of power loss in the rotor.
Use of the variable-speed drives is the most efficient way to control the air or
liquid flow. Control systems for synchronous motors are much simpler than those
used to control asynchronous drives.
Prior to design of efficient synchronous motor and its control system it is
necessary to determine load characteristics of fan or pump, i.e. the relation between
its angular speed.
139
L. Nowak, K. Kowalski, D. Stachowiak …/ Determining the optimal working point…
2. Determination of the working point of the ventilation system
The rated flow rate QN [m3/s] at rated speed nN is the basic parameter
characterizing the fan or pump.
Value of the fan flow rate demand and characteristic of the ventilated system Q
= Q(Δp), where Δp is the differential pressure, must to be given prior to the fan
system design process. The aerodynamic resistance of the ventilated objects
strongly depends on the rate flow Q, therefore the characteristic Q = Q(Δp) is
strongly nonlinear. This characteristic is usually presented in an inverted
coordinate system, i.e.: Δp = f (Q ) – Fig 1.
The characteristic of the turbo-machine (fan or pump) is also given in the same
coordinate system. The pressure difference between the two sides of the fan
decreases as a result of decreasing aerodynamic resistance of the ventilated system,
i.e. with the flow Q increase. Fan characteristics Δp = f (Q) is decreasing function –
Fig. 1.
The point of intersection of these two characteristics (point KN in Fig. 1)
determines the working conditions of the system fan-ventilated object, i.e. system
source-load.
Fan power output is equal to the product of differential pressure and flow rate
values: Pw = Δp ⋅ Q [5]. If Q = 0 or Δp = 0 then the power Pw = 0 . This means,
(
that there is an optimal working point K opt Qopt , Δp opt ) with the maximum fan
output power – Fig. 2.
Fig. 1. Determination of the optimal working point of the system: fan-ventilated object
140
The motor rated power can be expressed as follows:

PwN Δp N ⋅ Q N
PN = = (1)
η wN η wN
and its rated torque:
PwN Q ⋅ Δp N
TN = = N (2)
ω N ⋅ η wN ω N ⋅ η wN
where ω N , η wN are the rated values of angular speed and efficiency, respectively.
The rated working point of selected fan should correspond to the maximum output
power.
Rated power and rated torque of the driving electric motor can be determined on
the basis of the relations (1) and (2).
Fig. 2. Determination of the fan characteristics
3. Determination of load characteristics
These considerations apply to the fan motors with adjustable flow rate achieved
by control of the motor angular speed. Flow rate of the fan at constant differential
pressure is proportional to the speed and differential pressure at constant flow rate
is proportional to the square of the speed [1, 5], i.e.:
2
ω ⎛ ω ⎞ (3)
Q= QN Δp = ⎜⎜ ⎟⎟ Δp N
ωN ⎝ ωN ⎠
Fan characteristic and optimal working point of the system for any angular
speed that differs from rated value, can be calculated on the basis of relation (3).
141
Figure 3 shows the set of the fan characteristics for ω = ω N and ω = 0,9; 0,8;
0,7; 0,6 ω N . Points of intersection of these curves and the characteristic of the
ventilated object determine operating states (conditions) of the system at relevant
speed values.
Fig. 3. Determination of characteristics for different angular speeds
Assuming that the aerodynamic resistance is proportional to the speed and

efficiency of the fan is speed independent than equations (3) show that the torque is
proportional to the square of speed. In the real system, however, flow rate and
differential pressure values change with variations of the angular speed of the fan.
Relationship of these two parameters is highly nonlinear in relation to both: fan and
ventilated system [1, 5].
In order to determine the fan generated full-load characteristic Tm (ω) of the
driving motor, using the Borland-Delphi programming environment, the computer
software has been developed. The software allows to:
− enter data (e.g. from file) describing the characteristics of the fan and the
system – Fig. 2 and 3;
− determine fan characteristics Δp = f (Q ) , Pw = f (Q ) and calculate an
optimal operating point – Fig. 2;
− determine set of fan characteristics for a selected range of angular speed and
compute the set of operating points {Qi , Δpi } – Fig. 3;
− determine the load characteristic Tm (ω) – Fig. 4.
142
Points (ωi , Tmi ) of load characteristics are calculated according to the formula:
q
Q ⋅ Δp i Q ⋅ Δp N Q ⋅ Δp N ⎛ ωi ⎞
Tmi = i + m0 N + βt N ⎜⎜ ⎟⎟ (4)
ωi ωN ωN ⎝ ωN ⎠
where m0 = ω N T0 QN ⋅ Δp N − relative static friction torque,
βt =η−wN
1
− 1 − m0 − relative friction coefficient, q − exponent characterizing
dependence of friction torque on speed.
An exemplary load characteristic constructed on the basis of equation (4) is shown

in Fig. 4.
Fig. 4. Load characteristics Tm(ω)
4. Characteristics of impeller pumps
In the mining industry mainly the impeller (rotodynamic) pumps are employed. In
the underground mines the pumps are used for drainage of mine faces and drawing
levels. Drainage pumps up to 6 kW are used on the rock walls, pumps up to 20 kW are
used for dewatering of mine faces. Units up to 90 kW are used for pumping water
between the intermediate levels. Pumps with power output of more than 1 MW are
used to extract water directly onto the surface (even more than 600 m). In strip mines
pumps are used to active drainage and boreholes dewatering [3].
Operation of the impeller pump is very similar to the axial flow fan operation. It
crates differential pressure between the inlet (suction) and the outlet (discharge) of
the pump. A characteristic aspect of the centrifugal pump is reverse flow of liquid
143
after the pump stops, caused by e.g. water column pressure. In such case the pump
start-up is often done at considerable high initial load and therefore, the driving
motor must have high starting torque. On the other hand, the inertia of the rotating
parts of the pump is noticeable smaller than inertia of the comparable capacity fan
and in this respect the pump star-up process is less demanding.
The basic parameter characterizing the pump is its rated output capability (i.e.
flow rate) QN [m3/s] at rated speed ωN. The other key parameters of the pump are:
pressure expressed in height H of the water column (called the pump pressure head
or lifting head or elevating head or delivery head), power P and efficiency η. They
are presented as a function of the pump flow rate Q and are called pump
characteristics. The most important characteristic is dependence of the pump lifting
head H on the flow rate Q, at constant angular speed. It is so called the throttle
curve. This characteristic can be stable (ACD curve, in Fig. 5) or non-stable (BCD
curve in Fig. 5). Stable characteristic is monotonically descending curve and
maximum of the pressure head H occurs at flow rate Q = 0. In the case of non-
stable characteristics, the same value of pressure head H corresponds to two values
of flow rate Q.
H
A C
D
E
B
Fig. 5. Flow rate-pressure head characteristic
The pump systems (installation) consist of numerous components. Hydraulic

losses are caused by viscous friction between the liquid and the installation. System
is characterized by the dependence of hydraulic losses (expressed in meters of
height of water column) on the flow rate. These losses depend on the diameter and
length of pipes, type and dimension of valves and pipe fittings and, of course, on
the velocity of flow. The operating point of the system is located at the intersection
of pump and installation characteristics – Fig. 6. Hydraulic losses ΔHr represent
sum of losses ΔHl on the sections of straight elements (pipes) and losses ΔHm on
144
other local obstacles (valves, elbows, etc). The pump pressure head Hc have to be
equal to hydraulic losses an can be expressed by the following equation [4]:
l c sr2 c2
H c = ΔH r = ΔH l + ΔH m = λ + ∑ ζ m sr (5)
d 2g 2g
were: λ is the dimensionless hydraulic resistance coefficient; l, d are is the length
and diameter of linear-elements, respectively; c is the flow velocity, g is the
gravitational acceleration; ζm is the dimensionless local hydraulic resistance
coefficient dependent on Reynolds number, type of obstacles, inner surface
roughness and valve opening degree.
After substituting c sr = 4Q πd 2 , equation (5) can be written as:
⎛ l 1 ∑ζm ⎞ 2 8 ⎛ l ⎞
H c = ⎜⎜ λ + ⎟⎟c sr = 2 4 ⎜ λ + ∑ ζ m ⎟Q 2 = CQ 2 (6)
⎝ d 2g 2g ⎠ π gd ⎝ d ⎠
where constant C is the characteristic coefficient of the installation, which is equal
to:
8 ⎛ l ⎞
C= 2 4⎜
λ + ∑ ζm ⎟ (7)
π gd ⎝ d ⎠
H
characteristic of the pump
working
point
characteristic of the installation
Fig. 6. Determination of operating point of the pump-installation
If there is difference in level Hg between the beginning and the ending of the
installation, than:
H c = H g + ΔH r = H g + CQ 2 (8)
145
Characteristics of the installation and the pump are shown in Fig. 7.

H
C
Hp’
B
Hp
A
Hg
Qp’ Qp Q
Fig. 7. Optimal working point of the system pump-installation, at constant speed
Point B of intersection of these curves is the optimal working point of

coordinates Q p , H p . If the hydraulic resistance of installation will increase, e.g.
due to the pipe bend or connecting smaller diameter pipe, the optimal working
point moves to point C. The pump lifting head will increase to the value H ′p and
its flow rate will decrease to the value Q ′p – Fig. 7.
5. Determination of the rated torque and pump generated load characteristics

of the driving motor
Useful output power Pp of the pump is equal to [2,4]:

Pp = γ ⋅ Q ⋅ H (9)
where γ is the specific gravity of liquid. If Q = 0 or H = 0, then Pp = 0 . This
(
means, that there is an optimal working point point Qopt , H opt , in which the )
impeller pump has a maximum power. Rated torque of the driving motor can be
expressed as:
PpN γ ⋅ QN ⋅ H N
TN = = (10)
ω N ⋅ η pN ω N ⋅ η pN
146
where ω N , η pN are the rated angular velocity and rated pump efficiency,
respectively.
Pump, in respect to the entire system, should be selected in such a way, that the
nominal working point should close to the optimal point, which in turn corresponds
to the maximum power output and the maximum efficiency of the pump.
The pump flow rate (at constant pressure head) is proportional to the angular
speed and pressure head (at constant flow) is proportional to square of the speed
[2, 4]. On that basis, the pomp characteristics and the optimal working point for
any speed ω different than the rated value ω N can be determined. Fig. 8 shows
the set of characteristics for ω = ω N and ω1 < ω N ; ω2 < ω1 . Points KN, K1, K2,
define the states of the system relevant to these speeds.
In the real systems the pump flow rate and pump pressure head vary freely with
the change of the pump speed. Relationship between these parameters in respect to
both: the pump and the installation is highly nonlinear.
Fig. 8. Effect of the angular speed on the shape of the pump flow (rate) characteristics H=f(Q)
In order to determine the load torque and load characteristic Tm (ω) generated
by the pump the computer software has been developed. The software allows to:
enter data describing the pump and the installation, determine the optimal working
point of system, determine the set of pump characteristics for selected range of
speed (Fig. 9), determine the set {Qi , H i } of optimal working points for different
speeds ωi and corresponding values of torque:
147
γ ⋅ Qi ⋅ H i
Tmi = (11)
ωi η pi
The points calculated according to (11) allow to determine load characteristics
Tm (ω) – Fig. 10. The expression (11) is valid only for angular speeds ωi greater
then critical value ωg, which corresponds to the point of intersection of the pump
and installation curves at Q = 0.
Fig. 9. Determination of the set of characteristics for various speeds
Fig. 10. Determination of load characteristics Tm(ω) for freely various flow rate and pressure head
148
After selecting the pump which ensures the desired flow at rated speed, the
motor rated torque can be determine according to (10). Computed load
characteristic (shown in Fig. 10) can be the basis for the variable speed drive
design.
6. Conclusions
The developed software allowing to determine the optimal working point of fan
or pump system can be used in the design of efficient driving motors. It also allows
to determine the load characteristics, which can be used in the design of the engine
control system.
References
[1] Gundlach W. R., Fundamentals of turbo machinery and energy systems (in Polish),
WNT, Warszawa, 2008.
[2] Jędral W., The impeller pumps (in Polish), PWN, Warszawa, 2001.
[3] Katalog firmy Flygt: Drainage pumps. Pumps for the mining industry (in Polish),
http://www.ittwww.pl/, 06.05.2010.
[4] Stępniewski M, Pumps (in Polish), WNT, Warszawa, 1985.
[5] Szklarski L., Skalny A., Strycharz J., Dynamics and control of stationary electric drives
in mining (in Polish), PWN, Warszawa, 1984.
149
Influence of disturbances in power network on torsional torques

in big power turbines-generator shafts
Józef Wiśniewski
Technical University of Lodz
90-924 Łódź, Stefanowskiego 18/22, e-mail: jwisniew@p.lodz.pl
The mechanical data of turbines and a generator of 1000 MW reference unit, working at
overcritical parameters, were established. Using the modal method, the characteristic
frequencies of torsional oscillations in shafts of rotating masses were calculated. The values
of torsional torques in shafts which can arise during faults in electrical networks, like close
distance short-circuits, the action of automatic reclosing, synchronization, as well influence
of electromagnetic torques from the network side causing the resonance oscillations of unit
were investigated.
1. Introduction
Planned in nearest future implementation of 1000 MW unit working with

overcritical parameters (program 50+) into the National Power System poses new
problems in the area of cooperation between the unit and the power system [1].The
planned parameters of units are the following: temperature 560÷580°C, and
pressure 25.8 MPa.
The issues of the rotating set of turbines and generator masses modeling are
presented. The aim of the modeling is to calculate torsional torques in shafts
between the individual masses during faults appearing in the power network.
Calculations were performed using the EMTP/ATP program.
The elements of the turbine - generator unit are connected with shafts (Fig. 1),
with limited mechanical endurance which can be exceeded during some faults.
HP IP LP1 LP2 LP3

G
Fig. 1. Structure of 1000 MW reference unit
Due to transport limits, it is not possible to use a single three phase transformer
with power over 1000 MVA in the generator-transformer unit. Power station
designers and producers of big power transformers take into consideration the use
of two three-phase transformers in parallel work or more reliable system of three
150
J. Wiśniewski / Influence of disturbances in power network on torsional …
one-phase transformers (and reserve one) with the summary power adequate to the
unit power, Fig. 2.
The EMTP/ATP [2] program enables us to simulate dynamics of the
synchronous generator with any number of turbines fixed on the common shaft.
The mechanical system is considered as linear, therefore spring-connected rotating
masses can be described by the second Newton law:
d2 d
J⋅ δ + D ⋅ δ + K ⋅ δ = Tturb − Tgen (1)
dt 2 dt
where: δ - angular positions, J - moments of inertia, D - damping coefficients, K -
stiffness coefficients of shafts between rotating masses, Tturb - torques applied to
turbines, Tgen - electromagnetic torque of a generator.
a) b)
2×600 MVA
1200 MVA
3
3
G G
3
1000 MW 1000 MW
c)
2
3×400 MVA (+1 in reserve)
2
G
1000 MW 2
Fig. 2. The scheme of high power unit work: a) traditional system, b) system with two three-phase
transformers, c) system with three one-phase transformers (and reserve one)
To determine the expected mechanical data of a similar power unit, the review of
literature describing these problems was presented. On the basis of many publications,
for example [3÷7], the mechanical parameters of units were compared. Inertia
constants of generators Hgen in [p.u.*s] units and their mean value Hgen_mean (about 0.82
p.u.*s) as well as sums of inertia constants of turbines propelling the generator Hturb_sum
and the mean value of these sums Hturb_sum_mean (about 2.96 p.u.*s) for the range of unit
power 500÷1100 MVA were shown in Fig. 3.
151
6 H gen
H [p.u.*s]
H turb sum
5
H gen_mean
4
H turb_sum_mean
3
2
Sgen [MVA]
1
0
500 600 700 800 900 1000 1100 1200
Fig. 3. Inertia constants of generators Hgen , sums of inertia constants of turbines Hturb_sum
and their mean values Hgen_mean ,and Hturb_sum_mean for different powers of units
The similar analysis for establishing stiffness coefficients K and damping

coefficients D was performed. Mean values of stiffness coefficients of shafts
between rotating masses Kmean in [p.u./rad] units for the same range of unit power
and their average value Kmean_average (about 83.5 p.u./rad) were presented in Fig. 4.
250
K [p.u./rad] K mean
200 K mean_average
150
100
50 Sgen [MVA]
0
500 600 700 800 900 1000 1100 1200
Fig. 4. Mean values of stiffness coefficients Kmean of shafts between rotating masses
and their average value Kmean_average
Parameters of the 1000 MW referent unit were set as: Inertia constants
(p.u.*s): HP=0.17, IP=0.4, LP1=0.6, LP2=0.6, LP3=0.6, GEN=0.8, Stiffness
152
coefficients (p.u./rad): HP-IP=150, IP-LP1=200, LP1-LP2=250, LP2-LP3=300,

LP3-GEN=350.
Using the transformation modal matrix Q with columns being eigenvectors of
expression J −1 ⋅ K , the equation (1) can be transformed to the modal formula. Its
solution enables to find the modal frequencies of system oscillations [2, 3, 5]. For
the considered unit these frequencies are: f = [1.44 13.37 23.77 32.98 38.17
44.9] Hz.
1
Mode 0 0
f0 = 1,44 Hz -1
1 1 2 3 4 5 6
Mode 1 0
f1 = 13,37 Hz -1
1 1 2 3 4 5 6
Mode 2 0
f2 = 23,77 Hz -1
1 2 3 4 5 6
1
Mode 3 0
f3 = 32,98 Hz -1
1 2 3 4 5 6
1
Mode 4 0
f4 = 38,17 Hz -1
1 2 3 4 5 6
1
Mode 5 0
f5 = 44,9 Hz -1
1 2 3 4 5 6
Fig. 5. The shapes of individual modes (normalized values of elements

of eigenvectors of transformation modal matrix Q)
The shapes of individual modes (i.e. normalized values of elements of

eigenvectors of the transformation modal matrix Q) are presented in Fig. 5. The
mode shape illustrates the maximal relative displacement of rotating masses during
the appearance of resonance with the given modal frequency.
Fig. 6 presents the dependence of maximal values of torsional torques Ti
(related to nominal torques of shafts) in individual shafts on frequency of the
sinusoidal external input torque with amplitude 1% of nominal electromagnetic
torque Tn_gen, affecting on the generator rotor.
It should be noticed that even this small effect of the network disturbance on the
rotor at some resonance frequencies results in the torsional torques exceeding
nominal values for these shafts.
153
10 T1
Ti [p.u.] T2
T3
T4
1
T5
0.1
0.01
f [Hz]
0.001
0 10 20 30 40 50
Fig. 6. Dependence of maximal values of torsional torques in individual shafts on frequency

of the sinusoidal external input torque with amplitude 1% Tn_gen, affecting a generator rotor
2. Calculations of torsional torques
Influence of network on generator rotor. Susceptibility of rotating masses to the

influence of a disturbing signal with the close to natural frequency was examined.
Such a disturbance can come from the network side as the generator current with
this frequency component. More and more power electronic devices in the network
encourage to such a situation.
The presence of the disturbing sinusoidal torque with the frequency equal to the
resonance frequency and amplitude of 1% of the rated torque was assumed. Using
the EMTP/ATP program the simulations were performed for the rated loaded
generator as well as for the synchronized one without a load. In the shaft sections
significant values of torsional torque appeared and their values were higher in the
case of the generator loaded. The modeled generator-transformer-line unit power
lead is shown in Fig. 7.
1000 MW Q1 TB Q2 Q3
400 kV
G Sk = 20 GVA
Fig. 7. Scheme of the generator - transformer unit for calculations

of mechanical shaft torques during selected faults
154
The amplitudes of torsional torque, expressed in relative units [p.u.], related to

the rated torque of each shaft sections, during the action of the disturbing torque
are presented in Fig.8. The generator is rated loaded.
5
Timax [p.u.] T1max
4 T2max
T3max
T4max
3 T5max
fres [Hz]
0
13.4 23.8 33.0 38.2 44.9
Fig. 8. The amplitudes of torsional torque, during the action of disturbing torque
with the value of 1% of the rated rotor torque Tr_gen
Short-circuits in power network. Fig. 9 presents the maximal values of torques

in shafts of the rotating unit at some faults in the network:
− 3-phase short-circuit on generator voltage bus bars for a rated loaded (A), (Fig.
10) and unloaded generator (B).
− 3-phase short-circuit on HV side of the unit transformer for a rated loaded
generator (C) and unloaded generator (D).
− 1-phase short-circuit in 400 kV unit line. 1-phase effective automatic reclosing
with time 0.4 s (E).
The calculations show that values of torques significantly exceed permissible
limits.
The calculations of torsional torques in the turbines - generator unit while
carrying out the generator synchronization using a generator breaker Q1 were
made. The analysis was performed for the difference in frequencies of the
generator and the network Δf = 0.1 Hz and for differences Δϕ of voltage phases in
the range 0°÷180°. Fig.11 presents the differential voltage Udiff (on Q1 breaker) and
torsional torques Ti in the shaft sections during synchronization with differences
Δϕ=5° of phase voltages.
155
Timax [p.u.] T1max

6 T2max
T3max
5 T4max
T5max
4
0
fault A fault B fault C fault D fault E
Fig. 9. The maximal values of torques in shafts of the rotating unit

at some short-circuits close to generator
5.0 Ti[p.u.]
2.5
0.0
-2.5
-5.0
4.98 5.00 5.02 5.04 5.06 5.08 5.10 5.12 [s] 5.14
(f ile 1000MW.pl4; x-v ar t) t: T1 t: T2 t: T3 t: T4 t: T5
3060
ω [res/min]
3030
3000
2970
4.98 5.00 5.02 5.04 5.06 5.08 5.10 5.12 [s] 5.14
(f ile 1000MW.pl4; x-v ar t) t: VEL_1 t: VEL_2 t: VEL_3 t: VEL_4 t: VEL_5 t: VEL_6
Fig. 10. Mechanical torques Ti and the generator and turbine rotation speed ωi during 3-phase short-
circuit on generator voltage bus-bars for a rated loaded generator (fault A)
156
12
Udiff [kV]
6
0
-6
-12
0.0 0.2 0.4 0.6 0.8 1.0 [s] 1.2
(f ile 1000MW.pl4; x-v ar t) t: DUD
0.6 Ti [p.u.]
0.3
0.0
-0.3
0.95 1.00 1.05 1.10 1.15 [s] 1.20
(f ile 1000MW.pl4; x-v ar t) t: T1 t: T2 t: T3 t: T4 t: T5
Fig. 11. The differential voltage Udiff (on Q1 breaker) and torsional torques Ti
during synchronization with differences Δϕ=5° of phase voltages
The dependence of maximal torque values on differences Δϕ of voltage phases

during synchronization is presented in Fig. 12.
10
Timax [p.u.]
4
T1max
T2max
2 T3max
T4max
T5max
Δϕ [°]
0
0 50 100 150
Fig. 12. Dependence of maximal torque values on differences Δϕ of voltage phases during
synchronization
157
The results of the analysis indicate that for the studied unit the maximal values
of torques for individual shafts appear at synchronization with the difference Δϕ of
voltage phases in the range 110°÷130°.
The phase of voltage at the moment of breaker closing does not influence the
value of torsional torques in shafts.
Calculation of shaft fatigue. The degree of the shaft fatigue caused by the
torsional oscillations depends on the amplitude of torsional stress and the number
of oscillations. The number of oscillations is influenced by the level of oscillations
damping. Figure 13 shows an example of the curves for determining the fatigue
caused by torsional oscillations [8].
Fatigue per incident
i=n
F = ∑ fi
⎛ ΔTi ⎞ i =1
1.0 ⎜⎜ ⎟⎟
⎝ TR ⎠
Relative torsional stressamplitude
Relative average ⎛ Tav ⎞

⎜⎜ ⎟⎟
0.5
torsional stress ⎝ TR ⎠ 0.5
1.0 [%]
0
10-5 10-4 10-3 10-2 10-1 10 0 10 1 10 2
Fatigue fi per one cycle
Fig. 13. Example of the curves for determining the fatigue caused by torsional oscillations [8]
This Figure allows to read the value of fatigue fi [%] caused by one swing
(horizontal axis) in relation to the value of the relative torsional stress amplitude
ΔTi / TR (vertical axis). The value fi can be read on the curve corresponding to the
relative average torsional stress Tav/TR. TR is the rupture stress causing the
destruction of the material, Tav means average stress (constant component) and ΔTi
means amplitude of an alternating component of torsional stress.
In Table 1 the impact of different system switching operations and fault
conditions on the fatigue of turbine-generator unit shafts is presented [8].
158
Table 1. Impact of different system switching operations and fault conditions on the fatigue
of turbine-generator shafts [8]
Fatigue per incident [%] negligible moderate severe

0.001 0.01 0.1 1 10 100
ΔP < 0.5 p.u.
Normal line switching
ΔP > 0.5 p.u.
Automatic
Synchronizing Manual (Δϕ < 10°, s = 0.7%)
Faulty (90°< Δϕ < 120°)
Full load rejection
Triple pole successful
reclosing unsuccessful
Line-to-ground fault
Single pole successful
reclosing unsuccessful
Clearing
Line-to-line fault successful
Reclosing
unsuccessful
Clearing
Line-to-line-to-ground
successful
fault Reclosing
unsuccessful
Clearing
Three phase fault successful
Reclosing
unsuccessful
Terminal-to-terminal unit HV side(system bus)
short circuit LV side (gen. leads)
HV side(system bus)
Three phase unit
short circuit LV side (gen. leads)
3. Conclusions
Calculations of system turbines and generator natural frequencies of oscillation

are an important element of the shaft designing process. They allow to avoid
working of the unit in conditions of hazard of arising oscillations caused by
external influences.
Calculations of torsional torque values can be useful for finding out a reason for
shaft damage, applying preventing means and for determining exploitation rules for
unit work.
159
The studied cases of short-circuits, synchronization and the effect of external

electromagnetic torque with resonance frequency on a generator rotor were
characterized by high values of torsional torques exceeding the permissible values.
Acknowledgment
The results presented in this paper were obtained from research work co-financed by the
National Centre of Research and Development in the framework of Contract
SP/E/1/67484/10 – Strategic Research Programme – Advanced technologies for obtaining
energy: Development of a technology for highly efficient zero-emission coal-fired power
units integrated with CO2 capture.
References
[1] Zagadnienia projektowania i eksploatacji kotłów i turbin do nadkrytycznych bloków

węglowych. Praca zbiorowa. Wyd. Politechniki Śląskiej, Gliwice 2010 (in Polish).
[2] EMTP Theory Book. Bonneville Power Administration, 1987.
[3] Machowski J., Białek J., Bumby J.: Power System Dynamics and Stability. John
Wiley & Sons Ltd, 1997.
[4] Jennings G., Harley R.: New index parameter for rapid evaluation of turbo-generator
subsynchronous resonance susceptibility. Electric Power Systems Research, 37, 1996.
[5] Jose A. Castillo J.: Turbo-generator torsional behavior using the participation factors
and considering the static loads model. Transmission and Distribution Conference and
Exposition: Latin America, IEEE/PES, 2008.
[6] Maljkovic Z., Stegic M., Kuterovac L.: Torsional oscillations of the turbine-generator
due to network faults. 14th International Power Electronics and Motion Control
Conference, EPE-PEMC, 2010.
[7] Tsai J.: A new single-pole switching technique for suppressing turbine-generator
torsional vibrations and enhancing power stability and continuity. IET Gener. Transm.
Distrib., 5, 2007.
[8] Joyce J., Kulig T.: Torsional Fatigue of Turbine-Generator Shafts Caused By Different
Electrical System Faults and Switching Operations. IEEE Transactions on Power
Apparatus and Systems, Vol. PAS-97, No. 5, Sept/Oct 1978.
160
Basic equations for analysis of a three-phase three-winding

transformer subjected to asymmetric load
Leszek Kasprzyk, Zbigniew Stein, Maria Zielińska
60 - 965 Poznań, ul. Piotrowo 3A, e-mail: {Leszek.Kasprzyk, Zbigniew.Stein,
Maria.Zielinska}@put.poznan.pl
The paper presents a set of equations enabling analysis of operation of three-phase

three-winding transformers under asymmetric load. For a selected asymmetric state of such
a transformer the computation leading to analysis of characteristic output parameters of the
transformer is carried out with the help of the Mathcad software.
1. Introduction
Asymmetric condition of a transformer is caused by asymmetry of the phase

loads or supplying voltages. The three-winding transformers used in electric-power
systems usually are not loaded with single-phase receivers (with neutral
conductor), but often with two-phase receivers. In industrial applications the
transformers are used under various receiver connections. The paper presents a set
of the equations that describe the effect of various load types on the distribution of
currents in the transformer windings and, in consequence, on voltage values in both
secondary windings. Making use of the analogies with the power transformers the
following definitions were introduced: the high (g), medium (s), and low (d)
voltage winding, irrespective of actual values of these voltages. In order to enable
researching the effect of three various values on the output parameters of the
transformer, the k1, k2, and k3 coefficients are used in the equations. Other
coefficients u1, v1, and w1 are introduced with a view to consider various values of
the angle ϕ of the receivers.
2. Basic relationships describing the considered problem
The transformer load is characterized by the impedances connected to particular

phases. Therefore, knowledge of the impedance values is necessary for analyzing the
transformer operation. It is assumed in the paper that the impedances are known. For
the medium voltage winding the impedances are determined as follows (1):
π
j0.307 u1 2
Z zsu (k 1 ) = (k 1 Z odng e 3 )
161
L. Kasprzyk, Z. Stein, M. Zielińska / Basic equations for analysis of a three-phase…
π
jk 2 0.307 v1 2
Z zsv (k 2 ) = ( Z odng e 3 ) (1)
π
j0.307 w1 2
Z zsw (k 3 ) = (k 3 Z odng e 3 )
Application of the transformation matrix (2):
⎛ Z1(k1, k2 , k3) ⎞ ⎛⎜1 a a ⎞⎟⎛ Utg1 ⎞
2
⎜ ⎟ 1 ⎜ ⎟ 1
⎜ Z2 (k1, k2 , k3) ⎟ = ⎜1 a a ⎟⎜ Utg2 ⎟
2
(2)
⎜ Z (k , k , k ) ⎟ 3 ⎜⎜1 1 1 ⎟⎟⎜ U ⎟ D(k1, k2 , k3)
⎝ 0 1 2 3⎠ ⎝ ⎠⎝ tg0 ⎠
enables calculating the symmetrical components of the impedance of the medium
voltage side (3):
1
Zs1 (k1 , k 2 , k 3 ) = ( Z zsu (k1 ) + aZ zsv (k 2 ) + a 2 Z zsw (k 3 ))
3
1
Zs 2 (k1 , k 2 , k 3 ) = ( Z zsu (k1 ) + a 2 Z zsv (k 2 ) + aZ zsw (k 3 )) (3)
3
1
Zs 0 (k1 , k 2 , k 3 ) = ( Z zsu (k1 ) + Z zsv (k 2 ) + Z zsw (k 3 ))
3
The currents of particular symmetrical components of the medium and low
voltage windings were calculated based on the relationship (4):
⎛ I1(k1, k2 , k3 ⎞ ⎛ M11 M12 M10 ⎞⎛ Utg1 ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟ 1
⎜ I2 (k1, k2 , k3 ⎟ = ⎜ M21 M22 M20 ⎟⎜ Utg2 ⎟ (4)
⎜ I (k , k , k ⎟ ⎜ M M M ⎟⎜ U ⎟ D(k1, k2 , k3)
⎝ 0 1 2 3 ⎠ ⎝ 01 02 00 ⎠⎝ tg0 ⎠
The values Utg1, Utg2, Utg0 in these formulas are the symmetrical components of
the high voltage winding (the primary one), while M and D are the elements of the
impedance matrix.
Particular terms of the impedance matrix M/D for the medium voltage winding
take a form (5):
Ms11(k1, k 2 , k3 ) = (Zugs + Zs0 (k1, k2 , k3 ))(Zs0 (k1, k2 , k3 ) + Zμ0 ) − Zs1(k1, k 2 , k3 )Zs2 (k1, k2 , k3 )
M s12 (k1 , k 2 , k 3 ) = Zs1 (k1 , k 2 , k 3 ) 2 − Zs 2 (k1 , k 2 , k 3 )( Zμ 0 + Zs 0 (k1 , k 2 , k 3 ))
M s10 (k1 , k 2 , k 3 ) = Zs 2 (k1 , k 2 , k 3 ) 2 − Zs1 ( k1 , k 2 , k 3 )( Zs 0 (k1 , k 2 , k 3 ) + Z ugs )
M s 21 (k1 , k 2 , k 3 ) = ( Zs 2 (k1 , k 2 , k 3 )) 2 − ( Zs1 (k1 , k 2 , k 3 ))( Zs 0 ( k1 , k 2 , k 3 ) + Zμ 0 ) (5)
Ms22 (k1, k2 , k3 ) = (Zugs + Zs0 (k1, k2 , k3 ))(Zs0 (k1, k2 , k3 ) + Zμ0 ) − Zs1 (k1, k2 , k3 )Zs2 (k1, k2 , k3 )
M s 20 (k1 , k 2 , k 3 ) = Zs1 (k1 , k 2 , k 3 ) 2 − Zs 2 (k1 , k 2 , k 3 )( Zs 0 ( k1 , k 2 , k 3 ) + Z ugs )
M s 01 (k1 , k 2 , k 3 ) = Zs1 (k1 , k 2 , k 3 ) 2 − Zs 2 (k1 , k 2 , k 3 )( Zs 0 (k1 , k 2 , k 3 ) + Z ugs )
162
M s 02 (k1 , k 2 , k 3 ) = Zs 2 ( k1 , k 2 , k 3 ) 2 − Zs1 (k1 , k 2 , k 3 )( Zs 0 (k1 , k 2 , k 3 ) + Z ugs )

M s 00 (k1 , k 2 , k 3 ) = ( Zs 0 ( k1 , k 2 , k 3 ) + Z ugs )) 2 − Zs1 (k1 , k 2 , k 3 ) Zs 2 (k1 , k 2 , k 3 )
The matrix determinant (6):

D s (k1 , k 2 , k 3 ) = D s1 (k1 , k 2 , k 3 ) + D s 2 (k1 , k 2 , k 3 ) + D s3 (k1 , k 2 , k 3 ) (6)
where Ds1, Ds2, Ds3 are presented as (7):
D s1 ( k1 , k 2 , k 3 ) = ( Zs 0 (k1 , k 2 , k 3 ) + Z ugs )( Zs 0 (k1 , k 2 , k 3 ) + Z ugs )( Zs 0 (k1 , k 2 , k 3 ) + Zμ 0 )
D s 2 (k1 , k 2 , k 3 ) = − Zs1 (k1 , k 2 , k 3 ) Zs 2 (k1 , k 2 , k 3 )[3Zs 0 (k1 , k 2 , k 3 ) + ( Z ugs + Z ugs + Zμ 0 )
D s 3 (k1 , k 2 , k 3 ) = Zs1 (k1 , k 2 , k 3 ) 3 + Zs 2 (k1 , k 2 , k 3 ) 3 (7)
Particular terms of the impedance matrix M/D for the low winding voltage take a
form (8):
Md11(k1, k 2 , k3 ) = (Zugd + Zd0 (k1, k 2 , k3 ))(Zd0 (k1, k 2 , k3 ) + Zμ0 ) − Zd1 (k1, k 2 , k3 )Zd2 (k1, k 2 , k3 )
M d12 (k1 , k 2 , k 3 ) = Z d1 (k1 , k 2 , k 3 ) 2 − Z d 2 (k1 , k 2 , k 3 )( Zμ 0 + Z d 0 (k1 , k 2 , k 3 ))
M d10 (k1 , k 2 , k 3 ) = Z d 2 (k1 , k 2 , k 3 ) 2 − Z d1 (k1 , k 2 , k 3 )( Z d 0 (k1 , k 2 , k 3 ) + Z ugd )
M d 21 (k1 , k 2 , k 3 ) = ( Z d 2 (k1 , k 2 , k 3 )) 2 − ( Z d1 (k1 , k 2 , k 3 ))( Z d 0 (k1 , k 2 , k 3 ) + Zμ 0 ) (8)
Md22 (k1, k 2 , k3 ) = (Zugd + Zd0 (k1, k 2 , k3 ))(Zd0 (k1, k 2 , k3 ) + Zμ0 ) − Zd1 (k1, k 2 , k3 )Zd2 (k1, k 2 , k3 )
M d 20 (k1 , k 2 , k 3 ) = Z d1 (k1 , k 2 , k 3 ) 2 − Z d 2 (k1 , k 2 , k 3 )( Z d 0 ( k1 , k 2 , k 3 ) + Z ugd )
M d 01 (k1 , k 2 , k 3 ) = Z d1 (k1 , k 2 , k 3 ) 2 − Z d 2 (k1 , k 2 , k 3 )( Z d 0 (k1 , k 2 , k 3 ) + Z ugd )
M d 02 (k1 , k 2 , k 3 ) = Z d 2 (k1 , k 2 , k 3 ) 2 − Z d1 (k1 , k 2 , k 3 )( Z d 0 ( k1 , k 2 , k 3 ) + Z ugd )
M d 00 (k1 , k 2 , k 3 ) = ( Z d 0 ( k1 , k 2 , k 3 ) + Z ugd )) 2 − Z d1 ( k1 , k 2 , k 3 ) Z d 2 (k1 , k 2 , k 3 )
The matrix determinant is given by the equation (9):
D d (k1 , k 2 , k 3 ) = D d1 (k1 , k 2 , k 3 ) + D d 2 ( k1 , k 2 , k 3 ) + D d 3 (k1 , k 2 , k 3 ) (9)
where Ds1, Ds2, Ds3 are presented as (10):
D d1 (k1 , k 2 , k 3 ) = ( Z d 0 (k1 , k 2 , k 3 ) + Z ugd )( Z d 0 (k1 , k 2 , k 3 ) + Z ugd )( Z d 0 (k1 , k 2 , k 3 ) + Zμ 0 )
D d 2 ( k1 , k 2 , k 3 ) = − Z d1 (k1 , k 2 , k 3 ) Z d 2 (k1 , k 2 , k 3 )[3Z d 0 (k1 , k 2 , k 3 ) + ( Z ugd + Z ugd + Zμ 0 )
D d 3 (k1 , k 2 , k 3 ) = Z d1 (k1 , k 2 , k 3 ) 3 + Z d 2 (k1 , k 2 , k 3 ) 3 (10)
Assuming that only asymmetric impedances of the receivers are considered and
only positive-sequence voltage of the value equal to the rated voltage Utng is taken
into account, the currents of particular symmetrical components of the medium
voltage winding are determined by the relationships (11):
1
I s1 (k1 , k 2 , k 3 ) = (M s11 (k1 , k 2 , k 3 ) U ntg )
D s ( k1 , k 2 , k 3 )
163
1
I s 2 ( k1 , k 2 , k 3 ) = (M s 21 (k1 , k 2 , k 3 ) U ntg ) (11)
D s (k1 , k 2 , k 3 )
1
I s 0 (k1 , k 2 , k 3 ) = (M s 01 (k1 , k 2 , k 3 ) U ntg )
D s (k1 , k 2 , k 3 )
The load impedances of particular phases of the low voltage winding are written
down in the form (12):
π
− j0.45⋅2
Z zdu (k1 ) = ( k1 Z odng ⋅ 1 e 3)
π
− jk 2 0.2⋅2
Z zdv ( k 2 ) = ( Z odng ⋅ 1.5 e 3) (12)
π
− j0.3⋅2
Z zdw (k 3 ) = ( Z odng ⋅ 0.8 e 3)
Application of the transformation matrix enabled calculating symmetrical

components of the load impedance of the low voltage side defined by the equations
(13):
1
Z d1 (k1 , k 2 , k 3 ) = ( Z zdu (k1 ) + aZ zdv (k 2 ) + a 2 Z zdw (k 3 ))
3
1
Z d 2 (k1 , k 2 , k 3 ) = ( Z zdu (k1 ) + a 2 Z zdv (k 2 ) + aZ zdw (k 3 )) (13)
3
1
Z d 0 (k1 , k 2 , k 3 ) = ( Z zdu (k1 ) + Z zdv ( k 2 ) + Z zdw (k 3 ))
3
Under the assumptions identical to the ones formerly adopted for the medium
voltage side, the currents of particular symmetrical components of the low voltage
winding are determined by the relationship (14):
1
I d1 (k1 , k 2 , k 3 ) = ( M d11 (k1 , k 2 , k 3 ) U ntg )
D d (k1 , k 2 , k 3 )
1
I d 2 (k1 , k 2 , k 3 ) = (M d 21 ( k1 , k 2 , k 3 ) U ntg ) (14)
D d (k1 , k 2 , k 3 )
1
I d 0 ( k1 , k 2 , k 3 ) = (M d 01 (k1 , k 2 , k 3 ) U ntg )
D d (k1 , k 2 , k 3 )
Once the currents of medium and low voltage sides are calculated, the current
of high voltage winding (Ig) should be calculated. In order to find these currents the
currents of medium and low voltage windings should be summed up, that is
presented by the equations (15):
I g1 (k1 , k 2 , k 3 ) = I s1 ( k1 , k 2 , k 3 ) + I d1 (k1 , k 2 , k 3 )
I g 2 ( k1 , k 2 , k 3 ) = I s 2 ( k1 , k 2 , k 3 ) + I d 2 ( k 1 , k 2 , k 3 ) (15)
164
I g 0 (k1 , k 2 , k 3 ) = I s 0 (k1 , k 2 , k 3 ) + I d 0 (k1 , k 2 , k 3 )

The phase currents (a,b,c) of particular windings are obtained in result of
transformation of the symmetrical components to the phase form. This is made
with the help of the transformation matrix (16):
⎛ Ia (k1, k2 , k3) ⎞ ⎛⎜1 a a ⎞⎟⎛ I1(k1) ⎞
2
⎜ ⎟ 1 ⎜ ⎟
⎜ Ib (k1, k2 , k3) ⎟ = ⎜1 a a ⎟⎜ I2 (k2 ) ⎟
2
(16)
⎜ I (k , k , k ) ⎟ 3 ⎜⎜1 1 1 ⎟⎟⎜ I (k ) ⎟
⎝c 1 2 3 ⎠ ⎝ ⎠⎝ 0 3 ⎠
The phase currents of the medium voltage winding are determined by the
relationships (17):
I sa (k1 , k 2 , k 3 ) = I s1 (k1 , k 2 , k 3 ) + I s 2 (k1 , k 2 , k 3 ) + I s 0 (k1 , k 2 , k 3 )
I sb (k1 , k 2 , k 3 ) = a 2 I s1 (k1 , k 2 , k 3 ) + aI s1 (k1 , k 2 , k 3 ) + I s 0 (k1 , k 2 , k 3 ) (17)
2
I sc (k1 , k 2 , k 3 ) = aI s1 ( k1 , k 2 , k 3 ) + a I s 2 (k1 , k 2 , k 3 ) + I s 0 (k1 , k 2 , k 3 )
The phase currents of the low voltage winding are described by the equations (18):
I da (k1 , k 2 , k 3 ) = I d1 (k1 , k 2 , k 3 ) + I d 2 (k1 , k 2 , k 3 ) + I d 0 ( k1 , k 2 , k 3 )
I db (k1 , k 2 , k 3 ) = a 2 I d1 (k1 , k 2 , k 3 ) + aI d1 (k1 , k 2 , k 3 ) + I d 0 (k1 , k 2 , k 3 ) (18)
2
I dc (k1 , k 2 , k 3 ) = aI d1 (k1 , k 2 , k 3 ) + a I d 2 ( k1 , k 2 , k 3 ) + I d 0 (k1 , k 2 , k 3 )
The phase currents of the high voltage winding are determined by the
relationships (19):
I ga (k1 , k 2 , k 3 ) = I g1 (k1 , k 2 , k 3 ) + I g 2 (k1 , k 2 , k 3 )
I gb (k1 , k 2 , k 3 ) = a 2 I g1 ( k1 , k 2 , k 3 ) + aI g1 (k1 , k 2 , k 3 ) (19)
I gc ( k1 , k 2 , k 3 ) = aI g1 (k1 , k 2 , k 3 ) + a 2 I g 2 (k1 , k 2 , k 3 )
The equations (20) present the relationships necessary for calculating the
symmetrical components of the output voltages of the medium and low sides.
E1 (k1 , k 2 , k 3 ) = U ntg − (I s1 (k1 , k 2 , k 3 ) + I d1 ( k1 , k 2 , k 3 )) Z ug1
U 2 tg (k1 , k 2 , k 3 ) = I g 2 (k1 , k 2 , k 3 ) Z ug 2
U 0 tg (k1 , k 2 , k 3 ) = (I s 0 ( k1 , k 2 , k 3 ) + I d 0 (k1 , k 2 , k 3 )) Z ug 0
U s1 (k1 , k 2 , k 3 ) = E1 ( k1 , k 2 , k 3 ) − I s1 (k1 , k 2 , k 3 ) Z ug1
U d1 (k1 , k 2 , k 3 ) = E1 ( k1 , k 2 , k 3 ) − I d1 (k1 , k 2 , k 3 ) Z ud1 (20)
U s 2 (k1 , k 2 , k 3 ) = I s 2 (k1 , k 2 , k 3 ) Z us1
U d 2 (k1 , k 2 , k 3 ) = I d1 (k1 , k 2 , k 3 ) Z ud1
U s 0 (k1 , k 2 , k 3 ) = I s 0 (k1 , k 2 , k 3 ) Z us1
U d 0 (k1 , k 2 , k 3 ) = I d 0 (k1 , k 2 , k 3 ) Z ud1
165
The phase voltages of the medium voltage side are calculated according to the
formulas (21)
U sA (k1 , k 2 , k 3 ) = U s1 (k1 , k 2 , k 3 ) + U s 2 (k1 , k 2 , k 3 ) + U s 0 (k1 , k 2 , k 3 )
U sB (k1 , k 2 , k 3 ) = a 2 U s1 (k1 , k 2 , k 3 ) + aU s 2 ( k1 , k 2 , k 3 ) + U s 0 (k1 , k 2 , k 3 ) (21)
U sC ( k1 , k 2 , k 3 ) = aU s1 (k1 , k 2 , k 3 ) + a 2 U s 2 ( k1 , k 2 , k 3 ) + U s 0 (k1 , k 2 , k 3 )
The phase voltages of the low voltage side are calculated according to the
relationship (22)
U dA (k1 , k 2 , k 3 ) = U d1 (k1 , k 2 , k 3 ) + U d 2 (k1 , k 2 , k 3 ) + U d 0 (k1 , k 2 , k 3 )
U dB (k1 , k 2 , k 3 ) = a 2 U d1 (k1 , k 2 , k 3 ) + aU d 2 ( k1 , k 2 , k 3 ) + U d 0 ( k1 , k 2 , k 3 ) (22)
U dC (k1 , k 2 , k 3 ) = aU d1 (k1 , k 2 , k 3 ) + a 2 U d 2 (k1 , k 2 , k 3 ) + U d 0 (k1 , k 2 , k 3 )
3. Example calculation results

The formulas derived above have been used for calculating selected output
parameters of a three-winding transformer of the power of 20/10/10 MVA and the
voltages 110, 33, 6.6kV. In order to carry out the example calculation it was assumed
that the transformer is subject to asymmetrical load at the low voltage side, and to
symmetrical load at the medium voltage side. The values of the output parameters are
the best illustrated by the characteristics plotted against the load parameters.
The Figures 1-3 present the relationship between the asymmetry factor Ud2/Ud1
and the parameters k1, k2, and k3, respectively. Figures 4-6 show the plots of the
currents in the high, low, and medium voltage sides, respectively, against the k1
parameter. The Figures 7 and 8 show the plots of the currents in the medium and
low voltage sides, respectively, against the k1 parameter.
1.9
( )
Ud2 k1 , 1 , 1 ⋅ 100
(
Ud1 k1 , 1 , 1 )
( )
Us2 k1 , 1 , 1 ⋅ 1000
1.25
(
Us1 k1 , 1 , 1 )
0.6
0.8 1 1.2
k1
Fig. 1. The relationship between the asymmetry factor and the k1 parameter
166
1.3
(
Ud2 1 , k2 , 1 ⋅ 100)
(
Ud1 1 , k2 , 1 )
0.95
( )
Us2 1 , k2 , 1 ⋅ 1000
(
Us1 1 , k2 , 1 )
0.6
0.8 1 1.2
k2
1.3
(
Ud2 1, 1, k3 ⋅ 100 )
(
Ud1 1, 1, k3 )
0.95
(
Us2 1, 1, k3 ⋅ 1000 )
Us1 1, 1, k3 ( )
0.6
0.8 1 1.2
k3
1.4
(
I ga k 1 , 1 , 1 )
I ntg
(
I gb k 1 , 1 , 1 )
I ntg 1.25
(
I gc k 1 , 1 , 1 )
I ntg
1.1
0.8 1 1.2
k1
Fig. 4. The relationship between the current in the high voltage side and the k1 parameter
167
0.95
(
Ida k1 , 1 , 1 )
Intg
(
Idb k1 , 1 , 1 )
Intg 0.75
(
Idc k1 , 1 , 1 )
Intg
0.55
0.8 1 1.2
k1
Fig. 5. The relationship between the current in the low voltage side and the k1 parameter
0.55
(
Isa k1 , 1 , 1 )
Intg
(
Isb k1 , 1 , 1 )
Intg 0.43
(
Isc k1 , 1 , 1 )
Intg
0.3
0.8 1 1.2
k1
Fig. 6. The relationship between the current in the medium voltage side and the k1 parameter
1.045
(
U sA k 1 , 1 , 1 )
U ntg
(
U sB k 1 , 1 , 1 )
U ntg 1.042
(
U sC k 1 , 1 , 1 )
U ntg
1.038
0.8 1 1.2
k1
Fig. 7. The relationship between the voltage at the medium voltage side and the k1 parameter
168
(
U dA k1 , 1 , 1 )
U ntg
(
U dB k1 , 1 , 1 )
U ntg 0.98
(
U dC k1 , 1 , 1 )
U ntg
0.96
0.8 1 1.2
k1
Fig. 8. The relationship between the voltage at the low voltage side and the k1 parameter
4. Conclusions
Impedance asymmetry of the receivers at the medium and low voltage sides
induce asymmetric current distribution in the transformer windings. In result
thereof the output voltages and, in consequence, the voltage asymmetry factors are
asymmetric too. Therefore, the load impedances should be so chosen as to avoid
not only the excess of the rated current level at the high voltage side caused by the
asymmetric load but, first of all, as to keep allowable values of the asymmetry
factors U2/U1.
References
[1] Kasprzyk L., Stein Z., Zielińska M., Analiza wpływu niesymetrycznych obciążeń
transformatora na jego charakterystyczne wielkości wyjściowe. Materiały Konferencji
„Zastosowania Komputerów w Elektrotechnice”, ZKwE 2010, Poznań, 2010.
169
Model predictive control of the two-mass drive system

with mechanical backlash
Piotr J. Serkies, Krzysztof Szabat
Wrocław University of Technology
50-372 Wrocław, ul. Smoluchowskiego 19, e-mail: Piotr.Serkies@ pwr.wroc.pl;
Krzysztof.Szabat@ pwr.pl
In the paper an application of the model predictive controller for the drive system with
elasticity and backlash is presented. In the introduction the control problem of the drive
system with mechanical backlash is introduced. Next, the mathematical model of the drive
is presented. Then the predictive algorithm is described briefly. The performance of the
predictive controller is examined in the simulation study. The influence of the changing of
the matrix Q is tested.
1. Introduction
The nonlinearities of the mechanical part of the drive, such as friction or backlash
decrease the performance the system [1]-[4]. The mechanical backlash evident in
gearboxes shortens the life-time of the whole drive system [2]-[4]. This is especially
evident in multi-mass systems in which the load (machine) is connected to a driving
motor through one or multiple flexible shafts. Excessive shaft twists and poorly
damped torsional vibrations are detrimental to the drive’s performance, greatly
decreasing product quality and system reliability, and in some cases leading to
instability and failure of the entire drive system. This problem commonly occurs in
rolling-mill drives, belt-conveyors, paper machines, robotic-arm drives including
space manipulators, servo-drives and throttle systems [1]-[12].
The control methodologies developed in this type of the drive may be based on
linear PI controllers with an additional feedback [1, 3, 5, 6], state controllers [6], or
robust controllers using the H∝ control law [7].
In recent years, model predictive control (MPC) has been widely investigated
for its potential in controlling modern electrical drives and power electronics
circuits [15]–[18]. Predictive control presents several advantages that make it
suitable for the control of power converters and drives. The central feature of MPC
is that it enables the process operating and physical constraints (due to e.g. resource
limitations, operational or safety concerns as well as limits arising from various
economic objectives) to be taken directly into consideration in the control problem
formulation so that any potential constraint violations are anticipated and
prevented. Additionally, as the control input is obtained by solving an optimization
170
P. J. Serkies, K. Szabat / Model predictive control of the two-mass…
problem at each sampling time, it can ensure truly optimal performance of the
closed-loop control system [8]-[12].
The main contribution of this paper is the design and validation of an explicit
model predictive controller for a two-mass elastic drive system with mechanical
backlash. The explicit version of the MPC algorithm presented here does not
involve complex optimization to be performed in a control unit but requires only a
piecewise linear function evaluation, which can be realized through a simple look-
up table approach. The effect of the additional input of the system, on the drive
dynamic not considered in the literature is examined. Also the robustness of the
control structure to the changes of the mechanical gap is considered.
2. The mathematical model of the drive
Many industrial drive systems can be modeled as two-mass systems, where the
first mass represents the moment of inertia of the motor and the second mass refers
to the moment of inertia of the load machine. In this paper, the commonly used
inertia-free-shaft dual-mass system model will be employed, which is described by
the following normalized differential equations
⎡ −1 ⎤
0 0 0 0 ⎡1⎤ ⎡0⎤
⎡ω1 ⎤ ⎢ T1Tc ⎥ ⎡ω1 ⎤ ⎢ ⎥
⎢ω ⎥ ⎢ ⎥ ⎢ ⎥ T1 ⎢−1⎥
⎢ 1 ⎥ ⎢ω2 ⎥ ⎢ ⎥ ⎢T ⎥
d⎢ ⎥ ⎢ 0 0 (1)
2
0 0 0 ⎢ ⎥
T2Tc ⎥ ⋅ ⎢α1 ⎥ + ⎢ ⎥ ⋅ [me ] + ⎢ 0 ⎥ ⋅ [mL ]
2
⎢α1 ⎥ =
dt ⎢ ⎥ ⎢ 1 0 0 0 ⎥
0 ⎢α ⎥ ⎢ ⎥
0 ⎢ ⎥
⎢α2 ⎥ ⎢ ⎥ ⎢ 2⎥ 0 ⎢0⎥
⎢⎣ms ⎥⎦ ⎢ 0 1 0 0 0 ⎥ ⎢m ⎥ ⎢ ⎥ ⎢ ⎥
⎢ϕ(Δθ ) −ϕ(Δθ ) ⎥ ⎣ s⎦ ⎢ 0⎥ ⎣⎢ 0 ⎦⎥
0 0 0 ⎦ ⎣ ⎦
⎣
where: ω1 ,ω2 – motor and load speeds, me, ms, mL – electromagnetic, shaft ad load
torques, α2 – shaft position (on the load side), T1, T2 – mechanical time constant of
the motor and the load machine, Tc – stiffness time constant, α1, α2 – mechanical
angle of the motor and the load machine, φ(Δθ) – the function of the mechanical
couplings.
The backlash is described by the following function:
⎧Δθ − ε dla Δθ > ε
⎪
ϕ (Δ θ ) = ⎨ 0 dla Δθ < ε (2)
⎪Δθ + ε dla Δθ < ε
⎩
Δθ = α 1 − α 1 (3)
where: ε is the backlash width. The block diagram of the considered system is
presented in Fig.1. The main parameters of the considered system are as follows:
T1=T2=203ms oraz Tc=1,2ms.
171
Fig. 1. The block diagram of the considered system
3. The predictive control structure
The control structure for the considered electrical drive system is shown in
Fig. 2. The inner torque loop (here for simplicity represented as a single block) is
composed of the power converter, electromagnetic part of the motor, current sensor
and respective current or torque controller. As this control loop is designed to
provide sufficiently fast torque control, it can be approximated by an equivalent
first order term with a small time constant. For a well-tuned torque controller, the
drive machines could be AC or DC motors without any impact on the outer speed
control loop. The outer loop consists of the mechanical part of the motor, speed
sensor and speed controller with reference ωr. For the estimation of signals
required in the MPC control structure, the Luenberger observer is used in this
paper.
Δθ
T1 T2
mer me Tc
ω1 ms ω2
ω ref
ωˆ1 , ωˆ 2 , mˆ s , mˆ L
Fig. 2. The schematic diagram of the considered control structure
In its core, Model Predictive Control employs an explicit process model that is
used to predict the effect of future actions of the manipulated variables on the
process output. The model choice is open but typically the following linear
discrete-time state-space form is considered [13]-[14]:
172
x(t + 1) = Ax(t ) + Bu(t )

(4)
y (t ) = Cx(t )
where x(t)∈ℜn, u(t)∈ℜm, y(t) ∈ℜp are the state, input and output variables,
respectively, and A∈ℜn×n, B∈ℜ n×m C ∈ℜ p×n are known time-invariant system
matrices.
Let yk denote the prediction of the output variable at a future time k, given the
input sequence uk, an initial state x0 and the model (2). At each time step k, an
MPC algorithm attempts to optimize future plant behavior while respecting the
system input/output constraints by solving the following optimization problem
[13]-[14]
min Np N c −1
u 0 ,..., u TN c −1
T ∑ y kTQy k + ∑ u kT Ru k
k =0 k =0
(5)
subject to: u min ≤ u ≤ u max k = 0,1,L , N c −1 (6)

x min ≤ x ≤ x max k = 0,1,L , N p
x k +1 = Ax k + Bu k , k ≥ 0,
y k = Cx k k ≥ 0,
x 0 = x(0),
where: Q ≥ and R > 0 denote the weighting matrices and Np and Nc are the
prediction and control horizons, respectively. Constraints umin and umax arise
naturally from physical restrictions of control actuators whereas restrictions due to
safety, quality, environmental and economic targets may be incorporated in xmin
and xmax.
The implementation of the MPC controller amounts to solving problem (3) on-
line for a given x0 in a receding horizon fashion. This means that, at time k, only
the first element u0* of the optimal input sequence is applied to the plant and the
remaining future control actions u1* ,...,u*Nc-1 are discarded. At the next time step
the whole procedure is repeated for the new measured or estimated output y(k + 1)
[13]-[14]. This strategy can be computationally intensive for systems with fast
sampling requirements thus greatly limiting the scope of applicability to systems
with relatively slow dynamics. Alternatively, rather than using the initial state x0 to
“update” the optimization problem (3) at each time k, the idea is to treat the state
vector as a parameter vector and then solve problem (3) off-line for all realizations
of x0 within a predefined set of states using multi-parametric programming [15]-
[20]. In this strategy, the parameter space is subdivided into characteristic regions
where the optimizer is given as an explicit piecewise affine (PWA) function of the
parameters:
u 0* ( x0 ) = K r x0 + g r , ∀x ∈ Pr (6)
where Pr are polyhedral sets defined as
{
Pr = x ∈ ℜ n | H r x ≤ d r , } r = 1,...N r (7)
173
Here Nr > 0 represents the total number of polyhedral regions in the partition. The
main advantage of this approach is that the optimal input u0* for a given initial state
x0 can be obtained by evaluating a PWA function in the control unit thus greatly
simplifying the controller implementation process as numerical optimization is no
longer required [15]-[20].
As an internal model of the plant used for the optimization problem, a linear
model of the two-mass system is used (7). Consequently, the mechanical backlash
is omitted for the prediction of the future response of the object.
⎡ −1 ⎤
0 0 0 0 0⎥
⎡⎡ω ⎤⎤ ⎡ ⎤
1
⎡ω1 ⎤ ⎢ T1
⎢ω ⎥ ⎢ ⎥ ⎢⎢ 1 ⎥⎥ ⎢T ⎥
1 −1⎥ ω
⎢ 2⎥ ⎢ 0 0 ⎢⎢ 2 ⎥⎥ ⎢ 0 ⎥
1
0 0
d ⎢α1 ⎥ ⎢
⎢ T2 T2 ⎥ ⎢⎢α ⎥⎥ ⎢ ⎥ (8)
0 ⎥ ⋅ ⎢⎢ ⎥⎥ + ⎢⎢ 0 ⎥⎥ ⋅ [me ]
1
⎢ ⎥ 1 0= 0 0 0
dt ⎢α2 ⎥ ⎢ ⎥ ⎢⎢α2 ⎥⎥
⎢ms ⎥ ⎢ 10 −11 0 0 0 0 ⎥ ⎢⎢ ⎥⎥ ⎢ 0 ⎥
⎢ ⎥ ⎢ ⎥ ms ⎢ ⎥
0 0 0 0 ⎥ ⎢⎢ ⎥⎥ ⎢ 0 ⎥
⎣⎢mL ⎦⎥ ⎢TC TC ⎢⎣⎢mL ⎦⎥⎦⎥ ⎢ 0 ⎥
⎢ ⎥ ⎣ ⎣ ⎦
⎣0 0 0 0 0 0⎦
The considered model includes two additional variables, namely the reference
speed and the load torque. The main design parameters of the MPC controller are
the following: prediction horizon, the selection of the inputs, and the values of the
cost matrices Q and R.
3. The study
In this section, the proposed single-loop explicit MPC control strategy for the
drive system with an elastic coupling will be evaluated through simulations. A
primary design objective for the MPC controller is to ensure that the load speed
response follows the set-point with the desired dynamics. This needs to be
achieved without generating excessive shaft torque responses and without violating
the input and output constraints of the drive. The first two requirements can be
addressed by defining the following auxiliary output variables:
(
⎧ y1 = ω1 − ω ref )
⎪
(
⎪ y2 = ω2 − ω
ref
)
⎨ (9)
⎪ y3 = (ms − mL )
⎪⎩ y 4 = (ω1 − ω 2 )
where: y1, y2 account for tracking performance, y3 relates to load-shaft torque
imbalance. The factor y4 reduces the difference of the speed which reduces the
maximal shaft torque. Due to (9) the reference speed variable and the disturbance
torque need to be directly incorporated into the drive system model.
174
The task of the MPC controller is to bring the output variables (9) to zero by
manipulating mer while respecting the safety and physical limitations of the drive
system:
− mer ≤ mer ≤ mer (10)
− ms ≤ ms ≤ ms (11)
The selection of the prediction and control horizons is a compromise between
the drive performance and computational complexity. In practice, Nc ≤ Np to avoid
large computational burden for the standard MPC and large number of regions for
the explicit MPC. In this paper, the state variables will be predicted using a 10ms
prediction window, whereas the control input will be calculated over 1ms intervals.
This translates to Np = 10, Nc = 2. The R is set to the 0,0001. The limit of the
electromagentic torque is set to ±3[p.u], and the shaft torque to ±1.5[p.u].
In the first step the influence of the fourth input and the changing of the value q4
on the value of the cost function and the location of the MPC controller regions is
presented. The investigated characteristic is depicted in Fig. 3. As can be
concluded from the presented characteristic, the variation of the q4 value influences
the cost function significantly (Fig. 3a). Also the controller regions are changing
according to the value q4.
diag(Q)=[7.5 15 1 0]
b)
1.2
ω [p.u]
0
2
-1.2
-1.2 0 1.2
q4=19 ω [p.u]
1
50 diag(Q)=[7.5 15 1 10]
c)
1.2
40
ω [p.u]
30
q4=0 0
2
J
20
10 -1.2
-1.2 0 1.2
0.2 ω [p.u]
0 1
0.1 d) diag(Q)=[7.5 15 1 19]
0.2 1.2
0.1 0
0 -0.1
ω [p.u]
-0.1 ω2 [p.u]
ω1 [p.u] -0.2 0
-0.2
2
-1.2
-1,2 0 1,2
ω [p.u]
1
Fig. 3. The relationship between the cost function value and the value of the speed for the two value
of the q4 (a), and controller regions (b,c,d) for diag(Q)=[7.5 15 1 0] (b), diag(Q)=[7.5 15 1 10] (c),
diag(Q)=[7.5 15 1 29] (d)
175
The transients of the system state variables for different q4 value are presented
in Fig. 4. The enlarged fragments of those transients are displayed in Fig. 5.
a) 3
b) 0.5
q4=0
2 0.4
1 0.3
me [p.u]
ω1 [p.u]
0 0.2
q4=10
-1 q4=0
q4=10 0.1 q4=29
-2 q4=29
0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t[s] t [s]
c)
1.5 d) 0.4
q4=0 q4=0
q4=29
0.3
1
ms [p.u]
ω2 [p.u]
0.2
0.5 q4=10 q4=10
0.1
0
q4=29
0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t[s] t [s]
Fig. 4. Transients of the system state variable: electromagnetic torque (a), motor speed (b), shaft
torque (c) and load speed (d) for different value of q4
a) b) 0.2
3 q4=0
q4=0
2 0.15
q4=10
1 q4=10
m e [p.u]
ω1 [p.u]
0.1
0
-1 0.05
-2
q4=29 q4=29
0
0 0.01 0.02 0.03 0.04 0.05 0 0.01 0.02 0.03 0.04 0.05
t[s] t [s]
c)
0.8 d) 0.2
q4=0
0.6 0.15
q4=10 q4=0
m s [p.u]
ω2 [p.u]
0.4 0.1
q4=10
0.2 q4=29 0.05
0
q4=29
0
0 0.01 0.02 0.03 0.04 0.05 0 0.01 0.02 0.03 0.04 0.05
t[s] t [s]
Fig. 5. Enlarged transients of the system state variable: electromagnetic torque (a), motor speed (b),
shaft torque (c) and load speed (d) for different value of q4
As can be deduced from the presented figures, the increase of the value q4
makes the torques decline when the gap of the backlash is closing (Fig. 5a,c). At
the same time, the increase of the q4 value slowest the system responses to the
176
changing of the load torque. The dynamics of the control structure is reduced and
the system needs more time to eliminate the static error. It stems from the fact that
the bigger value of the q4 counteracts the large difference between the motor and
the load speeds. Thus, that the motor cannot accelerate as dynamically as in the
case when the input q4 is neglected.
Then, the influence of the backlash width on the dynamics of the system has been
investigated. At this point the values of the matrix Q are constant and equal
(diag(Q)=[7.5 15 1 29]). The transients of the system are presented in Fig. 6.
Similarly as in the previous case the fragments of the transients are shown in Fig. 7.
It is obvious from the presented results that the considered system is robust to
changes of the backlash gap within the selected range without visible performance
degradation. The two backlash parts hit each other softly. Even the system with a
relatively big gap equal to 11,4 degree works correctly. The only noticeable
difference between the tested systems exists during the start-up. SO, the bigger the
gap of the backlash, the slower response time of the plant. This effect can be
reduced by decreasing in the value q4. In that case the speed of the motor will starts
more dynamical, which results in the shortening the settling time of the plant. The
reaction of the system to the changing of the load torque is the same in every tested
structure.
Then the system was tested for the nominal value of the speed. In this case the
shaft torque was limited at the maximal set value by the MPC algorithm. The
obtained results for two values of q4 are demonstrated in Fig. 8.
a) 3 b) 0.4
o
0
2
0.3 o
o 0
1 11.4
m [p.u]
ω [p.u]
0.2 o
11.4
0
o
-1
5.7 0.1
o
5.7
-2 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t[s] t [s]
c)
d)
1.5 0.4
o
o 0
0
1 0.3
o
11.4
m [p.u]
ω [p.u]
0.5 0.2 o
11.4
0 o 0.1
5.7 o
5.7
-0.5 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t[s] t [s]
Fig. 6. Transients of the system state variables: electromagnetic torque (a), motor speed (b), shaft
torque (c) and load speed (d) for different value of the backlash gap
177
a) 3 b) 0.15
o
2 5.7
0.1 o
5.7
o 11.4
m [p.u]
ω [p.u]
1
0.05
0
o o
o 11.4 0
-1 0 0
0 0.01 0.02 0.03 0.04 0.05 0.06 0 0.01 0.02 0.03 0.04 0.05 0.06
t[s] t [s]
c)
d)
0.8 0.15
o o
5.7 0
0.6 o
0 0.1
m [p.u]
ω [p.u]
0.4 o
5.7
0.05
0.2 o o
11.4 11.4
0 0
0 0.01 0.02 0.03 0.04 0.05 0.06 0 0.01 0.02 0.03 0.04 0.05 0.06
t[s] t [s]
Fig. 7. Enlarged transients of the system state variable: electromagnetic torque (a), motor speed (b),
shaft torque (c) and load speed (d) for different value of the backlash gap
3
q4=29
2 q4=29 1
1 0.8
q4=0
me [p.u]
0 0.6
-1 0.4
ω1 [p.u]
-2 q4=0 0.2
-3 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t[s] t [s]
1.5 q4=29
q4=0 q4=29
1
1
0.8
q4=0
ms [p.u]
ω2 [p.u]
0.5 0.6
0.4
0
0.2
-0.5 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t[s] t [s]
Fig. 8. The transients of the system state variable: electromagnetic torque (a), motor speed (b), shaft
torque (c) and load speed (d) for different value of q4 and nominal value of the system speed
178
The drive system works properly and the system speeds follow the reference
value with the desired dynamics (results from the values of the matrices Q and R)
taking into account the system constraints. At the times t1=35ms and t2=100ms the
electromagnetic torque decreased rapidly in order to avoid the validation of the
shaft torque constrains. In the electromagnetic torque transients some overshoots
(over the limit) are visible resulting from the neglected dynamic of the motor
torque control loop. The application of the load torque causes a quickly eliminated
speed fall.
4. Conclusions
The paper is devoted to issues related to the application of the MPC control
structure to the two-mass system with backlash are presented. In order to damp
torsional vibrations an MPC controller based on the linear model is applied. From the
presented study the following conclusions can be formulated:
- The application of the MPC control structure to the drive system with mechanical
backlash allows to damp the torsional vibrations effectively.
- The proposed MPC strategy allows to close the mechanical gap of the backlash
softly.
- The bigger value q4, the softer the mechanical gap closes, however, at the same time
the dynamics of the drive are affected.
- Despite of a linear model of the two-mass system is used in the optimization
algorithm, the set limits of the system state variable are not validated.
- The drive system works correctly for small as well as for big values of the reference
speed.
- For the system with a relatively big value of q4 the variation of the backlash width
has an insignificant effect on the performance of the system.
The future work will be devoted to experimental validation of the proposed MPC
control structure. Also a system with a nonlinear model of the mechanical part of the
drive in the MPC algorithm will be considered.
References
[1] B. Armstrong-Hélouvry, P. Dupont, and C. Canudas de Wit., A Survey of Models,

Analysis Tools and Compensation Methods for the Control of Machines with
Friction. Automatica, vol. 30, no. 7, 1994, pp. 1083-1138.
[2] M. Nordina, P-O. Gutmanb, Controlling mechanical systems with backlash—a
survey, Automatica 38 pp.1633 – 1649, 2002.
[3] K. Szabat, M. Dybkowski, Tłumienie drgań skrętnych w układzie dwumasowym z
luzem, Przegląd Elektrotechniczny R 86 NR2/2010, str. 376–379.
[4] H. Mokhtari, F. Barati, A New Scheme for a Mechanical Load Position Control
Driven by a Permanent Magnet DC Motor and a Nonzero Backlash Gearbox, IEEE
ISIE 2006, July 9-12, 2006, Montreal, Quebec, Canada.
179
[5] Y Wu, K. Fujikawa, H. Kobayashi, A control method of speed control drive system
with backlash, 4th International Workshop on Advanced Motion Control, 1996. AMC
'96-MIE. Proceedings., Volume 2 1996.
[6] S. Thomsen, F. W. Fuchs, Speed Control of Torsional Drive Systems with Backlash,
13th European Conference on Power Electronics and Applications, 2009. EPE '09.
[7] L. T. Aguilar, Y. Orlov, J.C. Cadiou, and R. Merzouki, Nonlinear H∞-Output
Regulation of a Multi-stable Drive System including Backlash with a Single-Stability
Approximation, Proceedings of the 2007 American Control Conference Marriott
Marquis Hotel at Times Square New York City, USA, July 11-13, 2007.
[8] J. M. Maciejowski Predictive Control with Constraints, Prentice Hall, UK, 2002
[9] P. Tatjewski, Sterowanie zaawansowane obiektów przemysłowych, struktury i
algorytmy, Akademicka Oficyna Wydawnicza EXIT, 2002.
[10] M. T. Cychowski, K. Szabat, T. Orłowska-Kowalska, Constrained Model Predictive
Control of the Drive System with Mechanical Elasticity, IEEE Trans. on Industrial
Electronics, vol. 56, no. 6, pp 1963-1973, 2009.
[11] T. Geyer, G. A. Beccuti, G. Papafotiou, M. Morari, Model Predictive Direct Torque
Control of permanent magnet synchronous motors, Energy Conversion Congress and
Exposition (ECCE), 2010 pp. 199 – 206.
[12] P. J. Serkies, K. Szabat, Predykcyjny regulator położenia z warstwą rozmytych
ograniczeń dla napędu dwumasowego, Przegląd Elektrotechniczny R. 87 NR 8/2011
str. 202–205.
[13] M. Kvasnica, P. Grieder, M. Baotic, M. Morari, Multi-Parametric Toolbox (MPT),
HSCC (Hybrid Systems: Computation and Control), Lecture Notes in Computer
Science, vol. 2993, 2004, pp. 448-46.
[14] A. Bemporad, M. Morari, V. Dua, E.N. Pistikopoulos, The explicit linear quadratic
regulator for constrained systems, Automatica, vol. 38, no. 1, pp. 3–20, Jan. 2002.
[15] M. T. Cychowski, Robust Model Predictive Control, VDM Verlag, 2009.
180
Discrete models of the NLFSR generators

Janusz Walczak, Rafał Stępień
Silesian University of Technology
44-100 Gliwice, ul. Akademicka 10, e-mail: janusz.walczak@polsl.pl,
rafal.stepien @polsl.pl
In this article a class of pseudo random signal generators, build with static nonlinear
feedback block, is described. The feedback block function is modeled as a linear
combination of sums and products of the register’s taps. A discrete model of the generator
is proposed as a nonlinear discrete circuit. The simulation results from Matlab-Simulink
and Multisim are shown. Also an experimental results of the example generator on FPGA
implementation is shown.
1. Introduction
The pseudo random signal generators are widely used in technical applications,
including key stream generators in cryptography [2], telecommunication [6], [7]
and in many other applications [6]. The wide class of pseudorandom signal
generators is based on linear shift registers that works with feedback loop. This
class can be divided into four groups depending on the used feedback loop, Fig 1.
Fig. 1. The classification of the generators based on shift registers
The linear feedback generators with static or dynamic feedback loop were
described in the following papers [1], [2], [5], [7] and the analysis of non linear
feedback loop generators (NLFSR) can be found in [4], [10]. The following article
is a continuation of [10] that was devoted to the NLFSR analysis.
181
J. Walczak, R. Stępień / Discrete models of the NLFSR generators
2. The binary model of NLFSR generator
The considered circuit (Fig. 2) is constructed from a linear shift register (LSR)
and a nonlinear feedback block, that can be made from the combinational logic.
Fig. 2. The NLFSR generator
The formula (1) that is often used to describe non inertial circuits cannot be
used to describe a nonlinear feedback loop of the NLFSR generator (fig 2.). The
shift register, that is a part of NLFSR generator, works with binary signals. The
feedback loop also works with binary values. Every cell of shift register contains
only one of two values {0;1}. For variables x1…xn that can only contain binary
values and for any values of coefficients aik, the nonlinear function (1) becomes a
linear function. This is caused by the exponent part of formula (1). For binary
values the exponent value does not matters because the formula (1) will return as
result also binary value.
M N
y = ∑∑ aik ( xi ) k , aik ∈ R, xi ∈ GF (2). (1)
i =1 k =1
In circuit shown on figure 2, the y value (output of the feedback loop) needs to
have a value from GF(2) finite field. Coefficients values from formula (1) and the
sum and product operators needs to belong to GF(2) finite field.
3. The time discrete model of the NLFSR
The proposed description of the non linear feedback loop (fig. 2) can be realized
with the following formula:
k n
y = ∑∑ mij x j y ∈ GF (2), x j ( j = 1,2,..., n) ∈ GF (2), (2)
i =1 j =1
where: y - binary signal from the feedback loop, xj - binary signals from selected
register’s taps, mij -zero-one elements of the M matrix, n – number of register cells,
k – number of sum components.
182
It follows that formula (2) has k (k≤n) components. Every one corresponds to
the product of j components (j≤n). Some of these components can be equal to zero.
That can be interpreted as a signal from register’s tap was not connected to the
feedback loop block. The M matrix defines which signals from linear shift register
are connected to the feedback loop. The following example shows this idea.
Example 1
The NLFSR generator contains a shift register with n=4 cells. The M matrix is
defined as follows:
⎡1 0 1 1 ⎤
M = ⎢⎢1 1 0 0⎥⎥ (3)
⎢⎣0 0 0 1⎥⎦ .
Determined from (2) output signal y of the feedback loop block is described
with following equation:
y = x1 ⊗ x3 ⊗ x 4 ⊕ x1 ⊗ x 2 ⊕ x 4 , (4)
where: ⊕ - the sum symbol in GF(2) finite field, ⊗ - the product symbol in GF(2)
finite field.
The scheme of the considered generator is show on Fig 3.
Fig. 3. Schematic of NLFSR generator described by matrix (3)
From the above considerations it follows that the feedback loop function (2) is a
sum of products in GF(2) finite field. This function can be considered as a modulo
2 sum (XOR) of component products from formula (2). The modulo 2 sum can be
written as a discrete form shown in the following formula:
A ⊕ B = A + B − 2 AB (5)
where symbols of sum and product that appears on the right side of formula (5)
needs to be considered as a classical operations in a set of real numbers. Formula
(5) can be used to transform the formula (2) to discrete time circuits and systems
relations that can describes the NLFSR generators. In order to perform that the
183
formula (2) needs to be decompose into a form of sums and products and then all
obtained components needs to be added as it is shown in formula (5). The example
2 shows the described idea.
Example 2
The M matrix that describes the feedback loop block is given as follows:
⎡1 0 0⎤
M = ⎢⎢0 1 0⎥⎥ (6)
⎢⎣1 0 1⎥⎦ .
Based on the formula (2) (see also example 1) the feedback loop block model
for signals from GF(2) finite field is described by the following formula:
y = x1 ⊕ x2 ⊕ ( x1 ⊗ x3 ) (7)
The formula described with (7) is made from three components: x1,x2 and
(x1 ⊗ x3). Every component needs to be added to the result of previous sum. The
algorithm of transformation is given as follows:
Step 1: sum of x1 ⊕ x 2 :
x1 ⊕ x 2 ⇒ x[n − 1] + x[n − 2] − 2 x[n − 1]x[n − 2] (8)
Step 2: sum of ( x1 ⊕ x 2 ) ⊕ ( x1 ⊗ x3 ) :
( x1 ⊕ x 2 ) ⊕ x1 x3 ⇒ x[n − 1] + x[n − 2] − 2 x[ n − 1]x[n − 2] + x[n − 1] x[n − 3]
(9)
− 2( x[ n − 1] + x[n − 2] − 2 x[n − 1]x[n − 2]) x[n − 1]x[n − 3] = y[n]
Figure 4 shows the structure of a discrete circuit that corresponds to the
difference equation (9).
Fig. 4. The feedback function structure that is obtained from equation (9)
184
4. The simulations results
Simulations of the exemplary NLFSR generator where conducted in the Matlab

Simulink environment and in the Multisim 9.0. The M matrix that describes the
simulated NLFSR generator is given as follows (10):
⎡1 1 1 0 ⎤
M =⎢ ⎥ (10)
⎣0 0 0 1 ⎦
If matrix elements are substituted into formula (2) the feedback loop formula is
describes as follows (11):
y = x1 ⊗ x2 ⊗ x3 ⊕ x4 (11)
The multiplication and sum operations in above formula should be interpreted
as operations in GF(2) finite field. Transforming formula (11) similarly as it was
shown in example 2 leads to the discrete form of (11). The discrete form of formula
(11) is defined as follows (12):
y[ n] = x[ n − 1]x[n − 2] x[n − 3] + x[n − 4] − 2 x[n − 1]x[ n − 2]x[n − 3]x[ n − 4] (12)
The feedback loop block of the considered generator, made from discrete systems
block, is shown on Fig 5.
Fig. 5. The structure of the feedback loop
The output sequence y[n] from the generator is shown on Fig 7. Simulated in
Matlab Simulink circuit is shown on Fig. 6.
185
The shift register is made from unit delay blocks z-1. The default state of the
shift register is set to 1111BIN. The sampling rate was set to 0,1s. The output signal
taken from the forth register’s tap is shown on Fig. 7.
Fig. 6. The simulated NLFSR generator
Fig. 7. The output signal of simulated NLFSR generator
Based on Fig. 7 and analytical calculations the obtained output sequence is as

follows: 1111011110111... This sequence is periodic with period 5. The next
simulation was conducted in the Multisim 9.0 environment. The simulated circuit
is shown on Fig. 8.
Fig. 8. Schematic of the simulated NLFSR generator
186
The simulated circuit is build from a shift register CD4015 supplied from 5V.
Logic gates U3, U4, U5 are used as a feedback block. The signal generator XFG1
is used as a clock source. The output signal is taken from the last, forth tap of the
shift register. The output signal and clocking signal are shown on figure 9. These
signals are identical as obtained from previous circuit (Fig. 7). This proves that
change from electronic circuit into discrete model is correct.
Fig. 9. The output signal of considered generator
5. The experimental results
The physical generator was constructed as a FPGA implementation. Used

FPGA integrated circuit XC3S200 [12] is produced by Xilinx and belongs to the
Spartan III family. This FPGA chip is mounted on a development board, ZL9PLD,
that is produced by Kamami [13], [14]. JTAG is used as a programming interface
and all schematics and codes are generated by the ISE WebPack environment.
The schematic of NLFSR generator is show on Fig. 10.
Fig. 10. The schematic of NLFSR generator implemented in FPGA

187
The initial state of the shift register is set exactly as it was set in simulation
results. The output signal obtained as a signal from buffered line Q3 and the
clocking signal are shown on Fig. 11.
Fig. 11. The output signal of NLFSR generator
The obtained waveforms are exactly the same as in previous cases. This proves
that carried out considerations and obtained formulas are correct. The output
sequence 1111011110111... is periodic with period 5.
6. Summary
In this article a method of describing the non linear feedback loop block of the
NLFSR generators is proposed. An effective method of transformation from the
binary model of the feedback loop to the model of time-discrete systems is given.
Obtained results are supported with examples. The simulation results of exemplary
NLFSR generators in Matlab and Multisim are given. More over a physical model
of the NLFSR was implemented on the Spartan III FPGA and some results of its
implementation were shown. A full consent was obtained between theoretical and
physical models. That proves that proposed method is correct.
REFERENCES
[1] Chen L., Gong G.: Communication Systems Security, Appendix A, Draft, 2008.
[2] Golomb S. W.: Shift Register Sequences. Laguna Hills, C A Aegean. Park Press,
1982.
[3] Schneier B.: Kryptografia dla praktyków, Vol. 2, WNT, Warszawa 2002.
188
[4] Dubrova E.: How to Speed-Up Your NLFSR-Based Stream Cipher, Royal Institute of
Technology (KTH), Stockholm, Sweden, 2009.
[5] Walczak J., Stępień R.: Modeling of the pseudo random signal generators using
digital filters. Proceedings of XXXIII Conference IC-SPETO May, 2010,pp. 85-86.
[6] Kotulski Z.: Generatory liczb losowych: algorytmy, testowanie, zastosowania,
Matematyka Stosowana 2,2001, pp.1-6.
[7] Mutagi R.N.: Pseudo noise sequences for engineers, Electronics & Communication
Engineering Journal,Vol.8 Issue 2, April 1996, pp.79-87.
[8] J. Massey, “Shift-register synthesis and bch decoding,” IEEE Transactions on
Information Theory, vol. 15, pp. 122–127, 1969.
[9] Walczak J., Stępień R.: Discrete Model of the NLFSR Generators, Proceedings of
XVI Conference ZKwE April, 2011,pp. 35-36.
[10] Walczak J., Stępień R.: Shift Registers with Dynamic Feedback Loop, Proceedings of
XXXIV Conference IC-SPETO May, 2011,pp. 125-126.
[11] Alfke P.: Efficient Shift Registers, LFSR Counters, and Long Pseudo-Random
Sequence Generators, Xilinx application note, July 7,1996, Vol 1.1.
[12] Xilinx corporation.: Spartan III family data scheet,
http://www.xilinx.com/support/documentation/data_sheets/ds099.pdf
[13] Kamami development tools, dipPLD module, www.kamami.pl/dl/zl10pld.pdf
[14] Kamami development tools, mother board for dipPLD modules,
www.kamami.com/dl/zl9pld_en.pdf
Sub-phase power electronics inverters

Michał Gwóźdź
60-965 Poznań, ul. Piotrowo 3A, e-mail: Michal.Gwozdz@put.poznan.pl
Dynamic changes of parameters of energy sources and receivers are reasons that
decrease exactitude of output signals towards reference signals. To improve these
parameters more advanced solutions of power electronics systems are necessary. Ones of
possible solutions are a sub-phase inverters. The output quantity regulator there is based on
Generalized Sampling Expansion, being crucial extension of Whittaker-Kotelnikov-
Shannon sampling theorem. Wide-band power electronics current source with regulator
based on that extension is analyzed further.
1. Introduction
Dynamic changes of parameters of electrical energy receivers as well as non-

linearity of power inverters are reasons that decrease exactitude of static and
dynamic output signals towards reference signals. To improve these parameters
more advanced solutions of power electronics systems are necessary. One of these
is power electronics wide-band voltage controlled current source [3, 4, 6, 7]. This
special inverter may found many applications as: execution block of compensators
of currents deformation in power grids [3, 6], systems with unified power flow
controllers (UPFC), power electronics reference generators [7], modern electrical
drives [16] and equipment for medicine [4, 17].
Proposed power electronics sub-phase inverters works with negative feedback
and contains output current regulator based on Generalized Sampling Expansion
(GSE) [14]. On base of such extension of Whittaker-Kotelnikov-Shannon (WKS)
sampling theorem, small-signal and simulation model of sub-phase current source
is proposed and analyzed further.
2. Generalized sampling expansion
Assume signal x(t ) ∈ L2 (ℜ) , the space of square-integrable functions,

where ℜ - real numbers domain, and its Fourier transform X ( jω ) exists. Since
the function x(t ) is band-limited, the support of X ( jω ) is within the interval
[− ωMAX , ωMAX ] . The function X ( jω ) lies within the space
L2 [− ωMAX , ωMAX ] and can, therefore, be written in terms of a Fourier series.
190
M. Gwóźdź / Sub-phase power electronics inverters
Taking into consideration WKS sampling theorem while sampling of x(t ) is

uniform and ideal, it means with utilization of Dirac series ∑ δ(t − nTS ) , where
n
π
TS ≤ is the sampling period, the x(t ) can be reconstructed using inverse
ωMAX
Fourier transform formula:
∞ ωMAX
1 jω (t − nTS )
x (t ) = ∑ x(nTS ) ∫ e dω . (1)
2ωMAX n = −∞ −ωMAX
A one of major extension of Whittaker-Kotelnikov-Shannon (WKS) sampling

theorem was formulated by Papoulis [14] which unifies a broad class of extensions.
The general setting is that a signal x(t ) is processed by a linear Multi-Dimensional
Sampling System (MSS) – Fig. 1.
Fig. 1. Multi-Dimensional sampling system (MSS)
Suppose now that x(t ) is a common input to M sampling systems, being sub-
1
systems of MSS. Each sub-system samples at a rate of times of the Nyquist
M
rate. Assuming individual delays τ i = iTS , the y (t ) can be reconstructed using
inverse Fourier transform formula:
191
M −1 ⎧ ⎫
⎛ i ⎞
1 ∞ ⎪ ⎡⎛ i ⎞ ⎤ ωMAX jω ⎜⎝ t − ( n + M ) MTS ⎟⎠ ⎪
y (t ) = ∑ ∑ ⎨ x ⎢⎜ n + ⎟ MTS ⎥ ∫ e dω ⎬ . (2)
2ωMAX n = −∞ i = 0 ⎪ ⎣⎝ M⎠ ⎦ −ωMAX ⎪⎭
⎩
The overall sampling rate still satisfies the Nyquist criterion. Finally, signal
y (t ) = x(t ) can be retrieved from extrapolator E on base a sum of x(ti ) samples.
The multi-dimensional sampling system, being a consequence of GSE, lets to
1
reduce the required sampling frequency to , comparing to 1-dimensional
M
sampling system (WKS sampling theorem).
3. Idea of sub-phase power electronics inverters
A general scheme of sub-phase inverter, for instance controlled current source,

is shown on Fig. 2. The sub-phase inverters are extension of idea of interleaved
inverters [10, 15] – toward more wide-band while final limitation of frequency is a
pulse modulation carrier frequency.
Fig. 2. General block scheme of power electronics wide-band controlled current source with GSE
based current regulator
The current source works with negative feedback and consists of two main
modules: R – regulator, PEM – power electronics module. CT is a current
transducer.
The regulator consists of M th order MSS and gain block ( k0 ) while the power
electronics module consists of M PWM based inverters and M output inductors
with common current summing node (N). The adder A produces error signal:
u ERR (t ) = u REF (t ) − u B (t ) . The output current iL (t ) is proportional to input
(reference) voltage u REF (t ) .
Since the sampled function does not carry any energy, additional module
converting sampled function into continuous-time domain, is necessary. Most wide
192
utilized in theory and practice converting module is 0-order extrapolator in form of

Sample-And-Hold Amplifier (SHA) [1]. Thus, for implementation of GSE in
regulator (R) crucial modification of MSS, according to Fig. 1, is necessary. In
each of sampling sub-system 0-order extrapolator (SHA) is placed now. All SHAs
T
sample at TS rate with individual τ i = i S time shift. Order of MSS is M .
M
Sub-phase inverters carry a lot of advantageous. Most common are: crucial
increasing the regulator gain k0 while system stability is behaved and significant
minimization of ripples in output signal caused by pulse (e.g. PWM) modulation.
4. Stability analysis of sub-phase power electronics current source
Assuming now hSHA (t ) is a pulse response and taking into consideration scale
1
factor [9], a transfer function of SHA is given by following equation:
TS
TS −ε TS
1 1 1 − e − sTS − jω
2 Sa ⎛
T ⎞ . (3)
H SHA (s ) = ℑ[hSHA (t )] = lim ∫ e − st dt = s = jω =e ⎜ω S⎟
TS ε →0 0 TS s ⎝ 2⎠
The integral in formula (3) is improper because of TS is excluded from the
sampling interval 〈 0, TS ) .
In fact, relationship between X ( jω ) and Y ( jω ) apart static (time invariant)
∞
∑ x(nTS )e
jωnTS
component (3), contains dynamic (time variant) component yet,
n = −∞
so formal transfer function of SHA doesn’t exist [2, 11, 12]. The dynamic
component is related to an aliasing effects also, by which the high frequency poles
are folded back into lower frequencies. Although respecting the static part only of
equivalent transfer function (ETF) of SHA gives, in most cases of system stability
analysis [3, 4, 7], satisfying results the crucial knowledge is that the aliasing
mechanism can cause loss of stabilization at critical frequencies [2, 5, 8, 11].
The equation (3) is valid assuming the SHA works in open loop, that is,
feedback has not an effect. However while feedback is present equation (3) has to
be modified toward ETF system with feedback. The modification follows a fact
that sampling system is a modulator producing an infinite number of aliases of
modulator input signal spectrum in basic band. A modulator input signal is a
difference of input (reference) and output signals [2, 5, 8, 11].
General case of sampling functions, in form of Dirac series while M =2, is
shown on Fig. 3. Sampling is uniform one and both of series are shifted each other
193
1 1
by 2aTS , where shift factor a ∈ − , . In particular cases: a =0 sampling is 1st
4 4
order, a = ± 1/4 sampling is 2nd order and orthogonal. Formally, sampling
functions are given by following equations:
∞
[ (
si (t ) = ∑ δ t − n − ( −1)i a TS ) ] (4)
n = −∞
where: i = 0 or i = 1 .
Fig. 3. Sampling functions for achieving 2nd order MSS
In order of stability analysis system should be more specified, that is, choice of
output circuit configuration is necessary. Thus, further consideration concerns the
small-signal (linear) model of controlled current source in form shown on Fig. 4.
The general assumption is that one works with negative feedback. Regulator (R) of
output current iL (t ) is 2nd order MSS ( M =2) and consists of two SHAs and gain
block k0 .
Fig. 4. Linear model of controlled current source with 2nd order MSS based regulator
194
Proposed small-signal model of such input-output system is one-dimensional,

time-invariant and can be completely described in terms of continuous-time system
I L ( jω ) K ( jω )
with transfer function: G ( jω ) = = . Stability analysis takes
U REF ( jω ) 1 + K ( jω )
advantage of Nyquist criterion [2,12,13]. Hence, characteristic equation of model:
R ( jω ) = 1 + K ( jω ) = 0 , while model transfer function K ( jω ) :
∞ ⎧⎪ − j(ω − nωS )TS ⎫⎪
K ( jω ) = 2rCT k0 ∑ ⎨e 2 cos[(ω − nω )aT ]Sa ⎡(ω − nω ) TS ⎤ 1
⎢⎣ ⎬ . (5)
2 ⎥⎦ j(ω − nωS )L ⎪
S S S
n = −∞ ⎪
⎩ ⎭
Effective regulator gain ke = 2k0 is increased by 2, comparing to gain of 1st

order MSS [9]. The function (5) possesses infinity number of singular points
ω = nωS .
The Nyquist diagram, on base of equation (5), shown on Fig. 5, concerns
following hypothetical conditions: rCT =1 V/A, L =5 mH, TS =0,1 ms and k0 =60
V/V (effective gain: ke =120 V/V). Since (5) is periodic function in ω with period
2π 2π ⎞
TS , the (P,Q) diagram is also periodic in ω intervals: n , (n + 1) ⎟⎟ .
TS TS ⎠
For hypothetical particular gain k0 =60 V/V ( ke =120 V/V) and a =0,10; 0,15;
0,20; 0,25 the (P,Q) phase curve encircles the Nyquist point ( − 1 + j0 ) on the right
side what means system is stable. Characteristic effect of 2nd order sampling is
decreasing gain of MSS while frequencies are closed to characteristic point
π
(Nyquist frequency) ω = . For other ones the gain is relatively high. Gain
TS
minimizing effect is maximal while sampling is orthogonal.
In terms of Nyquist criterion maximal regulator gain up to that system is stable
defines equation (6) while ω = −π :
rCT k0 2 ∞ ⎡π ⎤ 1
K ( jω ) ω =ω π = 2 ∑ cos ⎢ (1 − 2n )⎥ < 1. (6)
2
L π n = −∞ ⎣4 ⎦ (1 − 2n )2
By solving the equation (6), the maximal value of regulator gain k0, MAX can
be determined as follows:
L
k0, MAX < 2 , (7)
rCTTS
195
ke, MAX = 2k0, MAX . (8)
For given hypothetical system parameters ( rCT =1 V/A, L =5 mH, TS =0,1

ms): k0, MAX <100 , thereby: ke, MAX <200.
R ( jω ) of small-signal model of current source

Fig. 5. Nyquist diagram for characteristic equation
with MSS and M =2; k0 =60 V/V
5. Simulation model of current source
For acknowledgment of mathematical and small-signal model considerations,

simulation models of controlled current source with 2nd order MSS in ORCAD
environment has been proposed – Fig. 6. The PEM block there consists of two
pairs of inverters. Thanks to that, implementation of one-pole PWM strategy has
been possible.
196
Fig. 6. Block scheme of power electronics current source with 2nd order MSS
and one-pole PWM implemented
Selected investigation results of simulation model are shown on Fig. 7. Values of

system quantities are as follows: k0 =50 V/V, r =1 V/A, L X =5 mH, RL =0 Ω ,
TS =0,1 ms. The PWM is 2-sided and carrier frequency f PWM ,C =5 kHz.
a)
b)
Fig. 7. Selected waveforms in simulation model of current source with 2nd order MSS and one-pole
PWM implemented: a) u REF (t ) is rectangular and f REF =50 Hz, b) u REF (t ) is sinusoidal and
f REF =200 Hz
197
Simulation model investigation results point at fact that 2nd order MSS lets
crucially increase effective system gain, comparing to regulator on base of WKS
(or MSS and M =1). Another advantage of MSS based regulator is better stop-
band properties of this at frequencies closed to Nyquist frequency, what lets more
effective minimize aliasing effects.
Investigation results show that output current parameters are much improved.
Commonly, referencing the output current waveform within reference signal is
much better, comparing to conventional solution of such inverters. Output quantity
referencing error is understood as u ERR (t ) (Fig. 6) and can be easily measured.
6. Conclusions
Presented idea of sub-phase inverters with regulator on base of GSE lets

maximize significantly regulator gain and achieve more advantageous system
transfer function. Thanks of that affecting of aliasing effects on system work is
reduced. The ripples in output signal, caused by pulse (e.g. PWM) modulation, are
significantly minimized also. As a result, referencing the output quantity waveform
within input (reference) signal is much better, comparing to classical solution of
inverters. It seems necessary to continue investigations of such type of power
electronics equipments. The reason is their advantages and possibilities of direct
implementation in modern power electronics systems with acceptable today
complication of these.
References
[1] Dąbrowski A. (red.): Przetwarzanie sygnałów przy użyciu procesorów sygnałowych.

Wyd. III, Wydawnictwo Politechniki Poznańskiej, Poznań, 2000.
[2] Francis B., Georgiou T.: Stability theory for linear time-invariant plant with periodic
digital controllers. IEEE transactions on Automatic Control, 33 (9), 1988, pp. 820-
832.
[3] Gwóźdź M., Porada R.: Compensate for Loading Effects on Power Lines with a DSP-
Controlled Active Shunt Filter. Analog Devices, Inc. Analog Dialogue, 1999, Vol.
33, No 9.
[4] Gwóźdź M., Porada R.: Liniowe źródło prądu w generatorze przestrzennego pola
magnetycznego. Mat. XXVIII Międzynarodowej Konferencji z Podstaw
Elektrotechniki i Teorii Obwodów, Gliwice-Wisła, IC-SPETO2005, 26-29 maj 2005,
t. 2, ss. 329-332.
[5] Gwóźdź M.: Impact of Aliasing Effect on Work of Wide Band Power Electronics
Current Source. XIX Symposium Electromagnetic Phenomena in Nonlinear Circuits,
Maribor, Słowenia, 28-30.06.2006, ss. 147-148.
[6] Gwóźdź M.: Efektywność poprawy jakości prądu sieci za pomocą
energoelektronicznego kompensatora aktywnego. Przegląd Elektrotechniczny, nr 7-
8/2006, ss. 65-68, PL ISSN 0033-2097.
198
[7] Gwóźdź M., Porada R.: Wide-Band Power Electronics Current Source for Reference
Purposes. Proceedings of 12th European Conference on Power Electronics and
Applications, EPE2007, Aalborg, Denmark, 2-5 September 2007, full paper on CD-
ROM.
[8] Gwóźdź M.: Analiza pracy szerokopasmowych przekształtników energoelektronicznych.
Przegląd Elektrotechniczny nr 12/2010, ss. 216-221, PL ISSN 0033-2097.
[9] Gwóźdź M.: Stability of Discrete-Time Systems on Base of Generalized Sampling
Expansion. Mat. XXXIV Konferencji IC-SPETO 2011, Gliwice-Ustoń, 18-
21.05.2011, pp. 71-72.
[10] Huth S.: DC/DC converters in Parallel Operation with digital load distribution
control. Proceedings of the IEEE International Symposium on Industrial Electronics,
1996, 17-20 June 1996 pp. 808-813 vol.2.10.
[11] Kalman R., Ho B., Narendra K.: Controllability of linear dynamical systems.
Contribiutions of Differential Equations, 1, 1963.
[12] Kaczorek T.: Teoria Sterowania i Systemów. Wyd. Naukowe PWN, W-wa , 1999.
[13] Kaczorek T. (red.): Podstawy teorii sterowania. WNT, Wyd. II, W-wa 2005, 2006,
ISBN 83-204-3251-0.
[14] Papoulis A.: Generalized Sampling Expansion. IEEE Transactions on Circuits and
Systems, v.24, Nov. 1977.
[15] Saggini S., Ghioni M., Geraci A.: An innovative digital control architecture for low-
voltage, high-current DC-DC converters with tight voltage regulation. IEEE
Transactions on Power Electronics, Volume: 19, Issue: 1, Jan. 2004, pp. 210 – 218.
[16] Seliga R., Koczara W.: High Quality Sinusoidal Voltage Inverter for Variable Speed
AC Drive Systems. EPE-PEMC2002, Cavtat & Dubrovnik, 9-11 September 2002.
[17] Sieroń A.: Zastosowanie pól magnetycznych w medycynie. α-medica Press, Bielsko
Biała, 2002.
199
Assessment of the spatio-temporal vector median filtering

algorithms using colour video quality metrics
Krzysztof Okarma
West Pomerania University of Technology
71–126 Szczecin, ul. 26. Kwietnia 10, e-mail: okarma@zut.edu.pl
In this paper the application of the combined video quality assessment method as well
as some other recently developed objective metrics for the analysis of the results of the
nonlinear colour video filtering is discussed. The spatio-temporal versions of colour image
filtering methods, including the Vector Median Filter, can be obtained using frame-by-
frame approach but the proper choice of the spatio-temporal kernel weights and the colour
space used during filtration should be based on a reliable video quality assessment. In some
earlier papers the combined video quality assessment method has been proposed, which has
a highly linear correlation with subjective quality scores and can be extended into the
colour version. As the illustration of the problem, some results of the colour video
denoising using the spatio-temporal VMF, also in a weighted version, together with the
quality assessment results have been presented in the paper.
1. Introduction
Colour image processing is one of the most dynamically developing area of

image processing and analysis. One of the most important reasons is the hardware
progress, not only related to the colour image acquisition equipment, but also
specialized graphical processing units allowing fast parallel processing of large
amount of image data. Such progress requires some advances in colour image
processing algorithms, which are not always a straightforward extension of well-
known techniques used for greyscale imaging. A good example is nonlinear
filtration of colour images, in particular the vector approach to median filtration.
Another relevant area of research is reliable quality assessment of colour images
as well as colour video sequences. Most of quality metrics, even those recently
proposed, do not use colour data at all. Since colour dissimilarity is an important
element of image quality, the minimal requirement should be the calculation of an
image metric for three RGB channels or additionally for the chrominance,
depending on the colour model.
2. Nonlinear filtering of colour images
Classical approach to greyscale median filtering is used for the elimination of

an impulse noise. Analysing the local neighbourhood of the current pixel, e.g.
using the 3×3 pixels sliding window (mask), the most typical pixel from the mask
200
K. Okarma / Assessment of the spatio-temporal vector median filtering algorithms…
should be chosen as the output. For this purpose all the pixels within the current
mask are sorted and the central (median) element is chosen. Such approach is
characterised by an important advantage - the resulting image contains only the
pixels’ values which have been present in the original one.
In order to prevent such feature various versions of colour pixel’s ordering can
be used. A common idea is the choice of the most typical pixel from those which
are present within the current mask. For this reason no additional colours are
present in the resulting image.
One of possible approaches is the usage of the conditional ordering [3, 10],
where the pixels are sorted according to a single, arbitrary chosen, colour channel.
Only the pixels having the same value of that channel are additionally sorted
according to the second one. Such approach can be used e.g. using the HSV colour
model, using the value and saturation as first two ordering criterions, respectively.
Another group of colour image filtering methods is reduced (aggregate)
ordering, where the calculation of the additional values based on all channels is
necessary. Such values are used for ordering the pixels and the most typical one is
chosen as the output.
Another well-known algorithm for nonlinear filtration of colour images is
Vector Median Filter (VMF) [1], where the aggregated distance from each of the
pixels to the remaining ones within the current mask is calculated in the specified
colour space. As the result the pixel with the minimum aggregated distance
(usually calculated using the L2-norm) is chosen as the most similar to the other
ones. Instead of using the RGB colour space some other ones such as CIE XYZ or
CIE L*a*b* can also be used [2].
3. Spatio-temporal video filtering
Taking into consideration a specific character of the video data a relatively large
amount of noise can be observed, which is present for specified pixels in single
video frames. Such noise can be removed using independent filtering of each
frame, e.g. using Vector Median Filter, but for a large amount of an impulse noise
the obtained results are often unsatisfactory. A typical approach used for the noise
reduction from image sequences is the application of simple low-pass Arithmetic
Mean Filter (AMF), which unfortunately causes easily noticeable blurring and loss
of sharpness. Some other recently proposed ideas are based on sophisticated spatio-
temporal filters based on motion estimation, which can be relatively easily applied
for MPEG compressed sequences.
Nevertheless, the application of any spatio-temporal filter requires the proper
choice of a spatio-temporal kernel (mask) containing the analysed pixels and their
weights (used for the weighted filters). The choice of the kernel size and weights
should be performed in a way leading to the maximum possible quality of the
resulting image. Unfortunately, the optimisation is often performed using
201
traditional MSE-based image quality metrics, which are poorly correlated with
subjective evaluations. For this reason some modern image quality assessment
methods, such as Structural Similarity (SSIM) [11] should be used. Taking into
account the fact, that the kernel size and weights can be defined for each RGB
channel (or even using another colour space) separately, the colour image quality
index should be used (or in the last resort three quality metrics’ values for each
channel).
Recently some other modern approaches to objective image and video quality
assessment have been reported. Probably the most interesting ones are Multi-Scale
SSIM [12], Visual Information Fidelity (VIF) [9] and R-SVD metric [4] based on
the Singular Value Decomposition.
The MS-SSIM metric can be computed as the weighted product of three
components for M (typically five) scales according to the following formula:
M
MS − SSIM ( x, y ) = [l ( x, y )] ⋅ ∏ [c( x, y )] ⋅ [s ( x, y )] ,
αM βj γj
(1)
j =1
where the luminance (l), contrast (c) and structural (s) distortions are defined
similarly as for the single-scale SSIM:
α β γ
⎛ 2 x y + C1 ⎞ ⎛ 2σ yσ y + C2 ⎞ ⎛ σ xy + C3 ⎞
SSIM ( x, y ) = ⎜⎜ 2 ⎟⎟ ⋅ ⎜ 2 ⎟ ⋅⎜ ⎟ = [l ( x, y )]α ⋅ [c( x, y )]β ⋅ [s( x, y )]γ .
x + y 2
+ C ⎜ σ + σ 2
+ C ⎟ ⎜ σ σ + C ⎟
⎝ 1⎠ ⎝ x y 2 ⎠ ⎝ y y 3 ⎠
(2)
For both metrics the weighting coefficients (α,β,γ) are typically set to 1 for
simplicity.
The VIF metric is defined as:
S Nj
∑∑ I (ci , j ; fi , j )
j = 0 i =1
VIF = S Nj
. (3)
∑∑ I (ci , j ; ei , j )
j = 0 i =1
where I(a;b) denotes the mutual information between a and b. The denominator
and numerator denote the information extracted by human vision from the
reference and distorted images. Assuming that c is a block vector at a given
location in the reference image, e denotes the perception of block c by a human
viewer in the presence of an additive noise and f is the perception of distorted block
c [9].
Another interesting approach to image quality assessment is the usage of the
Singular Value Decomposition e.g. utilising the recently proposed R-SVD index
calculated as:
202
Nk
∑ (di − 1)
2
i =1
R-SVD = Nk
, (4)
∑ (di + 1)
2
i =1
where di denotes the singular values of the referee matrix R’=U’·Λ·VT, with
identity matrix Λ. The product of U·S·VT is obtained as the result of the SVD
decomposition of the matrix representing the original image A, and U’·S’·V’T is
obtained for the distorted image A’ respectively [4].
In one of the author’s earlier papers [5] a combined image quality metric
utilizing those metrics has been proposed and verified experimentally as highly
linear correlated with subjective scores. Such metric can be expressed as:
CQM = (MS − SSIM ) ⋅ (VIF ) ⋅ (R − SVD )
a b c
(5)
assuming the nearly optimal simple combination of a = 7, b = 0.3 and c = − 0.15 as
the result of the optimisation in terms of the linear correlation with subjective
quality scores using the TID2008 database [8], being the largest publicly available
image database containing the subjective quality scores for 1700 colour images
with many types of distortions.
Such approach has been further extended for the video quality assessment [6],
as well as for the colour image quality assessment [7]. Nevertheless, in this paper
an independent calculation of the metric for the RGB channels is used for
convenient comparison with results obtained for the other quality metrics.
Some other interesting approaches, presented recently, are the application of the
Riesz transform leading to the RFSIM metric [13], as well as the phase congruency
and gradient maps resulting in the Feature Similarity (FSIM) metric [14]. Both of
those metrics have good properties in the aspect of correlation with subjective
evaluations, so they have also been used in the paper.
The RFSIM metric is defined as:
∑∑ di (x, y ) ⋅ M (x, y )
5
RFSIM = ∏ x y , (6)
i =1 ∑∑ M (x, y )
x y
where M stands for the binary mask obtained as the result of the morphological
dilation of the image being the result of the Canny filter applied to the input image.
The mask is multiplied pixel by pixel with the feature similarity calculated for each
pixel with (x,y) coordinates as:
2 ⋅ f i ( x, y ) ⋅ g i ( x, y ) + c
d i ( x, y ) = , i = 1..5, (7)
fi 2 ( x, y ) + gi2 ( x, y ) + c
where fi and gi are the five consecutive coefficients of the Riesz transform of the
first and second order for the reference and distorted images.
The FSIM metric utilises the gradient map and the phase congruency instead of
Riesz transform, so the overall metric can be expressed as:
203
∑∑ S (x, y ) ⋅ PCm (x, y )

x y
FSIM = , (8)
∑∑ PCm (x, y )
x y
where PCm(x,y)=max(PC1(x,y), PC2(x,y)), so the higher from the two values of the
phase congruency for the reference and distorted image is used instead of the mask
value. The local similarity S can be determined as:
α β
⎛ 2 ⋅ PC1 ( x, y ) ⋅ PC2 ( x, y ) + TPC ⎞ ⎛ 2 ⋅ G1 ( x, y ) ⋅ G2 ( x, y ) + TG ⎞
S ( x, y ) = ⎜⎜ 2 2
⎟ ⋅⎜ 2
⎟ ⎜ G ( x, y ) + G 2 ( x , y ) + T
⎟
⎟ (9)
⎝ PC1 ( x, y ) + PC2 ( x, y ) + TPC ⎠ ⎝ 1 2 G ⎠
assuming the presence of two small stability constants TPC and TG and α =β =1 for
simplicity.
The filtering results obtained in the experiments have been assessed using most
state-of-the-art full-reference image quality assessment methods mentioned above,
including the combined metric, as well as the PSNR for comparison purposes.
(a)
(b)
(c)
Fig. 1. Example video frames with their zoomed fragments (a – original, b – 30% impulse noise after
VMF, c – 30% impulse noise after unweighted spatio-temporal VMF)
204
Fig. 2. Eight example consecutive frames of the reference test video
4. Experimental results
In order to illustrate the problem, some experiments have been performed using
the 27-pixels spatio-temporal Vector Median Filter (3×3 pixels mask for 3 video
frames) for the elimination of an RGB impulse noise from the well-known test
sequence “Crowd run” characterised by many changes in a lower part of each
frame. Exemplary video frames from the video sequence “Crowd run” used during
205
experiments are presented (luminance channel only) in Fig. 2. and some of the
results obtained for one of the noised and filtered frames are presented in Fig. 1
with zoomed critical fragments of images. After the filtration, the image quality
metrics’ values for each RGB channel have been calculated and compared to those
obtained using standard single-frame Vector Median Filter. The results presented
as the average quality metrics’ values for eight consecutive frames are presented in
Figs. 3-10.
Fig. 3. PSNR values for the Vector Median Filter (VMF), spatio-temporal Vector Median Filter
(STVMF) and weighted spatio-temporal one (STVMF) averaged for all frames
Fig. 4. Structural Similarity (SSIM) values for the Vector Median Filter (VMF), spatio-temporal
Vector Median Filter (STVMF) and weighted spatio-temporal one (STVMF) averaged for all frames
206
Fig. 5. Multi-Scale Structural Similarity (MS-SSIM) values for the Vector Median Filter (VMF),
spatio-temporal Vector Median Filter (STVMF) and weighted spatio-temporal one (STVMF)
averaged for all frames
Fig. 6. VIF values for the Vector Median Filter (VMF), spatio-temporal Vector Median Filter
Fig. 7. R-SVD values for the Vector Median Filter (VMF), spatio-temporal Vector Median Filter
(STVMF) and weighted spatio-temporal one (STVMF) averaged for all frames (smaller values
indicate better quality)
207
Fig. 8. FSIM values for the Vector Median Filter (VMF), spatio-temporal Vector Median Filter
Fig. 9. RFSIM values for the Vector Median Filter (VMF), spatio-temporal Vector Median Filter
Fig. 10. Combined Quality Metric (CQM) values for the Vector Median Filter (VMF), spatio-
temporal Vector Median Filter (STVMF) and weighted spatio-temporal one (STVMF) averaged
for all frames
208
Analysing presented results the advantages of using the spatio-temporal VMF

can be noticed, also for the combined metric, especially for the static or slow-
motion fragments of the video frames with large amount of noise, even using the
unweighted kernel. The improper choice of the spatio-temporal kernel weights may
lead to worse results than the application of unweighted spatio-temporal filter.
Most of the image quality metrics used in the paper lead to the correct conclusions
using the frame-by-frame analysis, except from the R-SVD metric.
A more detailed analysis can be conducted calculating the values of the chosen
quality metrics for the upper and lower parts of the frames separately. Since the
upper part is rather static and the upper one is much more dynamic (large number
of the nonzero motion vectors), the results obtained for those fragments of the
images differ significantly. In order to limit the amount of presented data, only the
MS-SSIM, FSIM and combined metric have been chosen for further analysis.
Fig. 11. MS-SSIM values for the upper and lower parts of the frames
209
Fig. 12. FSIM values for the upper and lower parts of the frames
Fig. 13. CQM values for the upper and lower parts of the frames
210
It can be easily noticed that the static parts of the frame should not be filtered
using the standard VMF filter in the presence of strong noise. A serious loss of
quality can be preserved using the spatio-temporal filtering, especially for the areas
with low variance. Lower parts of the frames containing many moving details can
achieve better quality using the VMF filter for small amount of noise.
The application of the weighting spatio-temporal filter does not lead to
significantly better results which could justify the increased computational
complexity. Many image quality metrics indicate even worse quality of such
filtered images in comparison to unweighted spatio-temporal VMF.
The usage of the R-SVD metric for the assessment of filtering results is at least
doubtful, since the metric indicate the increase of the quality for the images
containing more unfiltered noisy pixels. It is worth to notice that the higher values
of this metric are related to lower quality, similarly as for the MSE.
5 Conclusions
In the future research the application of an adaptive weighted spatio-temporal

filter utilizing also the motion vectors should reduce the motion blur visible in the
lower part of each frame and further improve the overall image quality. Since many
authors still use the MSE and similar metrics for the assessment of the filtering
results, the development of the new filters optimised in the aspect of some modern
image quality assessment methods seems to be an interesting area of further
research. Such approach should result in much better consistency of obtained
filtering results with human perception.
Presented results may also lead to the development of the hybrid adaptive filter,
which should estimate the local amount of noise in order to use the standard VMF
filtering procedure for less noisy areas and its spatio-temporal version for more
noisy fragments of each frame.
References
[1] Astola J., Haavisto P., Neuvo Y.: Vector Median Filters. Proceedings of IEEE, vol. 78
no. 4 (1990), pp. 678–689.
[2] Bartkowiak M., Domański M.: Vector median filters for processing of color images in
various colour spaces. Proceedings of the 5-th International Conference on Image
Processing and its Applications, Edinburgh, UK, 1995, pp. 833–836.
[3] Louverdis G., Vardavoulia M., Andreadis I., Tsalides Ph.: A new approach to
morphological color image processing. Pattern Recognition, vol. 35 (2002), pp. 1733–
1741.
[4] Mansouri A., Mahmoudi-Aznaveh A., Torkamani-Azar F., Jahanshahi J.: Image
Quality Assessment Using the Singular Value Decomposition Theorem. Optical
Review, vol. 16 no. 2 (2009), pp. 49–53.
211
[5] Okarma K.: Combined Full-Reference Image Quality Metric Linearly Correlated with
Subjective Assessment. Lecture Notes in Artificial Intelligence, vol. 6113 (2010), pp.
539–546.
[6] Okarma K.: Video Quality Assessment Using the Combined Full-Reference
Approach. Advances in Intelligent and Soft Computing, 2010, vol. 84 (Image
Processing and Communications Challenges 2), pp. 51–58.
[7] Okarma K.: Colour Image Quality Assessment Using the Combined Full-Reference
Approach. Advances in Intelligent and Soft Computing, 2011, vol. 95 (Computer
Recognition Systems 4), pp. 287–296.
[8] Ponomarenko N., Lukin V., Zelensky A., Egiazarian K., Carli M., Battisti F.:
TID2008 - a Database for Evaluation of Full-Reference Visual Quality Assessment
Metrics. Advances of Modern Radioelectronics, vol. 10 (2009), pp. 30–45.
[9] Sheikh H., Bovik A.: Image Information and Visual Quality. IEEE Transactions on
Image Processing, vol. 15 no. 2 (2006), pp. 430–444.
[10] Vardavoulia M., Andreadis A., Tsalides Ph.: A New Vector Median Filter for Colour
Image Processing. Pattern Recognition Letters, vol. 22 (2001), pp. 675–689.
[11] Wang Z., Bovik A., Sheikh H., Simoncelli E.: Image Quality Assessment: From Error
Measurement to Structural Similarity. IEEE Transactions on Image Processing, vol.
13 no. 4 (2004), pp. 600–612.
[12] Wang Z., Simoncelli E., Bovik A.: Multi-Scale Structural Similarity for Image
Quality Assessment. Proc. 37th IEEE Asilomar Conference on Signals, Systems and
Computers, Pacific Grove, California, 2003, pp. 1398–1402.
[13] Zhang L., Zhang L., Mou X.: RFSIM: A Feature Based Image Quality Assessment
Metric Using Riesz Transforms. Proceedings of the 17th IEEE International
Conference on Image Processing (ICIP), Hong Kong, China, 2010, pp. 321–324.
[14] Zhang L., Zhang L., Mou X., Zhang D.: FSIM: A Feature Similarity Index for Image
Quality Assessment. IEEE Transactions on Image Processing, vol. 20 no 8 (2011), pp.
2378–2386.
212
Sound recording with the application of microphone arrays

Eugeniusz Kornatowski
West Pomeranian University of Technology
71-126 Szczecin, 26 Kwietnia 10, e-mail: korn@zut.edu.pl
In this article the issues concerning sound recording with the use of three-dimensional
systems of several microphones were considered. The issues under study concern the so
called beamforming, which is modeling three-dimensional directivity patterns of
microphone arrays, as well as modern technologies of multichannel recording production
with the purpose of reproduction in surround sound systems.
1. Introduction
Modern technology of sound recording is, in most cases, based on multi-

microphone technology. For this purpose the microphones with specific directivity
patterns are used: omnidirectional (donut shaped), figure–8 and cardioid (including
the following subgroups: sub-, super-, hyper- cardioid). Application of a single
microphone - with a definite directivity pattern - to sound recording enables full
control of the recording: in all probability we can predict the magnitude and
geometry of an environment in which the sound emitted by particular sources will
be successfully recorded. However, while applying the multi-microphone
technology the final result is often difficult to predict. In such cases, in sound
engineering, the most popular method consists in conducting several test
recordings while changing the geometry of the microphone system. It is still more
complicated when a produced recording is supposed to be reproduced on a
surround sound system, e.g. 5.1. In such a case, apart from fulfilling the
requirement of recording the sound from particular sources with high quality, the
possibility of apparent sound source planar localization during multichannel
reproduction should also be ensured. Bearing in mind all these considerations, it
can be claimed that an analytic device enabling simulative microphone system
(microphone array) configuration would be extremely useful. In the literature this
process is often referred to as “beamforming” [1, 2].
2. Directivity patterns of microphone
The most important properties of microphones are determined by two

characteristics: the sensitivity and the directivity pattern. The sensitivity at a
specific frequency is the ratio of the voltage at the terminals (output) of a
213
E. Kornatowski / Sound recording with the application of microphone arrays
microphone loaded with nominal impedance to the acoustic pressure in the place of
location of the microphone [3]:
U ⎡V⎤
Su = , ⎢ ⎥ (1)
p ⎣ Pa ⎦
The directional properties of a microphone are determined by the ratio its of
sensitivity at any direction of incidence of a sound wave on the microphone to the
sensitivity at a perpendicular direction of incidence on the element receiving the
acoustic energy. The range of this ratio in the function of incidence angle of wave
is called the directivity pattern [3].
In simulation testing the most frequently modelled pattern is the directivity
pattern of a single microphone as a difference of the characteristics of an
omnidirectional (pressure) microphone and a figure-8 pattern (pressure – gradient)
microphone, located from one another at a distance of l. If τ is the delay between
signals of the two microphones, f determines the frequency and θ direction
(azimuth angle) from which the wave plane reaches both microphones, then the
directivity pattern of the microphone modelled is expressed by the following
equation [4]:
⎛ ⎛ l ⋅ cos θ ⎞ ⎞
U(f , θ) = 1 − A ⋅ exp⎜⎜ − j ⋅ 2 ⋅ Π ⋅ f ⋅⎜τ + ⎟⎟ (2)
⎝ ⎝ c ⎠ ⎟⎠
where: A∈< 0, 1 >, c – speed at which sound travels in the air is equal to 340 m/s.
Depending on coefficient A and delay time τ it is possible to simulate a directivity
pattern from an omnidirectional to a figure-8 pattern. Figure 1 shows an example
pattern of a microphone with a cardioidal pattern.
In Figure 1(b) the axes are dimensionless. Every point of the surface presented
represents the sensitivity of the microphone along direction of a surface point to the
centre of the co-ordinate system.
Modelling of directivity patterns of arrays (matrices) of microphones in general
can be performed for a near – field or for a far – field.
In case of a near – field it is necessary to take into account the distance of a
sound source from individual microphones during calculations. For the far – field it
is assumed that the distances between microphones in the array are much smaller
than the distance of geometrical centre point of the microphone array from the
sound source; the front of acoustic wave is flat.
Furthermore, it is assumed that the far – field case is considered. For an array of
microphones located on a horizontal surface, the directivity pattern of such an array
can be described using the following equation:
214
U M (f , θ) = ∑ U m (f , θ − α m ) ⋅
m
⎛ l ⎞ (3)
⋅ exp⎜ − j ⋅ 2 ⋅ Π ⋅ f ⋅ m ⋅ cos(θ m − θ) ⎟
⎝ c ⎠
a)
90
1
120 60
0.8
0.6
150 30
0.4
0.2
180 0
210 330
240 300
270
b)
0.5
0
1
-0.5
0.5
-1
0
-0.2
0
0.2 -0.5
0.4
0.6
0.8
1 -1
1.2
Fig. 1. Directional characteristics of cardioidal microphone for one frequency of 1000Hz:

(a) polar plot, (b) 3-D directivity pattern, as a function of elevation and azimuth angle
215
Formula (3) enables the determination of a directivity pattern UM for a planar

array of microphones, taking into account the directivity patterns Um of the
microphones forming the pattern. The position of each microphone is described by
the distance lm from the centre of the XY co-ordinate grid and the angle θm of
direction from the microphone to the sound source. Angles αm are the angles of
rotation around the axis of each microphone and are measured in relation to the
positive part of axis X in a counter-clockwise direction. The formula (3) can be
generalized for the case of a 3-D array including a number of microphones “m” ,
then:
U M (f , θ, φ) = ∑ U m (f , θ − α m , φ − β m ) ⋅
m
⎛ l ⎞ (4)
⋅ exp⎜ − j ⋅ 2 ⋅ Π ⋅ f ⋅ m ⋅ cos(θ m − θ) ⋅ cos(φ m − φ) ⎟
⎝ c ⎠
where: αm, βm – angles of azimuth and elevation of microphone “m”, θm, ϕm –
angles of azimuth and elevation of direction microphone “m” – sound source.
3. Experimental and simulation tests
Tests were performed using the “Atmos” microphone system comprising

Brauner’s five microphones VM1 mounted on five-armed planar stand ASM5.
The “Atmos” system is designed to make recordings dedicated to surround
sound systems 5.0 (5.1). The microphones, all having cardioidal pattern, record the
following signals: C – front center (center), LF and RF – front left and front right,
LR and RR – rear left and rear right respectively. Each microphone can be rotated
around its axis within the range +/– 900 in relation to the axis of a matching stand
arm. The construction of the array is symmetrical to the front-rear axis and its
geometry is shown in Fig. 2.
Fig. 2. Geometry of stand ASM5 with microphones VM1 of “Atmos” system
216
The producer recommends that during the recording the axes of the
microphones’ maximum sensitivity should overlay the axes of respective arms of
the stand (standard set-up).
The “Atmos” array was used several times to do test recordings of concerts
performed by the symphony orchestra at the Philharmonic in Szczecin. The signal was
recorded by the Zaxcom “Deva” digital recorder with the resolution of 24 bits and the
sampling frequency of 96 kHz. During the test the axes of the maximum sensitivity of
each microphone overlaid the axes of respective arms of the stand (standard set-up).
During the playback of the performance recorded in the way already described, it
turned out that, despite loudspeaker systems’ set-up fulfilling the standard ITU-R-
BS.775-1, in the front sound stage an inaccurate apparent sound source localization
could be noticed, with the clear domination of the background (rear channels) and
center channel. It can be assumed that the reason for this negative effect lies solely in
the wrong set-up of angles of particular microphones. The quality of the microphones
themselves is beyond question – the VM1 microphones are among the best studio
microphones in the world. In order to possibly validate this hypothesis, simulation tests
were conducted using the method described in Section 2 with the assumption that the
sound source is located far from the microphone. Hence the simulation tests were
conducted for the so called far field.
a)
b)
Fig. 3. “Atmos” system: a) system set-up in Studio S1 in Polish Radio in Szczecin, b) recording event
at the Philharmonic in Szczecin
217
The directivity pattern of the “Atmos” system with standard set-up has a shape
as shown in Fig. 4.
a)
2 Back
0 Left
-2
-3
-2
Front
Right 1.5
-1
1
0 0.5
0
1 -0.5
-1
2 -1.5
b)
90 3
120 60
150 30
1
210 330
240 300
270
Fig. 4. Directivity pattern of “Atmos” system with microphones with standard set-up angles for
the frequency f =500 Hz: a) 3-D graph, b) directivity pattern in the XY plane
In the 3-D graph the axes are dimensionless. Each point of the surface presented
represents the effectiveness of array along the direction: surface point – the center
of the coordinate system XYZ. The obtained patterns substantiated the
observations related to degradation of the front sound stage. The directivity pattern
218
is very uneven (far from omnidirectional) especially in the area of the front sound
stage. The following test phase consisted in searching for optimal angles of rotation
of particular microphones. While changing these angles, with the use of numerical
modeling, such set-ups were sought which would provide maximally
omnidirectional directivity pattern, especially in the front sound stage. The tests
yielded the following conclusion: the LF microphone should be turned through an
angle of +700 in relation to the standard set-up (clockwise direction of rotation), the
RF microphone -700, the microphones C, LR and RR should be left as in the
standard set-up. The pattern obtained is presented in Fig. 5.
a)
Back
2
Left
0
-2
-2
-1
Right
0
Front 1
0.5
1
0
2
-0.5
3 -1
b)
90
2.5
120 60
2
1.5
150 30
1
0.5
210 330
240 300
270
Fig. 5. Directivity pattern of “Atmos” system after the adjustment of angles

for the frequency f =500 Hz: a) 3-D graph, b) directivity pattern in the XY plane
219
Having altered the set-up of the microphones, the recording session in the
concert hall took place one more time. Subjective “readability” and the localization
of apparent sound sources in the front sound stage was considerably enhanced in
comparison with the standard set-up of the microphones.
The recordings were subsequently submitted for subjective verification to the
Laboratory of Sound Engineering and Ambiophonics at the Faculty of Electrical
Engineering, West Pomeranian University of Technology in Szczecin – Fig. 6.
Fig. 6. Laboratory of Sound Engineering and Ambiophonics at the Faculty of Electrical Engineering,
West Pomeranian University of Technology in Szczecin
The tests of that kind are commonly applied. Due to their vital importance in
quality assessment of multichannel signals and systems of recording and surround
sound playback, the procedure and conditions under which the tests are conducted
are standardized [5, 6]. The test - concerning a subjective assessment of the
“Atmos” system after its optimization – was conducted with 28 participants,
listeners-experts. The participants’ task was to compare the quality of recording
samples obtained using the “Atmos” system with the microphones set-up both in a
standard and optimized way. The assessment concerned the following parameters:
1. The sound quality of front channels understood as:
− stability of front sound image,
− width of front sound stage,
− precision of apparent sound source localization,
− sense of appropriate localization of sound sources depending on the type
and character of a recorded event (e.g. instrument groups’ localization
during the playback of symphony orchestra concert being in line with
expectations).
2. The sound quality of rear channels understood as:
− stability of rear sound image (analogous to Point 1.),
− coherence of sound space (no feeling of void in certain spot, e.g. directly
behind the listener, in the area of rear sound space),
220
− appropriate arrangement and localization of sound sources in the space.

3. The sense of spatiality, which is:
− feeling of the size of the place,
− appropriate length of reverberation time for a given event conditions (for a
given space),
− realism of sound space,
− sense of “presence”: feeling of being in a place where a recording was
done as a measure of sound realism,
− ratio of direct sounds to reflected sounds: if reflected sounds dominate, the
sense of the so called “artificial acoustic perspective” arises.
4. Clarity (lucidity), which is:
− speech clarity,
− ability to identify and differentiate between voices and sounds to be heard
simultaneously,
− separation of individual short sounds occurring in short time intervals.
5. Balance:
− dynamics of a played back track (appropriate to the nature of an acoustic
event),
− loudness ratio between front and rear channels.
6. Overall assessment: subjective assessment of a recording comprising formerly
described parameters as well as the quality of the recording as a whole and
general impression the recording made on the listener.
The listener-expert assessment consisted in completing a questionnaire form
and marking individual parameters from 1 (bad) to 6 (excellent).
The results of the tests proved (Fig. 7) that the optimization of the microphones’
set-up was especially beneficial to the quality of the front sound stage. Furthermore,
the feeling of spatiality of played back recordings was greatly enhanced. And the
balance of the sound surrounding the listener was highly marked.
180
160
140
120
100
80
60
40
20
0
1 2 3 4 5 6
Pa r a me t e r n u mb e r
St andar d Af t er opt i mi zat i on
Fig. 7. Test results for the “Atmos” system before and after optimization of rotation angles
of microphones
221
The tests in question concerned the case of recordings conducted in a large

concert hall. It can be claimed with all probability that for recording in spaces of
small cubic capacity and a small number of sound emitting objects, an optimal
configuration of the “Atmos” system will be undoubtedly different from the one
presented above. In such a case, simulation tests of patterns should be conducted
for the so called near field [7].
References
[1] Van Trees H.L.: “Optimum Array Processing”, John Wiley & Sons, New York, 2002.
[2] Raichel D.: “The Science and Applications of Acoustics”, Springer, New York, 2006.
[3] Leighton T.G. : “The Acoustic Bubble”, Academic Press, San Diego, 1999.
[4] Jamroz A.: “The Design and Use of a Double Cardioid Stereophonic Microphone”,
Journal of the Audio Engineering Society, vol. 8 No. 2, 1960, pp. 100 – 104.
[5] European Broadcasting Union Tech. 3286-E: “Assessment methods for subjective
evaluation of the quality of surround programme material – Music”, EBU Official
Technical Texts, Geneva, 1997.
[6] European Broadcasting Union Supplement 1 to Tech. 3286: “Assessment methods for
the subjective evaluation of the quality of sound programme material – Multichannel”,
EBU Official Technical Texts, Geneva, 2000.
[7] Chen H., Ser W., Yu Z.: “Optimal design of near field wideband beamformers robust
against errors in microphone array characteristics”, IEEE Transactions on Circuits and
Systems, No. 54(9), 2007, pp. 1950 – 1959.
222
Analysis of noise suppression techniques for Track-Before-Detect

algorithms using spatial noise estimators
Przemysław Mazurek
West Pomeranian University of Technology, Szczecin
71- 126 Szczecin, ul. 26. Kwietnia 10, e-mail: przemyslaw.mazurek@zut.edu.pl
In the paper the noise suppression algorithms for Track-Before-Detect (TBD) systems
are evaluated. Three estimators are considered: the global, local, and single local. The
estimator based on the impulse counting for large number of pixels is proposed. Estimation
of the noise parameters gives abilities of suppression of the salt-and-paper or high-valued
impulse noise.
1. Introduction
Tracking of dim objects needs specially designed systems. Conventional

tracking systems are based on the detection and tracking scheme. The detection
algorithm is based on the threshold algorithm. Detection is not possible if the signal
is below noise floor because multiple false detections are possible. Another
problem is that threshold level should be set at appropriate level. The adaptive
threshold techniques do not supports low SNR objects also.
Alternative approach, based on Track-Before-Detect (TBD) scheme, is the only-
one solution for tracking of such objects. TBD algorithms test all possible
trajectories and accumulate signal values over every trajectory (it is the tracking
phase). Accumulated values are tested using the threshold value and the detection
occur if the accumulated signal strength is higher then the threshold level (it is the
detection phase). TBD systems are computationally demanding because all
trajectories are tested even if no-one object is in the range. Multiple objects
tracking ability is the advantage of the TBD algorithms.
measurements trajectories
Tracking Detection Assignment
Fig. 1 Track-Before-Detect processing scheme
Tracking abilities of TBD algorithm are obtained by the application of the

accumulative approach [1]. The multiple measurements technique is simplest
choice for the noise suppression. TBD algorithms support the time-based and the
spatial-based signal accumulations (integration). The time-based accumulation
technique is important for single measurement sensor. The spatial-based
223
P. Mazurek / Analysis of nosie suppression techniques for…
accumulation technique is used for multiple sensor systems. Both of them are
combined together for the best results if it is possible.
Typical TBD algorithm [1,2] uses advanced integration of signal instead the
simple accumulation. Some systems accumulate only over assumed trajectories and
the computation cost is very high. Reduction of the computation cost is possible by
the support of the trajectories switching using the Markov matrix. The motion
model supported by this matrix allows reusability of the obtained values from
previous basis trajectory.
2. Recurrent spatio-temporal Track-Before-Detect algorithm
There are two variants of this algorithm. Non-recurrent version is based on the
implementation of the multiple FIR filters. The more important is the recurrent
version based on spatio-temporal IIR filters. Following pseudocode shows this
algorithm.
Start
P(k = 0, s ) = 0 // initialization (1a)
For k ≥ 1
P − (k , s ) = ∫ q k ( s | s k −1 ) P(k − 1, s k −1 )ds k −1 // motion update (1b)

S
P (k , s ) = αP − (k , s ) + (1 − α ) X (k , s ) // information update (1c)
EndFor
Stop
k – iteration number,
s – particular space,
X – input data,
P − – predicted TBD output,
P – TBD output,
α – weight (smoothing coefficient), range: 0-1.
Detection abilities of this algorithm depend on many parameters, especially on
the smoothing coefficient and the Markov matrix (q). High value of the smoothing
coefficient reduces noise influence, but it can be applied for the object with a slow
transition between trajectories only. Low value of the smoothing coefficient
improves system response for without fast transition between trajectories or higher
SNR scenarios.
224
3. Noise in TBD systems
Noise in TBD systems could be different types and the Gaussian noise is
considered mostly. Impulse noise is very important for real systems also. There are
a few sources of impulse noise: electromagnetic interferences, high energy
particles, and other. Such noise peaks should be suppressed. Noise related to the
objects is another possibility, especially if the objects make maneuvers (change
trajectory), the signal value may change rapidly. The good example is the signal
reflected or emitted by the airplane. Radar and vision sensors may measure large
peaks due to physical orientation of the plane surfaces in relation to the e.g. radar
source (for radar wavelengths) or Sun (for vision wavelength sensors). Satellites
also create complex light signal sources.
Such high peak related to the object should be carefully considered. The large
peak improves SNR for TBD systems, but disturbs a possibility of the trajectory
change. A previous trajectory is preserved over a next scans due to value, and the
current position is not observable and is hidden in impulse response of TBD related
to this peak.
Suppression of impulses of both types is important and the robust filtering
techniques are necessary. The background signal is useful for adaptive estimation
of the noise floor and selection of the threshold level of saturation filters.
4. Robust filtering techniques
Robust filtering techniques give abilities of the impulse noise suppression.

Simplest filtering technique is based on the saturation or removal such pulses.
Threshold based algorithm is the simplest one algorithm.
Another method is the median filtering of the signal, but temporal filtering is
very costly. Computation of the median value needs an implementation of the
sorting algorithm. A specific case, based on the recursive computation of the
median value by the removal of oldest and by insertion of the new value is
possible, but the computation cost due to reorganization of stored data in the table
is significant. Temporal median filtering is well fitted to the FIR type TBD
algorithms.
Spatial median filters are also important for signal preprocessing and detection
of impulses. They have similar computation problems like temporal filters. The
implementation of the sliding window median filter is possible but still there are
additional sorting costs. Spatial median filters are well fitted into recursive (IIR
type) and non-recursive (FIR type) TBD systems. The median filtering is
independent on the TBD processing. A new incoming data are filtered before
computation using (1c) formula, only. Median filter are robust filter, well fitted
into different cases, but the computation cost reduction is also important factor for
real-time systems.
225
Another possibility is the tracking of the values and removal impulses using the
predictor (1b) response. The predicted value corresponding to the expected level
(reduced little due to smoothing coefficient). The new incoming data for particular
trajectory should be inside the expected range.
The gatting algorithm with a fixed gate (clipping values) is very simple and
robust. The predictor gives abilities of the adaptive changes of the predicted signal
level and especially the upper gate level. More advanced techniques, based on
adaptive gatting, are possible using the Kalman filter, but the computation cost of
the Kalman filter is much higher in comparison to the fixed gate. Computation of
the large number of trajectories is typical for TBD systems. Assuming 1000x1000
measurement grid and 100 motion vectors, there are 100M trajectories.
Computation of huge number of Kalman filters is very demanding, using
conventional electronics devices. Even a processing of TBD algorithms in real-
time is very challenging task and special design is necessary for specific
computation platforms.
Detection of the peaks and clipping them using the saturation function is simple
and effective. Removal of the peaks from computations is similar, alternative
technique. They are discussed in [4-8].
5. Saturation based techniques for impulse noise suppression
Saturation techniques for impulse noise are simple and they have efficient
software and hardware implementations [4-8]. The saturation level should be
estimated using noise observations. Proposed solutions [7] are specific for TBD
systems - available signal has value similar to the background level.
c=ks
y
c
x i,j y i,j
Input c Output to TBD
image -c x image
-c
Fig. 2. Saturation function controlled by the scaled standard deviation
The saturation should be value estimated for the case when there are no
impulses. This is simple if the single target is observed and the impulses are related
to the target. Standard deviation is much higher is there are many impulses not
correlated to the target. There are a many possible control strategies of the clipping
level c.
226
The standard deviation should be computed for the case when there are not
available disturbing pulses and fixed. The application of the 3σ level saturation
allows preserving 99.7% signal values without clipping. The right tail of the
distribution and higher then 3σ positive pulses are suppressed to this level. The left
tail and negative pulses are clipped to the -3σ level. This clipping level should be
extended if the signal values are not weak for improving of tracking. The additional
clipping of the values reduce detection ratio due to lower SNR introduced by the
clipping to the signal part.
5.1. Global spatial noise estimator for TBD
Global spatial noise estimator is based on the standard deviation calculation.

The noise parameters are similar statistically independent on target availability and
location. Single measurement is sufficient for estimation of the standard deviation s
so all pixels (N) are processed in all image.
1 N
sB = ∑ (x i − m B )2 (2)
N − 1 i =1
The background (DC level) is equal to the zero for most applications so m B ≈ 0 .
This is very often a result of the background suppression algorithms or acquisition
technique (pass-band filtering). Simplified formula is obtained due to a zero valued
mean:
1 N
sB = ∑ (x i )2
N − 1 i =1
(3)
Saturations are based on the estimated variance. Not all values of the Gaussian
noise are inside range, defined by the variance, so additional constant k should be
considered for range extension if it is necessary (Fig. 2).
This simple estimator has very important and practical feature. The changes of
variance of the Gaussian noise are immediately calculated and applied for the
current measurements, because there is no delay that is typical for more advanced
algorithms, predictors especially.
The much better is the calculation of histogram, because measured probability
distribution has long tails, but is not considered here.
5.2. Local spatial noise estimator for TBD
Calculation of the global variance for all measurements is not necessary. There
are two interesting techniques from implementation point-of-view. The first is the
local estimation of standard deviation, depending on currently processed pixel
(x,y), that limit spatial range to neighborhood area only A. Overall measurements
space S is substituted by the region A.
227
1
s B ( x, y ) = ∑ x x, y 2
N A − 1 x, y ⊂ A
(4)
This technique should be carefully implemented, because multiple calculations

of the standard deviation using direct implementations repeat most temporal
results. Implementation techniques, similar to decomposition of the 2D FIR filter to
the vertical and horizontal filters, should be applied. Such local method is well
suited if the standard deviation depends on the position.
5.3. Single local spatial noise estimator for TBD
Simpler technique that assumes fixed standard deviation for overall image
(measurement space) is possible. Instead of global or local (position dependent)
computations a selected and small area should be used for estimation. There is a
small number of targets and the expected targets’ signal level is low. Estimation is
possible using small a fixed piece of measurement space so this technique is low
demanding.
6. Impulse noise statistics
Impulse noise influences the global or local standard deviation. The background
estimation algorithms suppress DC signal from the background and preserve
impulse level relative to background level. The result of the background
suppression algorithm is the zero mean Gaussian noise (mB=0), with preserved
standard deviation (sB) in the ideal case. Nonlinear effects, related to the pixel
value representation and non-ideal background suppression, are not considered.
Assuming stable strength of the signal, the fixed mean value (mO=const) is
observed and the standard deviation of the object’s signal is zero (sO=0). Such
situation is related to the point size object, where a single pixel or measurement
cell is excited. In real situation the excitation may be different and the Gaussian
profile of the signal is typical. The signal of the object is additive to the suppressed
background also.
The impulse noise, related to the object or false measurements, is a high value
peak. This high value is located on the trajectory, if it source is the object. False
measurement may exist in any pixel, including trajectory also. The impulse signal
is additive also, or may replace low-level value, related to the measurement
Gaussian noise.
The estimation of the background related parameters like mean or standard
deviation (or variance) of the specific area of the image is disturbed by the object’s
signal and impulse noise. There is only one signal’s related a weak peak for 2D
measurements if the object is in measurement area. The mean value of the signal
228
for TBD systems is at noise level values, and signal is hidden in ±3σ range very
often, and such weak signal is omitted from the analysis.
The impulse noise needs a special consideration, because this signal influent on
the mean and standard deviation value of the selected image area. Variance for
specific area is shown by the following formula:
s 2A = s B2 + s O2 + s I2 = s B2 + s I2 , (5)
and the mean by the following formula:
mO M
mA = + c, (6)
N N
where N is the number of pixels in the specific area A, and M is the number of
pixels excited by the impulse noise. The level of the impulse noise is the c value
and need additional description.
Assuming that object’s signal is smaller in comparison to the peaks an
approximated formula for the mean is obtained:
M
mA = c. (7)
N
The high value peaks are especially important as source of disturbances of TBD
algorithms. The high values are simple to clip using saturation function, but the
clipping level is not known. The raw settings of the clipping level reduce higher
pulses to the c level. Some pulses that are below c level are preserved. It is
assumed, that all peaks are higher then c level, and are clipped.
The variance of impulse noise is derived using the following formulas:
1 N
s I2 = ∑ ( x i − m I )2 ,
N − 1 i =1
(8a)
N
1 M
mI =
N
∑ xi = N
c. (8b)
i =1
Values of complete impulse signal xi are c and zero. After derivation the
following formula is obtained:
M (N − M ) 2
s I2 = c . (9)
N ( N − 1)
The following formula is related to the variance of the area signal including
background and impulse noises:
M (N − M ) 2
s 2A = s B2 + c . (10)
N ( N − 1)
The following example shows influence for the real case, tested using Monte
Carlo approach. The 1000 samples are used for two areas (3x3 and 7x7 pixels) and
medium value of the single peak limited to the c=10 level (Fig 3 and 4). The
empirical values are very close the theoretical values. The most valuable estimator
of the mean value without impulse is the median what is well known fact, but the
cost of estimation is high due to necessary values sorting algorithm. The variance is
229
higher for smaller area and it is very important for estimation of the current
standard deviation for clipping control.
The local estimation of the standard deviation is necessary, especially if the
standard deviation is different depending on the area. The global estimation of the
standard deviation is not adequate, estimation of the standard deviation using only
one arbitrary selected area is not recommended also.
1 1 1
P(med)
P(var)
P(m)
0.5 0.5 0.5
0 0 0
-2 0 2 -2 0 2 0 5 10 15
Mean (no impulse) Median (no impulse) Variance (no impulse)
1 1 1
P(med)
P(var)
P(m)
0.5 0.5 0.5
0 0 0
-2 0 2 -2 0 2 0 5 10 15
Mean (single impulse) Median (single impulse) Variance (single impulse)
Fig. 3. Histogram for 1000 tests of empirical parameters estimation

(3x3 pixels area, impulse noise level 10)
1 1 1
P(med)
P(var)
P(m)
0.5 0.5 0.5
0 0 0
-2 0 2 -2 0 2 0 5 10 15
1 1 1
P(med)
P(var)
P(m)
0.5 0.5 0.5
0 0 0
-2 0 2 -2 0 2 0 5 10 15

(7x7 pixels area, impulse noise level 10)
230
The additional variance for single impulse is depicted in Fig.5. for standard
deviation equal to unity of the background Gaussian noise.
Fig. 5. Additional standard deviation due to single pulse availability, dependent

on the area (number of pixels), and the impulse level c
7. Impulse suppression using impulse counting
Obtained formula (10) is very useful for another interesting impulse suppression
algorithm. This algorithm uses two saturation blocks (Fig.6).
SAT
c2 output
image
c2 M
SAT Est. of
input c1 c2
image
Fig. 6. Impulse suppression using impulse counting
First saturation block uses saturation c1 value set at high level for counting of
pulses (M). Application of derived formula (14) gives a possibility of estimation of
c2 for pulse suppression. It allows suppression of impulses to the 3sB range.
M (N − M ) 2
s B2 = s 2A − c1 (11)
N ( N − 1)
The saturation level is defined as:
3s B = c 2 (12)
231
so
c 22 M (N − M ) 2
= s A2 − c1 (13)
3 2 N ( N − 1)
and finally:
M (N − M ) 2
c 2 = 3 s A2 − c1 (14)
N ( N − 1)
The following Monte Carlo example shows results for the proposed
algorithm.
1 1 1
p(med)
p(var)
p(m)
0.5 0.5 0.5
0 0 0
-2 0 2 -2 0 2 0 5 10 15
1 1 1
p(med)
p(var)
p(m)
0.5 0.5 0.5
0 0 0
-2 0 2 -2 0 2 0 5 10 15

(7x7 pixels area, impulse noise level 100, first saturation level 10, suppression to 3sB range)
Impulses that are between 3sB range and first saturation values are not counted
and degrade suppression. The cost of median filtering is much larger especially for
large input set, so proposed algorithm is interesting. Proposed algorithm restores
quite well statistic of the original image (Fig.7) for large number of pixels. Small
number of pixels (e.g. N=9) is not well suppressed due to significant dependence
(number of pixel dependent) between impulse and data. The formula (10) is true
for independent data, only. Large and saturated value from impulse should be
additive to particular signal value, but it is not true, and particular signal value is
lost. Larger sets reduce dependence, but the check of square root of the formula
(14) is necessary, because a negative argument forbids calculation. Very small
values may also disturb signal.
232
6. Conclusions
Proposed techniques for salt and paper noise or even for any high valued
impulse noise are recommended due to low cost if the proper variant is selected.
Computation of giant amount of Kalman filters or median filters is not possible
using today available devices. Simpler techniques based on the standard deviation
and known characteristic of noise should be used. Saturation technique is very fast
and has direct implementations in hardware. In this chapter selected techniques are
presented. The statistical parameters for additive impulse and Gaussian noise are
derived and alternative impulse suppression algorithm is proposed, based on the
counting of the impulses.
The most promising techniques are based on the applications of the TBD
predictors for gatting, due to adaptive processing for local image areas. It will be
considered in the further works.
ACKNOWLEDGEMENTS
This work is supported by the MNiSW grant N514 004 32/0434 (Poland).
This work is supported by the UE EFRR ZPORR project Z/2.32/I/1.3.1/267/05 "Szczecin
University of Technology - Research and Education Center of Modern Multimedia
Technologies" (Poland).
References
[1] Blackman S., Popoli R.: Design and Analysis of Modern Tracking Systems. Artech
House 1999.
[2] Stone L.D., Barlow C.A., Corwin T.L.: Bayesian Multiple Target Tracking. Artech
House 1999.
[3] Boers, Y., Ehlers, F., Koch, W., Luginbuhl, T., Stone, L.D., Streit, R.L. (eds.): Track
Before Detect Algorithm, EURASIP Journal on Advances in Signal Processing,
Hindawi 2008.
[4] Mazurek P.: Improving response of recurrent Track-Before-Detect algorithms for small
and point targets. 14. Konferencja Naukowo-Techniczna “Zastosowania Komputerów
w Elektrotechnice” ZKwE’2009 Poznań, pp. 351-354, 2009.
[5] Mazurek P.: Impulse noise of small and point targets in recurrent Track-Before-Detect
algorithms, "Academic Journals: Electrical Engineering". Poznań University of
Technology no.61, pp. 53-62, 2010.
[6] Mazurek, P.: Suppresion of impulse noise in Track-Before-Detect algorithms using
saturation and pulse removal, 15. Konferencja Naukowo-Techniczna “Zastosowania
Komputerów w Elektrotechnice” ZKwE’2010 Poznań, 301-302, 2010.
[7] Mazurek, P.: Suppresion of impulse noise in Track-Before-Detect algorithms,
“Computer Applications in Electrical Engineering”, vol.8, 201-211, 2010.
[8] Mazurek P.: Noise suppression for Track-Before-Detect algorithms using spatial noise
estimators, 16. Konferencja Naukowo-Techniczna “Zastosowania Komputerów
w Elektrotechnice” ZKwE’2011 Poznań, pp. 281-282, 2011.
233
Parametric identification of dynamic imprecise models based on

relational fuzzy cognitive maps
Aleksander Jastriebow, Grzegorz Słoń
Kielce University of Technology
25-314 Kielce, al. Tysiąclecia P. P. 7, e-mail: jastri@tu.kielce.pl,
enegs@tu.kielce.pl
In the paper certain approach to the selection of identification parameters of relational

fuzzy cognitive map’s dynamic model, based on appropriate computational algorithms of
identification, is described. On the selected simulation results there is presented the
comparison of the efficiency of different methods of parameters selection (adaptation) in
fuzzy relations depending on the adaptation assumed goal.
1. Introduction
The identification problem the is an important element of synthesis and analysis

of models of dynamic systems [1]. At solving it important role plays: selecting the
model’s type, the information on its structure and topology and completeness of the
information on its characteristics (obtained e. g. on the way of measurement) and
other essential parameters.
In the paper the problem of identification of fuzzy relational cognitive maps
dynamic models [2, 4, 5, 6] is stated. Specificity of such constructions lies in a fact
that the fuzzyfication subjects not only values of concepts but also relations
between them. In such a structure, regardless of parametric identification (power
coefficients and dispersion coefficients in relations), also functional identification –
connected with possible (not only linear) relational connections between concepts
[2, 6] is needed.
Certain approach to the solving of such type identification problems is
described. In this paper, due to complexity of the problem of comprehensive
adaptation of fuzzy relations, parametrical identification is focused.
2. Formulating the problem
Let fuzzy relational cognitive map type (1) is given:

<X, R> (1)
where: X = [X1, ..., Xn]T – fuzzy values of concepts; R = {Ri,j} – fuzzy relations
matrix.
At it a fuzziness of Gauss type with member function (2)-(3) was selected.
234
A. Jastriebow, G. Słoń / Parametric identification of dynamic imprecise...
2
⎛ x− X i ⎞
− ⎜⎜ ⎟⎟
⎝ σi
μ X ( x) = e
i
⎠
(2)
2
⎛ x2 − ri , j ( x1 ) ⎞
−⎜ ⎟
⎜ σi, j ⎟
μ R ( x1 , x2 ) = e
i, j
⎝ ⎠
(3)
where: μ X i – member function of the i-th concept’s fuzzy value; μ Ri , j – member
function of fuzzy relation between concepts i and j; X i – center of the member
function of the i-th concept’s fuzzy value; σi – dispersion (fuzziness) coefficient of
the i-th concept; x1, x2 – axes of a fuzzy relation’s universum; σi,j – dispersion
coefficient of the fuzzy relation between concepts i and j; ri,j – power coefficient of
the fuzzy relation between concepts i and j (ri,j(1) corresponds with power of the
relation in equivalent crisp model)
There can be defined following vectors of unknown (exact to experts’
knowledge) parameters:
Q = [Q1, Q2]T (4)
where: Q1 = [ X 1 , ..., X n , σ 1 , ..., σ n ]T; Q2 = [{ri,j}, {σi,j}]T; i, j = 1, ..., n.

Dynamic dependencies between concepts for Xi (i = 1, ..., n) are described by
following fuzzy relational equations [2]:
n
X i (t + 1) = X i (t ) ⊕[ X
j =1
j (t ) X j (t − 1)] o R j ,i (5)
j ≠i
where: i - number of considered concept (i = 1, ..., n), t - discrete time

(t = 0, 1, ..., T), - fuzzy addition operation, - fuzzy subtraction
operation, Rj,i–individual fuzzy relation between fuzzy concepts numbered j
and i, –maxmin fuzzy composition operation.
The task of the identification of parameters vector Q can be divided into few
problems presented in Fig. 1.
In Fig. 1 the identification goals are also schematically presented. In approaches
from points a) and c) the identification (adaptation) goal is obtaining such
parameters of a cognitive map, so as selected (the i-th) concept’s value, in a
specific moment of discrete time, obtains a level, which fits into a canal with given
width, where Fig. 1a) concerns defuzzyfied value and Fig. 1c) – fuzzy value. An
approach from Fig. 1b) presents the situation where as exact mapping of the i-th
concept’s value as possible is significant.
235
Fig. 1. Methods of preparation of the identification criterion for defuzzified (a), b)) and fuzzy (c))
values of concepts based on expert knowledge, Xi max, Xi min – assumed (on the base of expert
knowledge) limits of the i-th concept’s value, di – assumed (crisp) course of the i-th concept’s value,
Xiw – defuzzyfied value of the i-th concept
Approaches presented in Fig. 1 can be taken as follow:
Fig. 2. Diagrams of the identification criterion from Fig. 1 in crisp (a), b)) and fuzzy (c)) form
236
Irrespective of chosen method, identification of a cognitive map’s parameters

amounts to the such a selection of them, which allows to minimize given criterion.
This criterion depends on the identification goals and on used approach. At this, it
can be assumed, that its base will be the value of only one – selected – concept, or
aggregated values of all concepts. Criteria from Fig. 2, in reference to all concepts
(for selected moment t), can be written in the form:
– crisp:
w
J1 =
1 n
∑
2 i =1
[( 2
) ( 2
X iw − X i max + X iw − X i min ⇒ min
Q
)] (6)
1 n
w
2 i =1
(
J 2 = ∑ X iw − d i ⇒ min
2
)
Q
(7)
– fuzzy:
n
J R = ⊕ [( X i di ) (Xi d i )] ⇒ min (8)
Q
i =1
Remark:
If criteria (6)-(8) refers to average distance between given intervals and
defuzzyfied values of the model’s concepts, then these criteria should be completed
with additional summing with respect to consecutive steps of discrete time t.
Criteria (6)-(8) allows, theoretically, applying any optimization algorithms [3,
7] (with restrictions or without restrictions dependently on changes of values of
vector Q). Typical gradient, no-gradient, genetic and other methods can be rated
among such algorithms. They can be also completed with additional teaching
procedures (similar to neural networks – e.g. by Hebb method) [4].
3. Selected algorithms of identification of model (5)
To optimize criterion (6), (7) or (8) one can use (as mentioned in chapter II)
different optimizing algorithms, which may be presented in following general
form:
Q new = Q old η ΔJ (9)
old new
where: Q – previous approximation; Q – next approximation of parameters
vector Q;
⎧ ΔJ1w
⎪
ΔJ = ⎨ΔJ 2 w (10)
⎪ ΔJ R
⎩
– different type of the criterion (e.g. gradient or random value) „increase”
estimation; – addition operation (fuzzy – for fuzzy optimization or normal –
for crisp optimization); – multiplication operation (fuzzy – for fuzzy
237
optimization or normal – for crisp optimization); η > 0 – algorithm’s step

coefficient.
Computer realization of random version of algorithm (9)-(10) for criterion (7)
showed certain efficiency of the identification process. Selected results of the
research are presented in the next chapter.
4. Selected results of simulation research
The research subject was exemplary fuzzy relational cognitive map composed
of 5 concepts, general diagram of which is shown in Fig. 3.
The simulation purpose was checking the work of the algorithm of
identification of two parameters of fuzzy relations: relation power (ri,j) and
fuzziness coefficients (σi,j). The identification process goal was to select parameters
of fuzzy relations, so as concepts in analyzed model of cognitive map could
achieve values (after defuzzyfication) corresponded with reference values.
Analyzed map has closed structure, so time courses of concepts’ values can
oscillate and achieve relative stabilization after certain number of discrete time
steps.
Fig. 3. Exemplary cognitive map being the research object. X1 – X5 – values of concepts; Ri,j, Rj,i –
general denotations of fuzzy relations between concepts; i,j = 1, ..., 5.
4.1. Reference courses
Reference values were obtained on the simulation way, by building crisp

cognitive map with the structure from Fig. 3 and freely set values of relation
powers. Values of these powers are put into Table 1.
238
Table 1. Values of relation powers in cognitive map crisp model, which was a source
of reference values of concepts
r X1 X2 X3 X4 X5
X1 0 0,5 0,4 0,2 0,4
X2 0,3 0 0,3 0 0
X3 0 0 0 0,4 0,4
X4 0 -0,6 -0,4 0 0
X5 -0,4 -0,4 0 -0,3 0
Crisp model with relations from Tab. 1 was examined for initial values of
concepts from Table 2.
Table 2. Initial values (crisp) of concepts in analyzed model
Concept X1 X2 X3 X4 X5
Value 0,6 0,3 0 0 0,3
Time courses of concept values in crisp reference model (reference courses) are
shown in Fig. 4.
Reference courses
0,7
0,6
Normalized values
0,5
X1
0,4
X2
0,3 X3
X4
0,2
X5
0,1
0
1 3 5 7 9 11 13 15
-0,1
t
Fig. 4. Reference courses of the concepts’ values
4.2. Selection of the identification method
For the identification purposes criterion (7) was chosen. It should be

remembered that equation (7) in principle presents only certain idea of the
criterion. Detailed solutions can be different, depending on current needs and
239
possibilities. From this reason three tests were conducted for following approaches
to the criterion problem:
1. The value in selected – the 15-th step of discrete time was used as a reference
value. In this approach it’s assumed that value of each course in selected point
of discrete time has fundamental meaning. Additionally, instead total criterion
for all concepts, partial criteria for each concept were used, what resulted that
the algorithm’s action lad in the achieving local minima for consecutive
concepts and the identification precision was increased by multiple repeating
the adaptation cycle.
2. Full time course of given concept’s value was a reference course and
coefficients yi, consistent with equation (11), were consecutively minimized.
∑ (X )
w 2
i (t ) − d i (t )
yi = t =1
(11)
T
where: t – discrete time step; T – discrete time being the test subject; Xiw(t) –
defuzzyfied value of the i-th concept in the t-th step of discrete time; d i (t ) –
value of reference course of the i-th concept in the t-th step of discrete time.
3. The value in selected – the 15-th step of discrete time was used as a reference
value, but the criterion was directly connected with equation (7).
Tests, carried out for each of above mentioned approaches, showed, that, for the
result accuracy, chosen method of the final result achieving has significant
meaning. In principle two approaches can be applied. In the first one the series of
the adaption cycles is made, checking, after finishing each of them, concept values
only in one, final, point of discrete time (moment T). Such a method is faster but it
doesn’t give possibility to achieve good mapping of full time course. The second,
more complex, approach is based on achieving local optima for consecutive steps
of discrete time, at this in each next step if this time the adaptation starting point is
the optimum achieved for the previous step. So, it’s a certain type of recursion,
which is generally presented in equation (12).
⎧Q dla t = 1
Q(t ) = ⎨ init (12)
⎩ f (Q(t − 1) ) dla t ∈ [2, T ]
where: f – function of the criterion optimization consistent with (9); T – final step
of discrete time.
The tests proved that approach according to (12), despite it’s much more time-
consuming, allows achieving much better mapping of full time courses of
individual concepts with keeping expected values in target point.
240
4.3. Parameters adaptation algorithm
During the adaptation of fuzzy relations’ parameters it is difficult to find

analytical method of determining optimal partial changes. It results from the nature
of applied fuzzy numbers. It should be remembered, that functional changes
influence on universum and deform it. On the other hand, for the correctness of
maxmin operations on fuzzy numbers and relations, keeping the universum
invariability has crucial meaning. The method to reconcile these two contradictions
can be making frequent defuzzyfication and repeated fuzzyfication of concepts’
values on the correct universum, but such an approach could false the results. For
this reason, on current stage of the research, for fuzzy relations’ parameters
adaptation purposes, method of „trial and error” was chosen. The method lies in
making, according certain algorithm, little changes of powers as well as fuzziness
coefficients of relations with simultaneous observing results of these changes. For
positive results, the changes are fixed, for negative – cancelled. The process is
conducted till it achieves planned identification (limiting criterion value) accuracy
or till it achieves the state, when consecutive cycles don’t bring any changes. The
algorithm has built-in additional mechanism of ending based on executing the set,
limiting for given stage, number of the adaptation cycles.
4.4. Simulation results
The process of fuzzy relation’s parameters identification has supervised

teaching nature, based on a comparison with reference values. Initial conditions
(initial values of identified parameters) can be, in principle, assumed in any way,
but here there can be distinguished two main methods depending on knowledge
level in modeled system. If this level is relatively high, one can use expert
knowledge and initially determine approximate relation parameters. The adaptation
process goal can be then boil down to „specify” individual values. Such a process
would be less time- and labor-consuming. In situation of little or none knowledge
of the modeled system’s structure, assuming zero initial values of the relations’
powers is safest because finally they can achieve positive and negative levels as
well. As for fuzziness coefficients, assuming too low as well as too high values
would be equally adverse. Series of tests results, that their initial values can be
assumed between 0.3 and 0.6. The method of fuzzyfication of concepts is a
separate problem. Here crucial role plays expert knowledge.
For the analyzed model needs it was assumed that concepts as well as relations
are fuzzyfied on the base of Gauss type functions, where for concepts the collective
coefficient σ = 0.6 was assumed, while for relations identical initial values of this
coefficient equal to 0.4 were taken. For the symmetry keeping, the universum was
expanded to the range [-2, 2]. On this universum 17 sampling points, being
concurrently centers of linguistic values’ membership functions, were evenly
241
located. There were also assumed zero initial values of powers for all relations.
Three tests were executed and their results were marked:
– “point” – for method 1 from point 4.2,
– “course” – for method 2 from point 4.2,
– “criterion J_2_w” – for method 3 from point 4.2.
In each test different final values of parameters in fuzzy relations were obtained,
what means, that similar courses of defuzzyfied values of concepts can be resulted
from models that are differed each other.
The simulation was executed using computer application which was created
especially to this purpose. The results are presented in Fig. 5.
X1
0,7
0,6
0,5
Reference
0,4
Course
Point
0,3
Criterion J_2_w
0,2
0,1
0
1 3 5 7 9 11 13 15
t
X2
0,6
0,5
0,4
Reference
Course
0,3
Point
Criterion J_2_w
0,2
0,1
0
1 3 5 7 9 11 13 15
t
242
X3
0,4
0,35
0,3
0,25 Reference
Course
0,2
Point
0,15 Criterion J_2_w
0,1
0,05
0
1 3 5 7 9 11 13 15
t
X4
0,09
0,08
0,07
0,06
0,05 Reference
Course
0,04
Point
0,03 Criterion J_2_w
0,02
0,01
0
1 3 5 7 9 11 13 15
-0,01
t
X5
0,8
0,7
0,6
0,5 Reference
Course
0,4
Point
0,3 Criterion J_2_w
0,2
0,1
0
1 3 5 7 9 11 13 15
t
Fig. 5. Comparison of results of fuzzy relational cognitive map’s parameters identification,
made by three different methods
243
5. Conclusions
The paper is devoted to the advance of certain approach to solving the problem
of parametrical and functional identification of dynamic models of fuzzy relational
cognitive maps. The identification problem in forms: crisp and fuzzy was
presented. Identification criteria were also described. General algorithm of
parametrical identification (9)-(10) was adduced. On the selected example there
was conducted analysis of the identification algorithm based on one of the
proposed criterions for three different approaches. From carried out comparison
(Fig. 5) consequents that the method of comparison individual outputs courses with
reference courses gives best results. It delivers expected values of concepts in
given step of discrete time as well as whole course shape closest to the reference
course. It means that this method can be applied not only to the inquiring current
state of modeled systems but it also can find application for the simulation of
complex objects work.
References
[1] Jastriebow A.: Synthesis and analysis of m-step adaptation algorithms of the parametric
identification of static and dynamic objects on the base of discrete measurement.
Zeszyty Naukowe Politechniki Świętokrzyskiej, Wyd. PŚk, Kielce 1990 (in Polish).
[2] Jastriebow A., Słoń G.: Granular calculations in modeling of imprecise object using
fuzzy relational cognitive maps. Pomiary, Automatyka, Kontrola vol. 56 nr 12/2010,
pp. 1449-1452 (in Polish).
[3] Jastriebow A., Wciślik W.: Optimization – theory, algorithms and their executing in
MATLAB. Skrypt nr 401 Politechniki Świętokrzyskiej, Wyd. PŚk, Kielce 2004 (in
Polish).
[4] Papageorgiou E. I., Stylios C. D.: Fuzzy Cognitive Maps, In: Pedrycz W., Skowron A.,
Kreinovich V.: Handbook of Granular Computing, Chichester, England, John Wiley &
Son Ltd, Publication Atrium 2008, pp. 755-775.
[5] Pedrycz W.: Fuzzy Sets as a User-Centric Processing of Granular Computing, In:
Pedrycz W., Skowron A., Kreinovich V.: Handbook of Granular Computing,
Chichester, England, John Wiley & Son Ltd, Publication Atrium 2008.
[6] Słoń G.: The analysis of selected algorithms of the adaptation of relations in fuzzy
cognitive maps. Pomiary, Automatyka, Kontrola vol. 56 nr 12/2010, pp. 1445-1448 (in
Polish).
[7] Stach W., Kurgan L., Pedrycz W., Reformat M., Genetic Learning of Fuzzy Cognitive
Maps, Fuzzy Sets and Systems, 2005, V. 153, pp. 371-401.
244
The significance of wireless communication

for the metering data transmission via Smart Grids
Robert Jędrychowski
Lublin University of Technology
20-618 Lublin, ul. Nadbystrzycka 38A,
e-mail: r.jedrychowski@elektron.pol.lublin.pl
The article presents technological solutions of the communication and data transmission
area that are indispensable for the organization of smart metering. As the term “smart
metering” is very broad, particular attention has been focused on solutions based on the
GSM-based wireless transmission that makes possible to flexibly model the process of
remote meter reading with the use of the existing technological infrastructure. The article
determines factors that can decide over the application of a such communication type as
well as technical specifications to be met in order to develop a system that could
successfully operate over an extensive network. Characteristics of the communication
equipment used in real-time metering and billing systems have also been presented.
1. Introduction
Among the most important smart grid components there is a technologically

advanced (smart) metering system that provides information about energy
consumption and consumer profiles. The realization of such a system requires
many technical and organizational actions to be performed in order to ensure
correct operation of the system and to secure the expected profits. The term smart
metering contains many issues [1] and among them the most important is an
advanced metering infrastructure (AMI) and its integral components such as
automated meter reading (AMR) and advanced meter management (AMM).
Smart metering makes possible to supplement the basic metering function by
such ones as the configuration of metering and communication circuits, energy
supply management and a real-time data analysis at the levels of a consumer or the
distribution system operator (OSD). Smart metering can be understood as an
automated process of data acquisition, management of the data and their
transmission realized by the remote meter control, bilateral communication as well
as taking actions that increase the system management efficiency.
The application of AMI is advantageous as it makes possible to obtain
additional information concerning the grid operation and the consumer
characteristics. The basic information includes readings of power and electrical
energy consumption as well as the load profile recording. Among supplementary
information data there are readings of voltage and its parameters, load control and
245
R. Jędrychowski / The significance of wireless communication for the metering data …
detection of power outages (failures) and of energy theft. The data will make
enable faster response to changes and events occurring at the consumer side as well
as better adjustment of tariffs to the needs of individual consumer groups. The end
results should be beneficial for both the energy supplier and the consumers.
Fig. 1. Communication model of the system installed in the buildings

of the Lublin University of Technology
Technical specifications presented in the article are based on the metering

system shown in Fig. 1 The system has been developed within the framework of a
research project carried out in the Department of Power Systems of the Lublin
University of Technology. Component elements of the system have been located in
the switchgear units of the University buildings and are sources of real-time data
concerning the buildings.
2. Advanced metering infrastructure
The development of advanced systems for automatic meter reading is possible

owing to fast developing information technologies and electronic applications. The
systems include equipment and consumers distributed over vast areas. They record
consumption of not only electrical energy but also of other media.
The AMR offers basic functions for the data access. The abbreviation,
according to its definition, describes distributed metering systems that enable
remote meter reading for residential and industrial consumers. Usually, such
246
systems are composed of meters that are equipped with incorporated or external
communication modules. Interrogation of the metering instruments is realized by
the data-acquisition server that makes a data source for later downloads to billing
applications of the energy supplier.
Within the AMR system structure the three below-given layers can be
distinguished:
1. Reading layer that is responsible for data acquisition. It is composed of the
consumer measuring circuits (meters) that collect metering data. Pulse meters,
electronic meters as well as the latest microprocessor meters can operate within
that layer. They cooperate with communication modules.
2. Communication layer, whose task is to secure data exchange between meters
and communication modules and subsequently – between the modules and the
data-acquisition server. The cooperating devices communicate based on the
master-slave rule and the data transmission is unidirectional. .
3. Information layer that is responsible for data processing Its task is to archive,
process and present the obtained data for the system operator.
As compared to the AMR systems the AMI ones realize much more developed
functions. They operate based on the assumption that the metering is realized by
smart meters and the obtained data are accessible for both the network operator
(who also is the energy supplier) and the energy consumers. Within such an
arrangement two below-given communication areas can be distinguished:
− between a meter (or a concentrator) and devices connected to the consumer
installation;
− between a concentrator and the operator billing system that processes the data.
It can be easily noticed that at upgrading the AMR system to the AMI,
unidirectional communication over one area has to be replaced by bidirectional
communication over at least two areas. It is more difficult technically-wise, but
seems to be advantageous as it brings benefits both for the system operator and for
the consumers.
The benefits at the operator side are the following:
1. Automatic reading of the basic data concerning electricity consumption
metering.
2. Reading of supplementary measuring data concerning:
− power consumption and the load profile record,
− feeding voltage and its parameters,
− load control,
− detection of power outages (failures),
− detection of energy theft.
3. Fast response to changes and events occurring at the consumer side.
4. Tariff adjustment to the needs of individual consumer groups.
5. Modeling of energy demand.
247
6. Easier forecasting of energy consumption in the operator system and balancing

of the system.
7. Load control for individual elements of the system.
Among benefits at the consumer side the following ones can be indicated:
1. Load operation time adjusted to the tariff (actual energy price).
2. Control of the supplied energy parameters.
3. Permanent control of energy costs.
4. Reduced cost of the energy purchase.
In order to obtain the above listed benefits it is necessary to secure
communication between the AMI component elements. The kind of
communication technology to be applied depends on the number of cooperating
devices, distances between them, accessibility of individual technologies and of the
considered AMI area. At the consumer area, data transmission between a smart
meter and the loads can be realized by:
− wireless transmission with the application of power line communication (PLC),
− Ethernet,
− local wireless networks (WLAN, Bluetooth),
− wireless network technologies used for the control of intelligent building
installation elements (ZigBee, EnOcean).
In the case of the concentrator application, data transmission between a meter
and the concentrator can be realized by pulse signals or with the use of RS232,
RS485, CLO connections. Data transmission over the area between the
concentrator and the acquisition system can be realized via:
− radio frequency networks,
− GSM/GPRS networks,
− public switched telephone network (PSTN),
− Ethernet,
− wireless transmission with the application of power line communication (PLC).
Additional legal and technical requirements arise when a consumer wants to
function at the Energy Market and choose his energy supplier. In such a case, a
consumer will have to adjust his metering-billing system to the required standards.
The requirements are specified in the Regulation of the Minister of Economy of 4
May, 2007 on the detailed conditions for the functioning of the electricity system
as well as in the IRiESD (Distribution Network Operating Manual) - Balancing.
The AMI is based on standards that meet the above assumptions. Particular
attention is paid to the correct exchange of data, which is ensured by data
transmission protocols. The Distribution Network Operating Manual assumes that
metering-billing data transmission protocols generated by electronic meters and
electrical energy recording meters should be publically accessible and their data
format should be concordant with the distribution system operator requirements.
The data should meet the DLMS requirements or their equivalents specified by the
248
IEC 61334 and IEC 62056 standards. The protocols are meant for AMI systems
both at the wired and wireless communication mode.
3. Electrical energy meters
Although electricity meters are measuring instruments that have been used since
electrical energy became a commodity, their design has essentially changed over
the last few years. Popular induction meters get gradually replaced by digital or
microprocessor meters.
Energy metering can be described by the following dependence:
t2
A = ∫ Pdt (1)
t1
where: P- power, whose value can be time-variable: t1 - t2 –energy metering time.

In the case of digital meters, current and voltage readings are the basic data.
Digital metering of current and voltage makes an essential issue of modern
measuring techniques. There are two fundamental solutions for that type of
instruments and they differ by the input data processing mode. Processing of the
signals at first is realized in analog functional units, then the output analog signals
get processed to the digital form by analog-to-digital converters or undergo
preliminary sampling and the ADC processing followed by processing with
numerical methods [7].
In a digital circuit, electrical energy of a single network voltage period can be
determined as follows:
t e +T
A= ∫ u ( t )i( t )dt (2)
te
where: - instantaneous value of alternating voltage, - instantaneous value

of alternating current, - network voltage period, - time.
analog digital basic output
signal signal data data
signal micro-
A/D procesor procesor
I1, I2, I3 +A, -A
ui , ii
U1, U2, U3 +R, -R
Fig. 2. Block diagram of the metering unit
Modern electric energy meters are manufactured as ever more complex devices.
Initially, mechanical electricity meters were equipped with attached
communication units to process their readings into signals. Modern meters realize
249
energy measurements by solely electronic methods. Example diagram of signal

processing is presented in Fig. 2. Analog signals of current intensity I1, I2 i I3 and
of voltage U1, U2 i U3 are supplied to the meter directly or via transformers. Out
of the analog input signals, the measuring circuit calibrates instantaneous digital
values of voltage and current for each phase. Based on the signals, a signal
converter determines the following measurands:
− active power P1, P2 i P3 together with the direction indicator,
− reactive power Q1, Q2 i Q3,
− phase voltages (U1, U2, U3), phase currents (I1, I2, I3), and neutral current I0,
− phase angles between voltages U1 and U2 as well as U1 and U3,
− phase angles between voltages U1 and currents I1, I2 and I3,
− network frequency .
Aside with the mentioned signals, a meter can provide supplementary data
calculated on the basis of the above given values and among them:
− total active, reactive and apparent power,
− the content of selected higher harmonics in current and voltage for each phase,
− THD coefficient for current and voltage.
Modern meters are also equipped with a real-time clock function, which makes
possible to realize complex tariff plans.
Such meters offer various communication arrangements. Usually, it is realized
via RS232, RS485 ports. Presently, it has become a norm to install optical port
according to the IEC62056 (previously IEC61107) standard. IEC62056, Modbus,
DLMS are applied as protocols. Apart from the above pulse outputs are also used.
4. Remote data reading and configuration of meters
Present-day meters are modern digital instruments characterized by a number of

metering and control functions. The functions determine quantity of data that can
be processed and stored in a meter, such as:
1. Measuring data that include information described in the previous chapter.
2. Control data that include information of external events (opening of the casing,
magnetic field action), voltage decay, changes in the meter parameters, tariffs or
the date and time as well as of critical errors.
Aside with the above mentioned control data, information that enables remote
reconfiguration of selected meter settings, tariff change or disconnection of a
consumer (by the cooperation with the switch module) can also be supplied to the
meter.
The concept of remote meter reading and reconfiguration is not novel, but
development of the applied solutions is in constant progress. Initially, the reading
process has been realized by the system shown in Fig. 3.
250
The above shown system enables remote data transmission from a meter to an
application and the reverse way for the meter reconfiguration. The application can
store data of various meters, but the communication model remains unchanged.
Modems that are used for the communication purposes are applied to organize the
communication channel, but at no modification of the transmitted data. Classical
data transmission methods based on remote access to single meters with the use of
modems present a solution that has evolved by the application of both
telecommunication networks and wireless ones. Standard equipment of the present-
day meters includes communication interfaces that meet the readout system
requirements. In the situation of reading many thousands of metering points, such
a communication model needs a developed communication unit at the system data
processing side (a readout server or a configuration server), in order to make it
possible to set a connection and interrogate the instruments. Although the solution
dominates in the presently used equipment, it is not efficient and cannot be
calibrated [8].
modem modem
communication
system
application
energy
meter
Fig. 3. One-to-one communication system
In the case of a large number of measuring circuits to be included into a smart

grid the above mentioned communication model would be impractical. Alternative
solutions that are proposed by some metering circuits make possible to differently
organize the data acquisition system. Such circuits include intermediate devices
such as data concentrators that supply data collected from many meters to the
readout server or an application A model of such a communication system is
presented in Fig. 4.
cmmunication
system
application
concentrator
Fig. 4. Communication system with an incorporated concentrator
251
Concentrators can be classified into two groups:

− Passive ones that usually realize one-way communication and enable reading of
a group of meters.
− Active ones that realize two-way communication and enable local data
acquisition, their storage and recording of events that occur in the feeding and
communication systems. They also make remote change of the meter
configuration and settings possible. Apart from the realization of communication
functions, a concentrator also stores data and manages their transmission and
access for various users [7].
5. Application of the GSM connection
The development of a communication system that could meet requirements of

the smart metering concept involves a solution of many technical problems with
the organization of communication infrastructure in that number. The applicable
technologies have been discussed in the Chapter 2. Technical characteristics and
implementation costs are decisive factors at the selection of a technological
solution.
Wireless networks seem to make a good solution for a few important reasons,
such as:
− small data size, quantity of the data that are sent from a single meter depends on
a kind of the transmitted information and usually ranges from 50 to 300 kB,
− in most cases the data are sent periodically, which can be precisely determined
in the schedule,
− in the case of data loss, there is a possibility of their retransmission,
− as compared to cable networks, their infrastructure does not have to be as much
developed and the number of formal documents and procedures that are
necessary for the legal validity of their operation is much smaller.
Wireless networks offer a selection of a few applicable technologies with the
following options among them:
− networks operating at the 2.4 GHz frequency range, such as local networks
WLAN or Bluetooth as well as standards of intelligent installations as ZigBee
or EnOcean,
− dispatching or trunking networks presently applied to billing systems for
commercial consumers,
− networks operating within the ISM band (433 and 868 MHz); investigations into
the application of that band are currently carried out by the APATOR S.A. [3].
− GSM networks operating within the frequency band of 900 and 1800 MHz.
In the presented article, the GMS (GPRS) wireless transmission method has
been selected for further discussion. It is applicable, where investments into
permanent connections are not cost-effective. Owing to the GSM telephone
252
network system development the service is available all over the country and the
signal can reach objects distributed over large distances. The GSM technology
development has contributed to the increase of data transmission rate.
Along with the GSM network development, a dynamic development of devices
used for data transmission over the network can be observed over the recent years.
It is driven by a few below listed factors:
− the GSM signal covers an area of almost the whole country,
− package communication GPRS or even newer solutions can be applied together
with the IP protocol,
− APN can be provided for a user,
− data can be buffered and stored in communication modules,
− various communication modes can be applied.
GPRS-based wireless communication makes an interesting application for
measuring systems that realize remote meter reading. It is a case when
communication equipment is used both as the basic channel and as a diagnostic
one. In the basic channel, meter-collected measuring data are cyclically
transmitted. The same interface can form a diagnostic channel that enables
configuration of the meter. It is the character of the measuring instrument operation
that makes it possible to couple the basic and diagnostic channels into one. It is
feasible because of the cyclic transmission of metering data and only sporadic use
of the diagnostic channel. In that situation metering and configuration data do not
have to compete for the shared communication channel.
Electricity metering performed by electronic meters is of periodic character.
The metering is realized by measuring cycles and a billing period is also defined.
Such cyclic character makes local data collection feasible and it is one of the
factors that has made the GSM communication applicable to transmit data of
measuring instruments such as meters and the cooperating equipment.
Data of electricity meters can be transmitted by three modes: cyclically, on
demand, and on-line. The cyclically transmitted data contain information about
metering results concerning billing periods such as 24 hours or other time-bound
events. The on-demand transmission is meant for the download of data that are
currently saved to the meter memory. The on-line metering first of all concerns
actual values of current and voltage as well as of active and reactive power.
In order to download the meter data it is necessary to determine its connection
to a concentrator or a readout server. Three methods can be distinguished there.
The first one applies meters with an incorporated communication module and then
the GSM-based communication is realized directly between the meter and a
readout server [4]. The second method uses a concentrator connected to the meter
via pulse outputs. A pulse number corresponding to the energy value is counted at
the concentrator and then is transmitted to a server by radio waves. The third
method uses the meter-concentrator connection via RS 232, RS 485 or CLO. A
253
concentrator can include many configurable readout interfaces, which enables its
flexible cooperation with many meters of different configurations[1], [7].
Fig. 5. iSerwer GP of the NUMERON company manufacture
An additional advantage of concentrators that use the GPRS transmission is that

they apply the IP protocol to the communication process. Owing to that, data are
accessible not only for the network operator but also for a consumer, which is the
basic requirement at the AMI development within smart grids. Another benefit of
the concentrator application is that it makes possible to avoid the GSM operation
disadvantages, such as the system overloads, as till the moment of setting another
network connection, all the data are stored in the concentrator memory.
Useful functions of concentrators operating within GSM networks can be
exemplified by a device called iServer, made by the NUMERON Ltd in
Częstochowa [8]. The device enables:
− data download via various interfaces like CLO, RS485 or RS232 and pulse
outputs,
− automatic send of its data to various users via the FTP server or e-mail,
− an access to the data of internal memory (buffer mode) or directly to the meter
(transparent mode),
− data transmission in their original format or PTPiREE,
− monitoring of the network parameters including those related to the energy
quality,
− monitoring of events (collapse/recovery of voltage or transmission),
− determination of 10 reading schedules and of the readout data type and range,
− configuration via the embedded www page, from the command line or with the
use of specialized software tools,
− update of the device software.
254
Fig. 6. GP Server of the NUMERON company manufacture
Data of the device can be transmitted via the GSM network or LAN, which
makes possible to reserve data transmission channels and to adjust transmission
methods to the operator and consumer needs.
Additionally, the device is equipped with a GPS module that can be used for
time synchronization. It is particularly important for industrial consumers and
institutions, whose settlements are based on balancing readings of a few metering
circuits. In such a case clocks of all instruments have to be synchronized.
As all devices that use the GPRS wireless transmission have to be identified by
the APN, all of them have IP addresses and can take advantage of services based
on that protocol. Their operation can be currently controlled. Fig. 6 presents an
example window that shows current operation state of a concentrator. With its use
it s possible to determine status parameters of the device as well as the information
of the GSM and LAN operation.
255
6. Conclusions
The GSM network used for metering data transmission makes possible to
flexibly develop an arrangement for the AMI. Regarding the broad GSM coverage,
the equipment can be installed at any site and owing to the cyclical transmission
character, little interruptions of the network operation or its overload do not affect
the process of data acquisition. The application of the IP protocol and data-
buffering concentrators make possible to freely define the time and methods for the
transmission of current data. The data can be accessed both by the distribution
system operator and by the energy consumers, Monitoring of the equipment
operation with the use of the GPRS communication and the IP protocol introduces
a number of supplementary functions that make possible to manage the entire
metering system.
There are also some disadvantages of the presented solution. The most
important of them are high costs of the GMS services and of the APN setting.
However, even with those costs taken into account, for now the available GSM
infrastructure makes the presented solution a preferable one for the wireless
communication.
References
[1] Bahramiazar A.: Automated Meter Reading Using RF Technology. IEEE ISGT,
Gothenburg 2010.
[2] Billewicz K..: Problemy przygotowania danych pomiarowych. Rynek Energii
1(92)/2011.
[3] Kiedrowski P., Boryna B., Dunalski B. Lis J.: Bezprzewodowa sieć telemetryczna
zrealizowana w oparciu o układy nadawczo-odbiorcze krótkiego zasięgu. Rynek
Energii 1(92)/2011.
[4] Słociński P.: Teleinformatyczne systemy akwizycji i analizy danych pomiarowych
energii elektrycznej – kierunki rozwoju.. Rynek Energii 1(86)/2010.
[5] Rosołowski E.: Cyfrowe przetwarzanie sygnałów w automatyce elektroenergetycznej.
Akademicka Oficyna Wydawnicza EXIT, Warszawa 2002.
[6] Szafran J., Wiszniewski A.: Algorytmy pomiarowe i decyzyjne cyfrowej automatyki
elektroenergetycznej. WNT Warszawa 2001.
[7] LandisGry: E650 Seria 3 (ZMD400AT/CT, ZFD400AT/CT), Dokumentacja techniczna.
[8] Pozyton: SOLEN - oprogramowanie narzędziowe do informatycznej obsługi urządzeń i
systemów pomiarowych energii elektrycznej. Dokumentacja techniczna.
[9] Numeron: iSerwer – Karta katalogowa. Dokumentacja techniczna.
256
Remote laboratory with WEB interface

Dominik Łuczak
Poznan University of Technology
60-965 Poznań, Piotrowo 3a, e-mail: Dominik.Luczak@put.poznan.pl
Presentation of selected structures of remote laboratory. The presented structure enables

remote control through the Internet and the acquisition of measurement data. The
measurement data are presented using a web interface. The final result is microcontroller
measuring and control system with web server. System flexibility and ease of access to its
resources allow parallel planning an experiment to multiple users simultaneously.
1. Introduction
The laboratory is an important element of student education. Allows the

verification of theoretical knowledge in practice. Properly prepared laboratory
makes it possible to solve problematic issues. An important feature is also a
collaboration on the issue. In order to ensure a sufficient level of training
laboratory is equipped with control and measuring equipment. Devices should be
safe for the user, and resistant to damage due to the very frequent use. Another
issue is the timely access to the room. It is limited to normal operating hours of
institution and laboratory supervisor work time. Execution of experiment by
students at home is not always possible. There is no doubt it is required a solution
to whole day access to the laboratory. The use of virtual or remote laboratory
allows all time access[1]. It also allows laboratory to obtain completely safe for the
user. Proposed to remotely execute the experiment. Web based user interface allow
prepare and present experiment. The solution allows multiple people to use the
remote laboratory at the same time [2].
2. Remote control structures
A. Direct control
The idea of direct control is the ability to directly interact with the hardware. A
general diagram of the network structure is shown in Fig. 1. System contains the
following elements: network client, network and control system witch network
module. Network modules are available in performing all the layers even the HTTP
(Hypertext Transfer Protocol) or just the physical layer of Ethernet (ENC28J60,
Digi Connect ME).
257
D. Łuczak / Remote laboratory with WEB interface
Fig. 1. Structure of system in direct control
Direct control allows immediate execution of the requested operations by the

system. A disadvantage of this structure is the limited resources of embedded
system that must fulfill two roles: WWW (World Wide Web) server and process
control system.
A simple control system consists of 8-bit microcontroller and Ethernet
controller. The entire user interface is included in the program memory chip as a
static HTML (Hyper Text Markup Language) page. Interactivity can be introduced
through the use of technology, which performs all operations on the client side.
JavaScript may be used for this purpose [3]. The disadvantage of presented
structure solution is additional task for control system that must support
microcontroller.
This structure is recommended for small applications, where users require a
direct response of the system. Please note that such systems can handle a small
number of users.
B. Control via dedicated server
Fig. 2. Structure of system with dedicated server
Figure 2 shows the structure extended from the point A. The system has been
expanded to include additional dedicated server directly connected to the control
system.
The presented system has all the advantages of direct control, but introduces
additional cost during installation and operation. Dedicated server performs the
function of a web server with dynamic web pages. Page content is dependent on
the state of the control system. In system, the controller containing microcontroller
performs only certain control functions. The entire bandwidth generated by the user
interface is handled by a dedicated server. It is required physical integration of the
server and control system. Such a connection is not always possible because of the
258
bigger dimensions of the server in relation to the size of the controller. Connecting
a server with a controller on long routes adds delay.
Use a proxy server allows to collect the history of the control system and
placing them in a database. Additionally, can be implemented any network
services, including the need to install and configure the appropriate software.
C. Control via virtual server
Structure of the system using a virtual server is shown on Fig. 3. Virtual server
provides own resources: processor, memory and disk space without specifying a
physical location. The user is connected by the network to the virtual server in the
same way as with dedicated server. The difference is imperceptible to the user.
Fig. 3. Structure of system with virtual server
The controller based on microcontroller must be equipped with a network

module that allows for communication over the Internet. Embedded system can
connect to the virtual server in two ways: demand-response or continuous.
Universal solution is to request-response mode, where the controller is a client and
the virtual server is a server. This configuration does not require assigning global
static IP (Internet Protocol) address for network module driver. Virtual server can
be any free server provided by hosting company. It is recommended to use the
XML-RCP (Extensible Markup Language - Remote Procedure Call) for data
transmission (Fig. 4).
Fig. 4. Data transfer using XML-RPC

259
Using XML-RCP standardizes how data is transferred. It allows to simplify the

client requests to the server. The driver calls the remote procedure with the
appropriate parameters allows to: put current state of the system in database virtual
server and download commands for execution.
An important design parameter is the frequency of queries performed by the
embedded system. Too frequent polling of the server on new tasks can generate
unnecessary network traffic and server load when the user does not change
anything in the system. The use of long periods of questioning may result in loss of
interaction with the hardware.
D. Choosing the structure of remote control
In carrying out the work was used the structure from the point B. It allows for
direct user interaction with the hardware and allows for the extension of the
hardware with new control systems. The control system was implemented as a
separate system coupled to a dedicated server. In the role of a dedicated server was
used NGW100 (Network Gateway) Atmel . The controller is connected to the
server via RS232 serial interface (Recommended Standard 232).
3. Example of remote laboratory
A. Basic design goals
Performed the experiment should allow the user to interact directly with the
object. The proposed scheme of experiment was shown on Fig. 5. System consists
of temperature sensor and two actuators: fan used for cooling and heater used to
heat the object. All components are used by the controller. Experiment can by
executed in automatic mode with implemented two-state controller. Available is in
the Control Panel to change the set point temperature of the object and the width of
the zone of hysteresis controller. Figure 6 shows the system realization.
Fig. 5. Scheme of experiment
260
Fig. 6. Hardware realization
The aim of this study was to design and execution of an electronic system
allowing remote control via the Internet. The object of the remote control is an
experiment showing selected physical phenomenon. It is assumed the limited size
of the entire system to compact closing in a cube of side 30 cm. User should be
able to interact with an object through a simple web interface. The interface should
allow the client to direct interaction with the object in order to monitor the effects
of certain parameters to change the object. The system should have an automatic
operation mode to record changes in the object for a specific set of parameters of
the experiment.
The author proposed the observation of temperature changes the object
depending on the settings of the cooling element at a constant heating of the object.
Cooling is forced convection. HMI (Human Machine Interface) allows a user-
friendly way to control object. The study used a web page as the user interface.
Temperature changes of the object are presented on the chart. Client has the ability
to control an object in manual mode or automatic mode with two state regulation.
B. Proposed project
Performed the experiment should allow the user to interact directly with the
object. The proposed scheme was an experiment equipped with a temperature
sensor and an object implementing two components: the fan used to cool and the
heater used to heat object. All components are handled by the driver.
Transmittance of the control object should have a large rise time - first order
inertial object. As object was used aluminum radiator. It is possible to measure
object temperature and control heating or cooling element.
The heating system was completed attached permanently without the possibility
of exemption by the user. Client observes the current temperature. Attaching a fan
on or off record changes in temperature at constant heating facility. Added are
three LEDs signal which an operator can control directly. So that with one click on
the website can change the state of the system. Discussed was manual mode. Two-
state controller was implemented to use in automatic mode. It is available in
261
control panel to change setpoint temperature of the facility and the width of the
hysteresis zone controller.
Regulation is carried out according to the formula 1. The fan is switched on
when the temperature of the object y(t) rises above the setpoint temperature x(t)
plus half of the hysteresis zone H/2. The fan turns off when the temperature of the
object y(t) decreases below the setpoint x(t) minus half zone of hysteresis -H/2.
⎧ H⎫
⎪ OFF when e( t ) > 2 ⎪
u(t) = ⎨ (1)
− H⎬
⎪ON when e( t ) < ⎪
⎩ 2 ⎭
where e (t) = x (t)-y (t), x - setpoint signal, y - output value object, H - width of the
hysteresis. Two-state controller have static characteristics shown in Figure 7.
Fig. 5. Static characteristics of two-state element
Changing the width of the hysteresis controller change the quality of the two-
state regulation. When setting up a narrow zone of hysteresis is observed and the
frequent switching off the fan. Using a broad zone of hysteresis is made smaller
number of switching windmill. The result is a deterioration in the quality of
regulation.
C. Implementation of the control system
Designed board corresponds to the location of joints J5, J6, J7 and J16
NGW100 board. The module is applied to the board from the top. Board is made
entirely by hand on a plate with universal elements mounted in THT (Through-
Hole Technology). The result of the work is board in Figure 8. System elements:
1st power outlet, 2nd ATmega8L microcontroller, 3rd programming interface
standard ISP-6 (In-System Programming), 4th three pin connector for fan, 5th fan,
6th thermistor, 7th object, 8th heater, 9th 3 pins available with voltage 3.3 V, 10th
3 pins available with the voltage of 0 V, 11th quick connector for the heater, 12th
quick connector for the fan, 13th NGW100 connection with the controller.
262
Fig. 6. Detailed hardware realization
D. User Interface
Application user interface only requires to installed browser (preferably Mozilla

Firefox). The interface was made in PHP, HTML and JavaScript - JS library Charts
and jQuery. JS Charts allows to draw multiple series charts. In order to place the
chart in the website code to create <div> </ div> and set its id. Id element is used
by JS Charts to be placed in the chart area. jQuery extends the basic elements of:
buttons, sliders, dialog boxes and tabs. All user input data are checking on the
server side using the validator class. Login can be a string containing the digits
with a minimum length of 3 characters. The password can be a string containing
the digits with a minimum length of 4 characters.
The system allows to log in. Without logging user can access to information
about the author, information about the project, links and logging capabilities or
263
registration. Figure 9 shows the page without a valid session: 1st login panel or
registration, 2nd Additional information is categorized in the tab.
Fig.7. Login screen
After properly logged in user can change settings (Figure 10). Session
information is stored on the server in PHP $_SESSION variable. Settings tab will
be opened by default after logging. Available options on the settings tab: 1st
setpoint change through the slider - the range 10 - 80 ° C, the pitch of 1, 2nd
change the width of the hysteresis zone through the slider - the range 0 - 4 ° C, the
pitch of 0.1, 3rd change chart history by slider - the range 10 - 300 s, the pitch of 1,
4th temperature from the current measurement, 5th setpoint set on the controller,
6th hysteresis width of the zone in the controller setpoint, 7th The status of the fan
1 - on, 0 - off, 8th sequence number of readout data.
Fig. 8. Control settings
User can become familiar with the operation of two-state controller or can
control the object manually. Manual control interface is shown in Figure 11.
264
Manual control interface elements: 1st logout button, 2nd sets / hide markers on the
graph to allow the temperature reading after moving the mouse in the area of
measurement, 3rd Manual Control diode SYS - On, Off, pulsation, 4th A Manual
Control diode - On, Off, pulsation, 5th Manual control of diode B - On, Off,
pulsation, 6th the inclusion of a manual fan control - the control is turned off -
there will be two buttons for switching on / off fan, 7th temperature chart drawing
area, 8th preview of the selected measuring mouse over the field measurement, 9th
temperature chart, 10th tabs.
Fig. 9. Control system screen
The user interface exchanges data with the server using AJAX (Asynchronous
JavaScript and XML). Ajax is best solution for remote laboratory[4]. All read
commands are sent to the script in the background read.php. From the PHP opens
the device to read the Linux / dev/ttyS3 before data transfer to the control system is
in the background send.php file is used. The PHP commands are executed directly
from the system console. Sending data via the serial port requires echo'command to
send'>>/dev/ttyS3. In PHP system performs this function, the system (“echo'
command to send '>>/dev/ttyS3”);.
The application user interface requires him only installed browser (preferably
Mozilla Firefox). The interface has been implemented in PHP, HTML and
JavaScript - JS Charts and jQuery library were used.
E. Preparing experiment
To perform an experiment, need to set operating conditions of the system,

connect necessary cable hardware to the computer and configure the client
software. The maximum temperature which was read on the object during testing
265
was approximately 50 °C. The minimum possible temperature of an object is equal

to ambient temperature, it is caused by cooling by forced convection.
Experiment should be carried out under the same environmental conditions.
Recommended room conditions: 20-22 ° C ambient temperature and relative
humidity of 70-80%. Ambient temperature influences characteristics of heating,
cooling, and the maximum temperature of the object. The system should be set in
place without additional air circulation (open window, door or the air
conditioning).
F. Performing experiment
Sets of experiments were made for manual and automatic control. The first
series shows system in the control manual. Object was cooled by the constant
switching on cooling. The experiment starts with recording excluding cooling and
heating characteristics. It was a measured manually time 200 seconds, after which
the attached cooling. The second step allows to record the cooling curve of the
object. Heating and cooling characteristics of the object is shown in Figure 12.
Fig. 10. Manual control
To prepare an experiment in automatic mode must be specified: setpoint

temperature of controlling object and width of controller hysteresis zone.
Figure 13 shows temperature regulation. It is part of web interface view. Set
point was changed from 35 °C to 40 °C with constant hysteresis width of 1 °C. The
change is made in 150 seconds.
266
Figure 14 illustrates the impact of changing the width of the hysteresis control
zone on the quality of regulation. H changed from a value of 1 ° C to 2 °C in 75
seconds. Increasing H causes a smaller number of cycles of cooling but reduce the
quality of regulation.
Fig. 11. Setpoint change
Fig. 12. Impact of changing the width of hysteresis zone for control quality
267
4. Conclusion
Embedded Systems with Web interface allow to control any object in a simple
manner. The web browser is required. The use of extended structure consisting of:
dedicated server and microserver controller creates many opportunities for designers. The
designer is relieved from the multiple implementations of different network protocols.
Modularity of the structure allows for the preparation of many different experiments.
Common element will be a dedicated server, while the controller is implemented
separately.
Embedded systems with interface web allow to control any object in a simple
manner. Just Internet browser. Using the extensive structure consisting of: a dedicated
server and controller offers many opportunities for designers. Designer is relieved from
multiple implementations of different protocols. By using Linux, there are a variety of
applications, including enriching the functionality of the whole network system.
Dedicated server supports queries made by users - the execution of these operations is not
critical. Separation of the control allows to increase the reliability of the control object. In
the event that the operating system will be suspended driver will continue to perform his
task at last entered data.
Performed experiments are characteristic for first-order inertial object with a high
time constant. Object has a long time determining the value of the temperature. Slowly
objects allow to observation of variables by a man of their parameters in real time.
System has been designed and constructed in a modular way. Controller of the object
can be disconnected from a dedicated server and replaced with a newer. Controller board
has two quick release to allow exchange of heater and cooling element. When changing
the components it is necessary to remembered about a rating of the power components.
All I/O of microcontroller ATmega8 were derived to a simple one-strip strip peg. Strip
allows to extend the system with additional modules. Hardware changes require
additional software updates. Sources of the website can be freely changed. A dedicated
server allows to be freely configured.
References
[1] Bogosyan S., Gomes L.: "Current Trends in Remote Laboratories", IEEE Transactions
on Industrial Electronics, Dec. 2009, Vol. 56(12), pp. 4744 – 4756.
[2] Henke K., Ubar R., Wuttke H.: "Remote and Virtual Laboratories in Problem-Based
Learning Scenarios", ISM 2010, IEEE International Symposium on Multimedia,
Taichung 13-15 Dec. 2010, pp. 377 – 382.
[3] Braune A., Hennig St., Koycheva E.: "Model Driven Approach for Remote Laboratory
Visualization Systems", ICELIE 2010, 4th IEEE International Conference on E-
Learning in Industrial Electronics, Glendale 7-10 Nov. 2010, pp. 67 – 72.
[4] Alves G.R., Garcia-Zubia J., Lopez-de-Ipina D., Orduna P.: "Addressing Software
Impact in the Design of Remote Laboratories", IEEE Transactions on Industrial
Electronics, Dec. 2009, Vol. 56(12), pp. 4757 – 4767.
268
AUTHORS INDEX
Erik Boeren 45 Tomasz Trawiński 45

Krzysztof Budnik 69 Bohuš Ulrych 1
Marek Chomnicki 60 Janusz Walczak 181
Ivo Doležel 1 Marcin Wesołowski 120
Stefan F. Filipowicz 23 Józef Wiśniewski 150
Ryszard Frąckowiak 112 Mirosław Wołoszyn 60, 78
Tomasz Gałan 112 Maria Zielińska 161
Michał Gwóźdź 190 Paweł Zimny 78
Jacek Hauser 120
Kazimierz Jakubiuk 78
Aleksander Jastriebow 234
Robert Jędrychowski 245
Leszek Kasprzyk 161
Krzysztof Kluszczyński 45
Janusz Kołodziej 129
Eugeniusz Kornatowski 213
Václav Kotlan 1
Krzysztof Kowalski 139
Marcin Kowol 129
Daniel Kucharski 120
Dariusz Kusiak 34
Jacek Łubkowski 60
Dominik Łuczak 257
Marian Łukaniszyn 129
Wojciech Machczyński 69
Przemysław Mazurek 223
Jacek Mikołajewicz 139
Ryszard Niedbała 120
Lech Nowak 139
Krzysztof Okarma 200
Zygmunt Piątek 34
Wojciech Pietrowski 139
Jan Purczyński 12
Kazimierz Radziuk 139
Tomasz Rymarczyk 23
Jan Sikora 23
Andre Skibniewski 45
Grzegorz Słoń 234
Dorota Stachowiak 139
Zbigniew Stein 161
Rafał Stępień 181
Tomasz Szczegielniak 34
Marcin Szczygieł 45
Jerzy Tchórzewski 85, 102
269

!computer Applications in Electrical Engineering - 2011 - 9

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

!computer Applications in Electrical Engineering - 2011 - 9

Uploaded by

Copyright:

Available Formats

COMPUTER APPLICATIONS

Published by ARP Promocja21 Sp. z o.o.

Chairman of the Programming Committee

Secretary of the Conference

Electromagnetic friction clutch for fixing accurate position

A complete mathematical model of an electromagnetic friction clutch is presented,

A lot of technical applications (linear motors, various actuators and other

Fig. 1. Fixing of prescribed position of movable element by friction clutch:

Fig. 2. Arrangement of the considered friction clutch:

These holes are covered by friction fillings made of metal-ceramic friction

Table 1. Physical parameters of particular elements of the friction clutch

Element Material Parameter Value Unit

Fig. 3. Magnetization curves of particular steel materials:

The model is solved in the axisymmetric arrangement. Unambiguousness of the

5. Results of illustrative computations and their discussion

5.1. Evaluation of available materials for the magnetic circuit

5.2. Evaluation of geometrical configuration of the clutch

The paper is devoted to one of possible arrangements of an axial

Application of trapezoidal integration method

1. Modified Dirichlet window determined from the trapezoidal rule

where: wm(n) is defined by equation (4).

Applying the DFT to the Dirichlet window:

a complex spectrum form is obtained ([2],equation (21c)):

Having done the calculations the following equation is obtained

The amplitude characteristic of the modified window (solid line) demonstrates

By applying the DFT:

to the Hamming window

As another example the Blackman-Harris window will be examined [2]:

a complex spectra of the Blackman-Harris window (equation (18)) and the

Much like in Fig. 4, the LBM (θ ) amplitude characteristic demonstrates

Paper [2] provides yet another set of coefficients:

Table 1. Maximum values of the first sidelobe in relation to window length N

Paper [4] describes the rules of a polynomial window construction. A relation

Figure 6 demonstrates the logarithmic characteristics (in decibels) of time

In accordance with the accepted assumption, the equality of maximum values of

Figure 8 demonstrates the comparison between the logarithmic characteristics (in

As yet another example the Gaussian window was considered [2]:

The final example is a harmonic signal of the form:

a complex spectra X (θ ) and XM (θ ) are determined.

Figure 10 presents the amplitude of the complex spectrum X (θ ) (dotted line)

0 0.5 1 1.5 2 2.5 3

Fig. 11. Logarithmic amplitude characteristics of the harmonic signal

Both figures 10 and 11 demonstrate faster falling of the modified spectrum of

0 0.5 1 1.5 2 2.5 3

Fig. 12. Logarithmic amplitude characteristics of the harmonic signal

Fig. 13. Logarithmic amplitude characteristics of the harmonic signal x( n) = cos(0,75π n)

For all the examined time windows: Dirichlet, Hamming, Blackman-Harris ( -

Comparing methods of image reconstruction

The problem of the image reconstruction in Electrical Impedance Tomography (EIT) is

Fig. 2. The reconstruction methods

2. Level set method

Fig. 3. The iterative algorithm

3. Mumford Shah model

The Mumford-Shah algorithm sets formulation and minimization problem in

Conductivity γ is represented as:

Level set function is updated the following iterative scheme:

4. Variational level set method

The formulation of the variational level set method consists of an internal

5. Variational level set method with Mumford-Shah model

Objective function (min)

Fig. 9. The objective function - 2 objects

Influence of the external proximity effect on the power losses in

Fig. 1. Power station [10]