Jcscs Vol 5 NR 2 October 2012

Academy of Romanian Scientists
University of Oradea, Faculty of Electrical Engineering and Information Technology
Vol. 5, Nr. 2, October 2012
Journal of
Computer Science and Control Systems
University of Oradea Publisher
2
Volume 5, Number 2, October 2012
__________________________________________________________________________________________________________
EDITOR IN-CHIEF
Eugen GERGELY - University of Oradea, Romania
EXECUTIVE EDITORS
Gianina GABOR
Helga SILAGHI
- University of Oradea, Romania

Daniela E. POPESCU - University of Oradea, Romania

Viorica SPOIAL
ASSOCIATE EDITORS
Mihail ABRUDEAN
Lorena ANGHEL
Gheorghe Daniel ANDREESCU
Angelica BACIVAROV
Valentina BALAS
Barnabas BEDE
Dumitru Dan BURDESCU
Petru CASCAVAL
Horia CIOCARLIE
Tom COFFEY
Geert DECONINCK
Ioan DESPI
Jozsef DOMBI
Toma Leonida DRAGOMIR
Ioan DZITAC
Jnos FODOR
Voicu GROZA
Kaoru HIROTA
Stefan HOLBAN
tefan HUDK
Geza HUSI
Ferenc KALMAR
Jan KOLLAR
Tatjana LOMAN
Marin LUNGU
Anatolij MAHNITKO
Ioan Z. MIHU
Shimon Y. NOF
George PAPAKONSTANTINOU
Dana PETCU
Mircea PETRESCU
Emil PETRIU
Mircea POPA
Constantin POPESCU
Dumitru POPESCU
Alin Dan POTORAC
Dorina PURCARU
Nicolae ROBU
Hubert ROTH
Eugene ROVENTA
Ioan ROXIN
Imre J. RUDAS
Rudolf SEISING
Ioan SILEA
Lacramioara STOICU-TIVADAR
Athanasios D. STYLIADIS
Lorand SZABO
Janos SZTRIK
Honoriu VLEAN
Lucian VINTAN
Mircea VLADUTIU
ahin YILDIRIM
ISSN 1844 - 6043
Technical University of Cluj-Napoca, Romania

I.N.P. Grenoble, France
"Politehnica" University of Timisoara, Romania
University Politehnica of Bucharest, Romania
Aurel Vlaicu University of Arad, Romania
The University of Texas at El Paso, USA
University of Craiova, Romania
"Gheorghe Asachi" Technical University of Iasi, Romania
University of Limerick, Ireland
Katholieke Universiteit Leuven, Belgium
University of New England, Armidale, Australia
University of Szeged, Hungary
Agora University of Oradea, Romania
Szent Istvan University, Budapest, Hungary
University of Ottawa, Canada
Tokyo Institute of Technology, Yokohama, Japan
Technical University of Kosice, Slovakia
University of Debrecen, Hungary
Technical University of Kosice, Slovakia
Technical University of Riga, Latvia
Technical University of Riga, Latvia
Lucian Blaga University of Sibiu, Romania
Purdue University, USA
National Technical University of Athens, Greece
Western University of Timisoara, Romania
University of Ottawa, Canada
University of Oradea, Romania
"Stefan cel Mare" University of Suceava, Romania
Universitt Siegen, Germany
Glendon College, York University, Canada
Universite de Franche-Comte, France
Tech Polytechnical Institution, Budapest, Hungary
European Centre for Soft Computing, Mieres (Asturias), Spain
Alexander Institute of Technology, Greece
Technical University of Cluj Napoca, Romania
Technical University of Cluj-Napoca, Romania
"Lucian Blaga" University of Sibiu, Romania
Erciyes University, Turkey
This volume includes papers in the following topics: Artificial intelligence and robotics, Real-time systems, Software
engineering and software systems, Advanced control of electrical drives, Dependable computing, data security and
cryptology, Computer networks, Modern control systems, Process control and task scheduling, Web design, Databases and
data mining, Computer graphics and virtual reality, Image processing.
Journal of Computer Science and Control Systems

3
__________________________________________________________________________________________________________
CONTENTS
DACHIN Tudor1, MEZA Serban2, NEMES Marian3 ,VODA Adriana4, BADILA Florin5 - 1University Lucian Blaga
of Sibiu, Romania, 2Technical University of Cluj-Napoca, Romania, 3Continental Automotive Systems S.R.L.,
Sibiu, Romania, 4iQuest Technologies, Cluj-Napoca, Romania, 5Wenglor Electronic, Sibiu, Romania
Complexity Appreciation for BLDC Flat Top Sinus Implementation..................................................................................5
GOYAL Sumit, GOYAL Kumar Gyanendra - National Dairy Research Institute, Karnal, India
Use of Artificial Neural Network for Testing Effectiveness of Intelligent Computing Models for Predicting Shelf
Life of Processed Cheese.................................................................................................................................................9
HANGIU Radu-Petru, FILIP Andrei-Toader, MARI Claudia Stelua, BIR Kroly goston - Technical
University of Cluj-Napoca, Romania
A Z-Source Inverter for an Integrated Starter Alternator.................................................................................................15
HARLICA Ciprian, SZAB Lornd - Technical University of Cluj-Napoca, Romania
Real-Time Simulation Environment for Embryonic Networks .........................................................................................19
MARGINEAN Calin1, MARGINEAN Ana-Maria1, VESE Ioana1, TRIFA Viorel1, TRIFU Emil2 - 1Technical
University of Cluj-Napoca, Romania, 2S.C. TRAMBUS S.R.L., Cluj-Napoca, Romania
Design and Implementation of a PWM Inverter for Reluctance Motors..........................................................................23
RAMAKRISHNAN Sumathi1, MAHALINGAM Usha2 - 1Pavai College of Technology, Namakkal, India, 2Sona
College of Technology, Salem, India
Microstructure Development by Controlling Grain Size ..................................................................................................27
STOJANOVIC Igor, ZDRAVEV Zoran, TASEVSKI Angel - Goce Delcev University, Stip, Macedonia
Progressive Wavelet Correlation as a Tool for Recognition of the Images.....................................................................33
VIATTCHENIN Dmitri - National Academy of Sciences of Belarus, Minsk, Republic of Belarus
Detecting Outliers in Interval-Valued Data Using Heuristic Possibilistic Clustering........................................................39
4
__________________________________________________________________________________________________________

5
__________________________________________________________________________________________________________
Complexity Appreciation for BLDC Flat Top Sinus Implementation

DACHIN Tudor1, MEZA Serban2, NEMES Marian3 ,VODA Adriana4, BADILA Florin5
1
Universitatea Lucian Blaga Sibiu, Bd. Victoriei Nr.10, Sibiu, Tudor.Dachin@ulbsibiu.ro
Universitatea Tehnica Cluj-Napoca, Str. Constantin Daicoviciu Nr. 15, Cluj-Napoca, Serban.Meza@com.utcluj.ro
3
Continental Automotive Systems S.R.L., Str. Salzburg Nr.8, Sibiu, Marian.Nemes@continental-corporation.com
4
iQuest Technologies, Str. Motilor Nr.6-8, Cluj-Napoca, Adriana.Voda@iquest.ro
5
Wenglor Electronic, Str. Caprioarelor nr. 2, Sibiu, Florin.Badila@wenglor.com
Abstract - The paper presents the usage of a better

commutation technique taking into account existing
simulation and calculation models. An implementation is
suggested, after performing an accurate study of the motor
and the control system. The need for predefining the needed
space and calculus capability is a must in complex projects,
therefore the proposal is to pre-calculate most of the
required inputs and use look up tables inside the software
processing.
Keywords: flat top sinus commutation,
mathematical model, driving technique.
I.
simulations,
INTRODUCTION
Automotive applications that make use of Brushless

Direct Current (BLDC) motors have gained popularity
due to energy efficiency advantages, energy/volume
ratio and increased diversity and availability of
application specific control chips. BLDC motors, in
comparison with DC motors, have a longer service life,
do not have high emissions (caused by brush sparking),
can be used in harsher environments (e.g. hydraulic or
transmission gear oil), require smaller available volume
and can be driven with very high powers. A BLDC
motor has the similar efficiency as permanent magnet
synchronous machines. Especially due to energy
consumption concerns and the need to improve overall
system efficiency, usage of BLDC motors is considered
a viable alternative and/or an upgrade for more and more
automotive control applications. In order to encourage
this growth, the manufacturers (motor manufacturers or
integrators) have to increase the knowledge and
experience of their customers, to help in deciding
between various market available solutions.
One critical aspect of building a system is the design
stage and over-designing some aspects (like speed or
torque) is the simple solution to ensure product success.
Proper design must include calculations, simulations and
measurements to choose an optimum solution that uses a
motor that fits all requirements, without above described
over-design. Once a motor has been chosen, a control
scheme that matches the required speed and torque
control is needed.
The control scheme requires a minimum of hardware

processing power, which together with the desired
purpose of the Electronic Control Unit sets the
microcontroller requirements. This paper presents a
method of mathematically describing the motor, applies
said model in a Pspice simulation and then proposes a
less resource demanding implementation of the control
algorithm. In order to build up a good motor model, the
existing ones have been studied and advantages and
disadvantages compared. The used mathematical model
is to be found in the existing literature. The usage of this
model in order to implement the flat top sinus
commutation is the target novelty.
II.
MATHEMATICAL MODEL
In order to build a simulation model [1][2], the

necessary mathematical background has to be defined.
The following notations, equations and concepts were
used to define the simulation:
(1)
(2)
(3)
Where:
[a,b,c] in [Wb]=[V s] is the complete interlinked
magnetic flux through the coil a, b or c. [a,b,c] in [A] is
the magnetic flux of the wiring a,b, or c. o in [A] is the
magnetic flux in the star point of the equivalent
magnetic circuit. It is an operand in the magnetic model
and is being used to calculate the coupling of the
inductors (mutual inductance in star point). PM in [A]
is the magnetic flux of the permanent magnet. PM[a,b,c]
in [A] is the magnetic flux of the permanent magnet
through the wiring a.b or c. RM[a,b,c] in [H-1] = [A/Wb] is
6
__________________________________________________________________________________________________________
the magnetic resistance of the magnetic circuit through

the wiring a, b, or c. N number of turns of a winding.
III.
SIMULATION MODEL
The simulation was built up in Cadence Orcad

Capture and simulated with PSpice. The motor model
was created with the following approximations:
No skin effect losses in conductors;
No core losses;
No eddy currents;
No dependency between the inductance and the

magnetic flux through a coil;
The driving 3 phased voltage system is

symmetrical, rigid and does not contain a zero
component;
The magnetic flux density in the air gap of the

machine is energized by a sinusoidal electric rotation be
(basic wave model);
During real measurements, it is not possible to
separately measure each magnetic resistance due to
motors properties influence on the other system
components (L, R). Hence simulations cannot be
relevant only by simulating which magnetic flux is being
generated at which moment in time and phase
displacement. The first part of the simulation model
(Figures 1 and 2) describes the magnetic loop with
magnetic flux caused by permanent magnets.
Figure 1. Model of magnetic loop with electric flux from

permanent magnets
Figure 2. Model of magnetic loop with magnetic flux from

permanent magnets.
In the above figures the naming of elements is as

following: Theta Coil A, B, C - Magnetic flux of the
wiring a, b, c Theta PM A, B, C - Magnetic flux of the
permanent magnet through the wiring a, b, c. PSI PM a,
b, c - Linked magnetic flux of the permanent magnet
through the wiring a, b, c RM a, b, c - Magnetic
resistance of the magnetic circuit through the wiring a,
b, c. These resistances are dependent on the position and

the saturation of the coils.
An electronic control unit [6] and its attached motor
are connected with real wires to the battery/alternator
supply system, which have finite resistance and parasitic
inductance/capacitance. The resulting model (Figure 3)
is used to simulate real behavior (e.g. oscillations in the
cable harness).
Figure 3. Battery and connections model
The controlling element is a 6 transistor bridge

(Figure 4), which must be driven in such a way that the
motor receives the intended commands and that does not
allow simultaneous switching of the high and low side
transistors in the same inverter leg. This is implemented
as a minimum dead time between transistor firing.
Figure 4. 6-transistor Inverter Bridge
When the signal PWM_123 has the value 1, the

multiport switch will receive the value 0V towards the
exit. This situation implements the opening of the low
side transistor and closing the high side one. When the
signal PWM_123 has the value 2, the multiport
switch will receive the maximum differential amplitude
of the 3-phase voltage system towards the exit. This
situation implements the opening of the high side
transistor and closing the low side one. When the signal
PWM_123 has the value 3, the multiport switch will
implement a voltage dependent on the current and on the
voltage from the motor. The condition 3 presents the
condition of the H bridge with both transistors closed. As
long as any transistor/diode from the H bridge is closed,
there will be no current flow towards or from the motor.
When the potential from the motor side will be with
+0.7V higher or will drop with -0.7V, the flow path will
open and the voltage drop on the H bridge will be
limited at one of this values. This is caused by the
voltage drop on the internal/external diodes.

7
__________________________________________________________________________________________________________
IV.
CONTROL TECHNIQUES
The driving technique is being implemented through

software. For the PSpice simulation, a repeating pattern
defined as a digital stimuli file control the transistor
bridge.
The way of driving is being applied taking into
account the type of application.[7]
The signals for the H Bridge are the PWM signals
used to open the transistors. In real life and in case of
simulations, the driver has to have implemented a dead
time sequence between the turn-on of the phases in order
to avoid damaging the transistors and in order to allow
correct measurement of transferred power from the
supply to the load.
PWM signals are being applied in the following way:
High side PWM, low side full ON or OFF.
In order to avoid overheating the transistors, a
switching technique between firing the high side with
PWM and low side with full ON or OFF and inverse can
be implemented.
A. 6-step PWM control
One possibility is using a 6 step PWM control
technique (Figure 5), graphically described below:
Figure 5. 6-step control technique.
The switching pattern is the same for each of the 3

phases, but with a 120 offset.
In order to implement the technique in software the
following have to be taken into account: at 6-step
control, it is sufficient if the microcontroller receives an
interrupt at every 60 electrical degrees to activate the
next state of the bridge.
These interrupts usually are generated by hall sensors
which indicate the real position of the rotor relative to
the stator.
It is very advantageous for the implementation of
various control method if the microcontroller recognizes
the following five states:
Bridge is inactive (tri-state, high impedance)
Terminal connected to ground
Terminal connected to supply

Low Side PWM (If the duty cycle is equal to
100%, with ground connected)
High Side PWM (If the duty cycle is equal to

100%, with supply connected)
By having the sensors on the motor, any kind of
driving technique can be implemented; either rotational
control or torque control.
In order to use the BLDC as a stepper motor, the Hall
sensors are indispensable.
In order to drive the motor only rotational with no
accurate control, the driving technique can be
implemented without hall sensors.
For the implementation of the illustrated six-step
commutation of the "Terminal 1" the following
information has to be used in the driving procedure, and
has to be stored somewhere (e.g. a look up table).
330 -30 : bridge inactive
30 -90 : high-side PWM
90 -150 : phase to supply
150 -210 : bridge inactive
210 -270 : low-side PWM
270 -330 : phase to ground

B. 12 Step PWM driving technique
The 12-step [2], [3] control driving technique needs
an interrupt to the microcontroller at every 30 electrical
degrees, in order to enable the next state of the bridge.
These interrupts can be triggered either by six hall
sensors, placed at 60 from each other or the same
phenomena can be simulated by the microcontroller, by
having simulated hall sensors which generate the
additional required information for the driving technique.
A 12 step control is graphically described below (Figure
6).
This technique is almost the same as the 6 step
technique, the difference being that the switching time
between the phases will be at 30 instead of 60. This
offers better control, leading to less vibrations of the
motor and a more accurate driving when speaking of
rotation speed and torque control.
In BLDC applications common problems are power
losses, driving frequency and current ripple.
In order to implement BLDC motor control in a
human related application we have to take into account
the driving frequency as a starting point.
For many applications, it is usually between 21 kHz
and 23 kHz (or lower if the risk of hearing the control
cycle is also lower e.g. if control unit is placed far from
the driver), because these values are above human
hearing range and allow fast enough reaction times from
the control algorithm.
The higher the frequency of switching the better the
current control we will have.
Also, power losses increase due to higher
frequencies, so a value in between has to be considered,
in order to satisfy control, low power losses, low current
ripple and user comfort.
8
__________________________________________________________________________________________________________
(11)
(12)
(13)
All three legs of the inverter must have individual
PWM duty cycles, dependent of the rotor angle. The
duty cycle can be calculated using the below equation:
(14)
Figure 6. 12-step control technique.
A proposed method in order to have a good current

control is the Sinus Flat top modulation [5], [6], [7]. The
voltages resulting by applying the duty cycle will have a
quasi-sinus form. The final effect on the motor has to
take into account the integration property due to the
motor inductances. The phase voltages will be calculated
as follows:
(4)
(5)
(6)
The supply voltage will be delivered by applying a
variable Duty cycle between 00% and 255100%
(with 8 bit PWM hardware units). The PWM step
number and the extreme values are chosen after taking
into account how fine the resulting control must be. For
the next step we have to calculate the sinus voltage
without the star null point:
(7)
(8)
(9)
Where U[1,2,3] in [V] is the calculated voltage value
without the zero value from star point connection at
supply 1, 2 or 3. Also, we have to calculate the zero of
the system:
(10)
Where U123Null in [V] is the offset on the 3 phase
voltages. It must be added to the voltages calculated
without the zero system:
The maximum values of the duty cycle [6] will be

written in a look up table and the value of the rotor angle
and two other values, with 120 shift, read. All these
values will be inputted in the above equation and the
resulting PWM commands forwarded to the
microcontroller PWM blocks and then to the inverter.
In order to change between 6 and 12 step control,
only the values in the look up tables must be different.
Choosing between one and the other is as simple as
running a different initialization script.
V.
CONCLUSION
Mathematical models and simulation models were

built to study, improve and implement a control strategy
for BLDC motors, with minimum hardware/software
resources.
By requiring two look up tables and minimal position
information, the implementation of each driving
technique is easy to use, relying on emulation of
additional sensors and performing like more complicated
solutions. The type of the application will allow deciding
on the method to use.
VI.
REFERENCES
[1] Michel,
Robert.
2009.
Kompensation
von
sttigungsbedingten Harmonischen in den Strmen
feldorientiert geregelter Synchronmaschinen. Dresden :
Vieweg + Teubner, 2009.
[2] Schrder, Dierk. 2009. Elektrische Antriebe - Regelung
von Antriebssystemen. Berlin Heidelberg : SpringerVerlag, 2009.
[3] Tarmoom, Osama. 2006. Beitrag zur Auslegung von
Permanent-Magnet-Motoren fr spezielle Einsatzgebiete
dargestellt am Beispiel einer Versuchsmaschine. Cottbus :
s.n., 2006.
[4] AVR928: Scalar sensorless methods to drive BLDC
motors.
[5] FCM8202 3-Phase Sinusoidal Brushless DC Motor
Controler.
[6] Diplomarbeit 1999 Dozent: L.Wobmann Diplomand:
Patrick Fuhrer Hochschule fr Technik und Architektur
Bern Abteilung Elektrotechnik und Elektronik
[7] Motion Controller mit Sinuskommutierung fr ECMotoren Serir MCBL2805

9
__________________________________________________________________________________________________________
Use of Artificial Neural Network for Testing Effectiveness of

Intelligent Computing Models for Predicting Shelf Life of
Processed Cheese
GOYAL Sumit, GOYAL Kumar Gyanendra
National Dairy Research Institute, Karnal -132001, India
E-mail:thesumitgoyal@gmail.com; gkg5878@yahoo.com
Abstract This paper presents the suitability of

artificial neural network (ANN) models for predicting
the shelf life of processed cheese stored at 7-8C.
Soluble nitrogen, pH; standard plate count, yeast &
mould count, and spore count were input variables,
and sensory score was output variable. Mean square
error, root mean square error, coefficient of
determination and Nash - sutcliffo coefficient were
used in order to test the effectiveness of the developed
ANN models. Excellent agreement was found between
experimental results and these mathematical
parameters, thus confirming that ANN models are very
effective in predicting the shelf life of processed
cheese.
Keywords: artificial neural network; artificial
intelligence; processed cheese; prediction; shelf life;
radial basis (exact fit)
I. INTRODUCTION
Processed cheese is a very popular variety of cheese.
It is manufactured from 4 to 6 months old ripened grated
Cheddar cheese. A part of ripened cheese is often
replaced by fresh cheese.
During its preparation required amount of water,
emulsifiers, extra salt, preservatives, food colorings and
spices (optional) are added, and the mixture is heated to
70 C for 10-15 min with steam in a cleaned double
jacketed stainless steel kettle (which is open, shallow
and round-bottomed) with continuous gentle stirring
(about 50-60 circular motions per minute) with a
flattened ladle in order to get unique body & texture and
desirable consistency in the product.
The determination of shelf life of processed cheese
in the laboratory is very cumbersome and costly affair,
and takes a very long time to give results.
Therefore, it was felt that ANN technique, which has
been vastly applied for predicting the shelf life of
various food products, be employed for processed
cheese as well.
Hence, the present research was planned with the
aim to develop feedforward ANN single and multilayer
intelligent models for predicting the shelf life of

processed cheese stored at 7-8C.
The first Artificial Neural Network (ANN) was
invented in 1958 by psychologist Frank Rosenblatt
called perceptron. It was intended to model how the
human brain processed visual data and learned to
recognize objects.
An artificial neural network operates by creating
connections between many different processing
elements, each analogous to a single neuron in a
biological brain.
These neurons may be physically constructed or
simulated by a digital computer. Each neuron takes
many input signals, then, based on an internal weighting
system, produces a single output signal that's typically
sent as input to another neuron. The neurons are tightly
interconnected and organized into different layers.
The input layer receives the input; the output layer
produces the final output [1]. A radial basis function
network is an ANN that uses radial basis functions as
activation functions. It is a linear combination of radial
basis functions.
They are used in function approximation, time series
prediction, and control. Radial basis function network
consists of one layer of input nodes, one hidden radialbasis function layer and one output linear layer [2].
As an increasing number of new foods compete for
space on supermarket shelves, the words speed and
innovation have become the watchwords for food
companies seeking to become first to market with
successful products.
Overall quality of the product is of prime importance
in this competitive market and needs to be built into the
speed and innovation system. How the consumer
perceives the product is the ultimate measure of food
quality.
Therefore, the quality built in during the
development and production process must last through
the distribution and consumption stages.
Shelf life studies can provide important information
to product developers and manufacturers enabling them
to ensure that the consumer will receive a high quality
product for a significant period of time after production.
Since long time taking shelf life studies do not fit with
10
__________________________________________________________________________________________________________
the speed requirement, hence new accelerated studies

have been developed [3]. The results of this research
concerning development of radial basis (exact fit) ANN
models for predicting the shelf life of processed cheese
would be very beneficial for consumers, dairy factories
manufacturing processed cheese, wholesalers, retailers,
food researchers, academicians and regulatory
authorities.
II. LITERATURE REVIEW
The literature survey revealed that the ANNs have
been implemented for predicting the wide range of
different physico-chemical characteristics of various
food products including the quality and shelf life.
A.
Fried Potato Chips

Quality of potatoes in chips industry is estimated
from the intensity of darkening during frying. This is
measured by a trained panel, subject to numerous factors
of variation. Gray level intensities were obtained for the
apex, the center, and the basal parts of each chip using
image analysis of frying assays.
Feedforward ANN was designed and tested to
associate these data with color categories. The
developed ANN showed good performance, learning
from a relatively small number of data values. The
model behaved better than multiple linear regression
analysis. Predicted categories appeared to reproduce the
pattern of the experimental data issued from the trained
panel, revealing nonlinear mapping, existence of sub
regions and partial overlapping of categories.
Moreover, the generalization capacities of the
network allowed to simulate plausible predictions for the
whole set of parameter combinations.
Marique et al. (2003) were of the opinion that this
work is to be considered as a 1st step toward a practical
ANN model that will be used for objective, precise, and
accurate online prediction of chips quality [4].
B. Honey
Seventy samples of honey of different geographical
and botanical origin were analyzed with an electronic
nose. The instrument, equipped with 10 Metal Oxide
Semiconductor Field Effect Transistors (MOSFET) and
12 Metal Oxide Semiconductor (MOS) sensors, was
used to generate a pattern of the volatile compounds
present in the honey samples.
The sensor responses were evaluated by Principal
Component Analysis (PCA) and ANN. Good results
were obtained in the classification of honey samples by
using a neural network model based on a multilayer
perceptron that learned using a backpropagation
algorithm.
According to researchers methodology is simple,
rapid and results suggested that the electronic nose could
be a useful tool for the characterization and control of
honey [5].
C. Beef
A series of partial least squares (PLS) models were
employed to correlate spectral data from FTIR (Fourier
Transform Infrared Spectroscopy) analysis with beef
fillet spoilage during aerobic storage at different
temperatures (0,5,10,15,and20C).
The performance of the PLS models was compared
with a three - layer feedforward ANN developed using
the same dataset. FTIR spectra were collected from the
surface of meat samples in parallel with microbiological
analyses to enumerate total viable counts.
Sensory evaluation was based on a three-point
Hedonic scale classifying meat samples as fresh, semifresh, and spoiled. The purpose of the modelling
approach employed in this work was to classify beef
samples in the respective quality class as well as to
predict their total viable counts directly from TIR
spectra.
The results obtained demonstrated that both
approaches showed good performance in discriminating
meat samples in one of the three predefined sensory
classes.
The PLS classification models showed performances
ranging from 72.0 to 98.2% using the training dataset,
and from 63.1 to 94.7% using independent testing
dataset.
The ANN classification model performed equally
well in discriminating meat samples, with correct
classification rates from 98.2 to 100% and 63.1 to73.7%
in the train and test sessions, respectively.
PLS and ANN approaches were also applied to
create models for the prediction of microbial counts. The
performance of these was based on graphical plots and
statistical indices (bias factor, accuracy factor and root
mean square error) [6].
D. Dairy products and sterilized drinks
Attention has been focused on the application of
neural networks for developing different models for
various dairy products and milk based sterilized drinks:
Cakes [7]; soft cakes [8]; kalakand [9]; instant coffee
drink [10]; instant coffee flavoured sterilized drink [11,
12]; milky white dessert jeweled with pistachio [13];
brown milk cakes [14]; soft mouth melting milk cakes
[15]; post-harvest roasted coffee sterilized milk drink
[16]; and processed cheese [17,18,19,20,21].
III. METHOD MATERIAL
The input variables used for developing the ANN
computing models were the experimental data of
processed cheese relating to soluble nitrogen, pH;
standard plate count, yeast & mould count, and spore
count; and sensory score assigned by the trained
panelists was output variable (Fig.1). All in all 36
observations for each input and output variables were
used for developing the models.
The dataset was randomly divided into two disjoint
subsets, viz., training set having 30 observations, and

11
__________________________________________________________________________________________________________
validation set 6. Mean Square Error : MSE (1), Root

Mean Square Error : RMSE (2), Coefficient of
Determination : R2 (3) and Nash - Sutcliffo Coefficient
: E2 (4) were applied in order to compare the prediction

ability of the developed models. Training pattern of
ANN models is illustrated in Fig. 2.
Soluble
nitrogen
pH
Standard
plate count
Yeast &
mould count
Sensory Score
Spore
count
Figure1. Input and output parameters for ANN models.
N Q Q
exp
cal
MSE
n
1
(1)
1 N Qexp Qcal
RMSE
n 1 Qexp
(2)
N Q Q
cal
exp
R 1
2
1
Qexp

2
N Q Q
cal
exp
E 2 1
1 Qexp Qexp

Where,
Q exp = Observed value;
Qcal = Predicted value;

Qexp =Mean predicted value;
n = Number of observations in dataset.
(3)
2

(4)
12
__________________________________________________________________________________________________________
Training
ANN models
Selecting
minimum
error
Evaluation
of error and
weights
d
Figure 2. Training pattern for ANN network.
IV. RESULTS AND DISCUSSION

ANN models performance matrices for predicting the sensory scores are presented in Table 1.
Table 1. Results of Radial Basis (Exact Fit) model.

Spread Constant
MSE
RMSE
R2
E2
10
0.002660019
0.051575367
0.948424633
0.997339981
20
0.002422522
0.04921912
0.95078088
0.997577478
30
0.001958471
0.044254619
0.955745381
0.998041529
40
0.001767319
0.042039494
0.957960506
0.998232681
50
0.002009656
0.04482919
0.95517081
0.997990344
60
1.04864E-05
0.003238266
0.996761734
0.999989514
70
8.32216E-07
0.000912259
0.999087741
0.999999168
80
7.6146E-06
0.002759456
0.997240544
0.999992385
90
1.44227E-05
0.003797717
0.996202283
0.999985577
100
2.00537E-05
0.004478131
0.995521869
0.999979946
110
2.48217E-05
0.004982141
0.995017859
0.999975178
120
2.90291E-05
0.005387869
0.994612131
0.999970971

13
__________________________________________________________________________________________________________
130
3.29264E-05
0.005738157
0.994261843
0.999967074
140
3.66713E-05
0.006055683
0.993944317
0.999963329
150
4.03612E-05
0.006353049
0.993646951
0.999959639
160
2.72909E-05
0.005224066
0.994775934
0.999972709
170
2.44465E-05
0.00494434
0.99505566
0.999975553
180
2.16412E-05
0.004652014
0.995347986
0.999978359
190
1.90509E-05
0.004364729
0.995635271
0.999980949
200
1.67078E-05
0.004087523
0.995912477
0.999983292
Radial Basis (Exact Fit) model was developed for

predicting the shelf life of processed cheese stored at 78o C. The comparison of Actual Sensory Score (ASS)
and Predicted Sensory Score (PSS) for Radial Basis
(Exact Fit) model is illustrated in Fig. 3. Radial Basis
(Exact Fit) model with spread constant 70 [MSE:

8.32216E-07; RMSE: 0.000912259; R2: 0.999087741;
E2:0.999999168] gave the best fit amongst all the
studied experiments (Table 1).
Figure 3. Input and output parameters for ANN models.

The modeling results showed that there was
excellent agreement between the experimental data and
predicted values, with a high determination coefficient
(R2 = 0.999087741) and Nash - Sutcliffo Coefficient (E2
= 0.999999168) showing that the developed model was
able to analyze nonlinear multivariate data with very
good performance. It is evident from the high R2 and E2
that the ANN computing models are very effective in
predicting the shelf life of processed cheese.
laboratory observations by employing mean square

error, root mean square error, coefficient of
determination and Nash - sutcliffo coefficient. Very
good correlation was found between experimental data
and the developed mathematical models, thus
confirming the suitability of radial basis (exact fit)
artificial neural networks for predicting the shelf life of
processed cheese.
REFERENCES
V. CONCLUSIONS
An effective new radial basis (exact fit) method
based on artificial neural network is proposed for
predicting the shelf life of processed cheese stored at 78o C. The results were verified by comparing them with
[1] http://www.computerworld.com/s/article/57545/Artificial
_Neural_Networks (accessed on 11.1.2011).
[2] F. Mateo, R. Gadea, A. Medina, R. Mateo and M.
Jimenez, Predictive assessment of ochratoxin A
accumulation in grape juice based-medium by Aspergillus
14
__________________________________________________________________________________________________________
carbonarius using neural networks, Journal of Applied

Microbiology, vol.107, no.3, pp. 915-927, 2009.
[3] Medlabs Website:
http://www.medlabs.com/Downloads/food_product_shelf_
life_web.pdf (accessed on 1.3.2011).
[4] T. Marique, A. Kharoubi, P. Bauffe and C. Ducattillon,
Modeling of fried potato chips color classification using
image analysis and artificial neural network, Journal of
Food Science, vol.68, no.7, pp. 2263-2266, 2003.
[5] S. Benedetti, S. Mannino, A.G. Sabatini and G.L
Marcazzan,Electronic nose and neural network use for the
classification of honey, Apidologie, vol.35, pp. 16, 2004.
[6] Z.P. Efstathios, R.M. Fady, A.A. Argyria, M.B. Conrad
and E. N. George-John, A comparison of artificial neural
networks and partial least squares modelling for the rapid
detection of the microbial spoilage of beef fillets based on
Fourier transform infrared spectral fingerprints, Food
Microbiology, vol.28, no.4, pp. 782790, 2011.
[7] Sumit Goyal and G.K. Goyal, Brain based artificial
neural network scientific computing models for shelf life
prediction of cakes, Canadian Journal on Artificial
Intelligence, Machine Learning and Pattern Recognition,
vol. 2, no. 6, pp.73-77, 2011.
[8] Sumit Goyal and G.K. Goyal, Simulated neural network
intelligent computing models for predicting shelf life of
soft cakes, Global Journal of Computer Science and
Technology, vol.11, no.14, version 1.0, pp. 29-33, 2011.
[9] Sumit Goyal and G.K. Goyal, Advanced computing
research on cascade single and double hidden layers for
detecting shelf life of kalakand: An artificial neural
network approach, International Journal of Computer
Science & Emerging Technologies, vol.2, no.5, pp. 292295, 2011.
[10] Sumit Goyal and G.K. Goyal, Application of artificial
neural engineering and regression models for forecasting
shelf life of instant coffee drink, International Journal of
Computer Science Issues, vol. 8(4), no. 1, pp. 320-324,
2011.
[11] Sumit Goyal and G.K. Goyal, Cascade and feedforward
backpropagation artificial neural networks models for
prediction of sensory quality of instant coffee flavoured
sterilized drink, Canadian Journal on Artificial
Intelligence, Machine Learning and Pattern Recognition,
vol.2, no.6, pp.78-82, 2011.
[12] Sumit Goyal and G.K. Goyal, Development of neuron
based artificial intelligent scientific computer engineering
models for estimating shelf life of instant coffee sterilized

drink, International Journal of Computational Intelligence
and Information Security, vol.2, no.7, pp. 4-12, 2011.
[13] Sumit Goyal and G.K. Goyal, A new scientific approach
of intelligent artificial neural network engineering for
predicting shelf life of milky white dessert jeweled with
pistachio, International Journal of Scientific and
Engineering Research, vol.2,no.9, pp. 1-4, 2011.
[14] Sumit Goyal and G.K. Goyal, Radial basis artificial
neural network computer engineering approach for
predicting shelf life of brown milk cakes decorated with
almonds, International Journal of Latest Trends in
Computing, vol.2,no.3, pp. 434-438, 2011.
[15] Sumit Goyal and G.K. Goyal, Development of
intelligent computing expert system models for shelf life
prediction of soft mouth melting milk cakes, International
Journal of Computer Applications, vol.25, no.9, pp. 41-44,
2011.
[16] Sumit Goyal and G.K. Goyal, Computerized models for
shelf life prediction of post-harvest coffee sterilized milk
drink, Libyan Agriculture Research Center Journal
International, vol.2 , no.6, pp. 274-278, 2011.
[17] Sumit Goyal and G.K. Goyal, Radial basis (exact fit) and
linear layer (design) ANN models for shelf life prediction
of processed cheese, International Journal of u- and eService, Science and Technology, vol.5, no.1, pp.63-69,
2012.
[18] Sumit Goyal and G.K. Goyal, A novel method for shelf
life detection of processed cheese using cascade single and
multi layer artificial neural network computing models,
ARPN Journal of Systems and Software, vol.2, no.2,
pp.79-83, 2012.
[19] Sumit Goyal and G.K. Goyal, Time delay simulated
artificial neural network models for predicting shelf life of
processed cheese, International Journal of Intelligent
Systems and Applications, vol.4, no.5, pp.30-37, 2012.
[20] Sumit Goyal and G.K. Goyal, Estimating processed
cheese shelf life with artificial neural networks.
International Journal of Artificial Intelligence (IJ-AI), vol.
1, no.1, pp.19-24, 2012.
[21] Sumit Goyal and G.K. Goyal, Performance of
generalized regression, radial basis (fewer neurons), and
linear layer (design) computational ANN techniques for
shelf life prediction of processed cheese, International
Journal of Artificial Intelligence and Knowledge
Discovery, vol.1, no.4, pp.12-15, 2011.

15
__________________________________________________________________________________________________________
A Z-Source Inverter for an Integrated Starter Alternator

HANGIU Radu-Petru, FILIP Andrei-Toader, MARI Claudia Stelua,
BIR Kroly goston
Technical University of Cluj-Napoca, Romania,
Department of Electrical Machines and Drives, Faculty of Electrical Engineering,
Memorandumului, 28, 400114, Cluj-Napoca, Romania, E-Mail: Petru.HANGIU@mae.utcluj.ro
Abstract This paper presents an overview of an

integrated starter alternator system used in mild hybrid
electric vehicles. Inverter configurations that are used
in HEV are assessed with an emphasis on the novel Zsource inverter topology. The final part presents a
simulation model of a bi-directional Z-source inverter,
developed in AMESim, and the simulation results.
Keywords:
AMESim.
ISA;
model;
simulation;
Z-source,
I. INTRODUCTION
The conventional internal combustion engine (ICE)
era is at its dawn because of the increase in fuel prices
and a more stringent legislation on greenhouse gas
emissions. The obvious alternative for personal
transportation is the electric vehicle, but the technology
needed to make this type of vehicles accessible to the
masses is still at an initial stage. The hybrid electric
vehicle (HEV) is a viable compromise until more
efficient batteries or fuel cells are developed.
The HEV comes in different configurations.
Depending on the power flow within the vehicle and on
the energy sources, they can be classified as: series,
parallel, series-parallel or as mild HEVs. The mild HEV
configuration represents the first step in the transition
from an ICE vehicle to a full HEV and onwards to a full
electric vehicle. The integrated starter-alternator (ISA)
unit represents the key difference between an ICE
vehicle and a mild HEV.
The ISA is in fact an electric machine that has a two
quadrant operation thus combining in one unit the starter
and the alternator of a conventional ICE vehicle. Besides
its main function of cranking the ICE and generating
electric power, the ISA can be used to implement other
functionalities that may improve fuel efficiency and ride
comfort, such as: start/stop functionality, power
boosting and regenerative braking.
The ISA is driven by a power converter which, in a
conventional system, is composed of an inverter and a
buck DC-DC converter. The inverter steps up the
voltage of the battery and provides AC to the motor and
the buck converter steps down the rectified voltage in
order to charge the battery when the ISA is in generator
mode.
The traditional inverter configurations used for HEV
are the voltage source inverter (VSI) and the current
source inverter (CSI). The VSI has the advantage of a

low cost and a simple control but has important
drawbacks such as the necessity of a bulky dc bus
capacitor, high electromagnetic noises, high frequency
losses and the necessity of a separate dc-dc converter in
order to boost the battery voltage. The CSI main
advantage comes from its capability to boost the dc
voltage of the battery without a separate boost converter.
Other advantages come from the fact that it can
withstand short circuits across any two of its output
terminals, it doesnt need a bulky dc bus capacitor or
anti-parallel diodes thus reducing the overall size of the
inverter. CSI disadvantages are the inability to reverse
the DC current in order to charge the battery, high
conduction losses and the need for forced commutation
which means that only switches capable of blocking
voltage in both directions can be used.
The Z-source (impedance source) inverter (ZSI) is a
novel inverter configuration, first proposed by F. Z.
Peng [1], that overcomes some of the limitations of
traditional inverters.
This paper presents an application of a ZSI for a
permanent magnet synchronous motor ISA to be used in
a mild HEV. First the ISA architecture and ZSI are
presented in detail. The final part presents simulation
results of the ISA driven by a ZSI.
II. ISA SYSTEM OVERVIEW
As mentioned before, a vehicle equipped with an
ISA is in fact a mild HEV. The main difference between
a mild HEV and a full HEV comes from the fact that in
the case of the mild HEV, the ICE is always on and is
driving the wheels as long as the vehicle is moving.
A.
Drive train configuration

There are several possible arrangements within the
vehicle drive train for the electric machine acting as an
ISA [2]:
Classic arrangement
Coaxial arrangement
Non-coaxial arrangement
Electric motor in auxiliary drive
Electric motor in the transmission

In the case of a coaxial arrangement there are three
possibilities, one with the electric machine on the
crankshaft, one with the machine sandwiched between
16
__________________________________________________________________________________________________________
two clutches and one with the machine mounted on the

input shaft of the transmission. Similar arrangements are
possible in the case of the non-coaxial solution where
the electric machine that is now mounted on the side of
the drive train will be connected via a drive.
The most widespread solution is that with the electric
machine mounted on the crankshaft. The rotor is
attached directly to the ICEs crankshaft and replaces
the flywheel. The advantage of such an arrangement,
besides its simplicity, consists in achieving a dual mass
flywheel effect by connecting the electric machine to the
friction plate for the clutch, with a torsion damper. This
will eliminate excessive transmission gear rattle, reduce
gear shift effort and further increase fuel economy.
B.
Operation modes
The ISAs main function is to generate power while
the ICE is on but it can also assist the engine when high
loads are requested. The systems modes of operation
are:
1) Internal combustion engine cranking
2) Power generation mode
3) Regenerative braking mode
4) Power boost mode
C.
Electric machines
For functioning as both a starter and an alternator the
electric machine must have a wide constant power speed
range. It must provide a high starting torque for engine
cranking even at very low temperatures and it must have
a high efficiency in generator mode at speeds ranging
from 1500 rpm to 4000 rpm. At the moment the two
competing electric machines for ISA applications are the
Permanent Magnet Synchronous Machine (PMSM) and
the Induction Machine (IM).
III. Z-SOURCE INVERTER
The Z-Source inverter, shown in Fig. 1, replaces the
dc-link present in a conventional VSI with an impedance
network composed of two capacitors and two inductors.
This enables the inverter to utilize the shoot-through
(short circuit) states of the phase legs in order to boost
the dc-bus voltage. By varying the active state and
shoot-through duty ratios, the ZSI can either buck or
boost the dc-bus voltage.
A.
Control methods
In order to utilize the shoot-through states the pulse
width modulation (PWM) control used for conventional
inverters needs to be modified. Several control methods
have been proposed:
1) Simple boost control (SBC)
This method was first proposed in [1] and it utilizes
an upper and a lower limit to control the shootthrough states. When the triangular carrier wave is
greater than the upper limit, or lower than the lower
limit, the inverter is in shoot-through state, the rest
of the time the control is the same as a normal
carrier-based PWM. This methods disadvantage is
the high voltage stress on the switches.
Figure 1. Z-Source inverter.
2) Maximum boost control (MBC)

This method was presented in [3]. Its working
principle is to turn all of the inverters zero states
into shoot-through states thus minimizing the
voltage stress on the switches
Its main draw back is that it produces low frequency
ripples across the Z-network because of the variable
shoot-through duty ratio.
3) Maximum constant boost control (CBC)
This control method was proposed in [4]. The
drawbacks of the previous two methods are
mitigated by keeping a constant shoot-through duty
ratio while maximizing the boost factor. This is
achieved either by utilizing a variable lower and
upper limit for controlling the shoot-through states,
either by injecting a third-harmonic component into
the reference signals. The envelopes used in the first
case are periodic signals that have a frequency three
times higher than that of the inverter output. The
injected third harmonic component is 1/6 of the
fundamental component. When using third harmonic
injection the upper and lower limits for controlling
the shoot-through states are straight lines. The
distance between this two limits is constant and
equal to
3M [4].
B.
Component rating
The inverters boost factor B and gain G can be
determined using the following [4]:
1
(1)
B
3M 1
Vm
M
MB
ES / 2
3M 1
(2)
Where Vm is the peak phase output voltage, M is the

modulation index and ES is dc-bus voltage.
For rating the ZSIs inductors and capacitors, for any
control strategy, the following equations can be used [5]:
2 E S d S TS (1 d S )
(3)
L
3k i MI m cos
3d S TS MI m cos
8k v ES (1 d S )
(4)
Where dS is the shoot-through duty ratio, kv and ki are

desired voltage and current ripple factors, Im is the peak
phase current and TS is the switching period.

17
__________________________________________________________________________________________________________
Figure 2. Control model.
C.
DC-DC converter
In order to accept a reverse power flow from the ISA
to the battery, a current fed Z-source dc-dc converter,
proposed in [5] and [6] has been adopted.
This converter consists of a bi-directional switch,
composed of two IGBTs and two diodes, a second
switch and an inductor placed between the Z network
and the 3 phase bridge and a smoothing capacitor. The
bi-directional switch replaces the Z-source network
diode so that, depending on the desired operating mode,
the converter can now accept a direct or a reverse power
flow. This converter can perform either as a buck
converter, either as a reversed polarity buck-boost
converter depending on the duty ratio.
By controlling the duty ratio of the switches the
converter output voltage can be regulated. The voltage
transfer ratio of the converter is determined with the
following equation [6]:
2D 1
(5)
G
D
Where G is the voltage gain and D is the duty ratio.

IV. SYSTEM MODELING AND SIMULATION
RESULTS
The system model has been implemented in
AMESim, which is a multi-domain simulation software
for the modeling and analysis of one-dimensional (1D)
systems.
The chosen electric machine is a three-phase, 7.5
kW, outer rotor PMSM presented in [7]. The outer rotor
configuration allows for an easy integration of the
machine in the vehicle drive train and also helps with
the cooling of the permanent magnet rotor. The electric
machine specifications are:
Table 1. Electric machine specifications.
Activepower
Voltage
Frequency
Polepairnumber
Ratedspeed
Ratedtorque
Statorcurrent
Efficiency
Powerfactor
7.5kW
72V
100Hz
15
400rpm
150Nm
66A
86.8%
0.65
The battery dc voltage (ES) is set at 12 V, the inverter

switching frequency (TS) at 9 kHz and the voltage (kv)
and current (ki) ripple factors are 5% and 1%,
respectively. Considering the electric machines rated
voltage and the batterys voltage the inverter must
provide a voltage gain (G) of 6. The modulation index M
is 0.621.
Based on this values the Z-network inductance and
capacity are determined using (3) and (4). The
calculated values are:
(6)
L 54.1H
C 3.9mF
(7)
The chosen control method is the constant boost

control method based on its advantages. The control
model of the system is presented in Fig. 2. It consists of
sine reference signal generators, a triangular carrier
generator and a series of signal comparison blocks used
to determine the control signals for the inverter switches.
The simulated reference signals and triangular
carrier, inverter output current and inverter output
voltage are presented in Fig. 3, Fig. 4 and Fig. 5.
18
__________________________________________________________________________________________________________
Figure 3. Reference signals and triangular carrier.
Z-source inverter can be used as the power converter for

an ISA system.
This paper first presents the ISA system
configuration followed by a description of the Z-source
inverter.
The Z-network components are rated and a
simulation is carried out in order to confirm the systems
expected behavior. The Z-source inverter is a suitable
candidate for driving the ISA in a mild HEV. By
combining the Z-source inverter with a Z-source dc-dc
converter the systems complexity is reduced and the
resulting converter can perform either as a buck or as a
boost converter.
ACKNOWLEDGMENT
Figure 4. Inverter output voltage.
This paper was supported by the project

"Improvement of the doctoral studies quality in
engineering science for development of the knowledge
based society-QDOC" contract no. POSDRU/107/1.5/
S/78534, project co-funded by the European Social Fund
through the Sectorial Operational Program Human
Resources 2007-2013.
REFERENCES
Figure 5. Inverter output current.
Fig. 3 shows the reference signal with the injected

third harmonic component and the triangular carrier
signal.
Fig. 4 and Fig. 5 present the inverter output voltage
and current waveforms. The peak output voltage is 104
V and the peak output current is 87 A.
V. CONCLUSIONS
The Z-source inverter is a novel inverter topology
that overcomes some of the limitations of conventional
inverters. Coupled with a Z-source dc-dc converter, the
[1] F. Z. Peng, Z-source inverter, IEEE Trans. Ind. Appl.,

vol. 39, no. 2, pp. 504-510, Mar./Apr. 2003.
[2] W. Reik, Electrical Motor in the Drive Train, LuK
GmbH & Co. KG technical presentation, pp. 3-8, Apr.
2004.
[3] F. Z. Peng, M. Shen, and Z. Qian, Maximum boost
control of the Z-source inverter, IEEE Trans. Power
Electron., vol. 20, no. 4, pp. 833-838, Jul./Aug. 2005.
[4] M. Shen, J. Wang, et al., Constant boost control of the
Z-source inverter to minimize current ripple and voltage
stress, IEEE Trans. Ind. Appl., vol. 42, no. 3, pp. 770-777,
May/Jun. 2006.
[5] S. Rajakaruna, and B. Zhang, Design and control of a
bidirectional Z-source inverter, in Proc. IEEE Power Eng.
Conf., 2009.
[6] X. Fang, A novel Z-source DC-DC converter, in Proc.
IEEE International conference on Industrial Technology
2008, pp. 1-4, Apr. 2008.
[7] F. Jurca, C. Martis, D. Fodorean, -Analysis of a radial
flux synchronous machine with outer rotor for integrated
starter-alternator, International Symposium on Power
Electronics, Electrical Drives, Automation and Motion,
SPEEDAM, Sorrento 2012, pp. 496-500, ISBN: 978-14673-1299-8.

19
__________________________________________________________________________________________________________
Bearing Faults Condition Monitoring A Literature Survey

HARLICA Ciprian, SZAB Lornd
Department of Electrical Machines and Drives, Technical University of Cluj-Napoca
400114 Cluj-Napoca, Memorandumului 28, Romania; e-mail: Ciprian.Harlisca@mae.utcluj.ro
Abstract Bearing related faults are one of the most

common causes of failure in electrical machines. By
means of advanced diagnosis methods it is possible to
detect these faults in their incipient phase, before the
catastrophic effects of the failures can occur. The aim
of this paper is to make a brief survey of the condition
monitoring techniques used in the field of bearing
fault diagnosis.
Keywords: bearing faults, diagnosis, induction
machines, fault detection, condition monitoring.
I. INTRODUCTION
In most industrial processes unplanned stops due to
failures have a high economic impact on the cost of the
process and it may result in significant process down
time. The fault can occur in any part of the machine or
even in the drive system. The faults of electrical
machine could be electrical or mechanical. The main
electrical faults are the stator and rotor windings (or
cage) faults [1], [2]. Mechanical faults include bearing
faults, air-gap eccentricity, misalignment, gearboxes
faults, etc.
The research on fault diagnosis has shown that the
most of the failures of induction machines (about 40%)
are related to the bearings [3]. The bearing related faults
do not cause immediate breakdown, they evolve in time
until they produce a critical failure of the machine.
Unfortunately these failures results both in costly repair
and downtime.
The bearings faults can be caused by material
fatigue, overheating, harsh environments, inadequate
storage, contamination, corrosion, wrong handling and
installation, etc. But the main cause of their failure is
due to poor lubrication, which can be easily avoided by
a correct maintenance plan.
Vibration based monitoring techniques are usually
applied for the diagnosis of the bearings. Unfortunately
these methods require vibration sensors and special
equipment for the condition monitoring. They also need
access to the machine under testing, which is not always
possible.
Compared to the methods above, the current
monitoring requires only (frequently already existing)
simple and cheap current sensors.
The current monitoring based techniques can be used
to detect a large number of faults: broken rotor bars [4]
[5], shorted windings, air-gap eccentricity [6], bearing
faults [7], load faults, etc.
These methods are non-intrusive and can be applied

both on-line and in a remotely controlled way.
II. BEARING FAULTS
A rolling-element bearing is generally composed of
two rings, between which a set of balls or rollers rotate
in raceways. In most cases, bearing failures are the result
of material fatigue of the bearing. Under normal
operating conditions fatigue failure begins with small
cracks, located inside the surfaces of the raceway and
rolling elements.
The repetitive impacts between the components of
the bearing and the faulted surfaces cause the cracks to
gradually propagate and expand, generating an increase
in vibrations and noise levels [8].
The repetitive stressing of the damaged area causes
the detachment of some small fragments of the material,
which produce a phenomenon known as flaking or
spalling [9]
The pattern of the vibration signal consists in a
succession of oscillations which repeat with each pass of
a moving component over the fault [10]. The repetition
frequency of the impact depends on the position of the
fault. The fault can be on the inner race, the outer race or
on the rolling element
The typical construction and sizes of a ball bearing is
shown in Fig. 1. The balls are fixed and held together by
a cage which prevents the contact between the balls and
ensures a uniform distance between them.
Figure 1. Main bearing dimensions and

characteristic fault frequencies [1].
In the literature the bearing faults are classified

according to:
20
__________________________________________________________________________________________________________
the location of the fault: inner race, outer race,

balls, and cage [7].
the fault signature: single-point defects and
generalized roughness [11].
A. SINGLE-POINT DEFECTS
A single-point defect can produce different
characteristic fault frequencies in the vibration spectrum
of the machine. These frequencies are predictable and
depend on the surface of the bearing which contains the
fault [12].
The single-point defects cause periodic impulses in
vibration signals. Amplitude and period of these
impulses are determined by shaft rotational speed, fault
location and bearing dimensions. Therefore, a specific
frequency can be attributed to each component of the
bearing [13].
The fundamental cage frequency is given by:
f
d
f c r 1 cos
2 D
D
d2
f r 1 2 cos 2
d D
f id n f r f c
nf r
2
1 cos
D
(2)
(3)
The formula for computing the outer race defect

frequencies is the following:
f od nf c
nf r
d
1 cos
2d D
(4)
where, fr is the rotor speed, n the number of balls, d the

diameter of the ball, D the pitch diameter of the bearing
and the contact angle as shown in Fig. 1. The typical
value of the contact angle is 0.
For most bearings with six to twelve balls, the
frequencies given by (3) and (4) can be approximated
with [13]:
f id 0.6 n f r
(5)
f od 0.4 n f r
(6)
It is known that any air-gap eccentricity produces

anomalies in the air-gap flux density, which is reflected
on the stator current. In the case of a bearing fault the
characteristic fault frequencies are modulated by the
electrical supply frequency at a predictable frequency
[13].
f bng f s m f V
Generalized roughness fault is the most frequent

cause of bearing failure. It usually occurs in the
industrial environment due to various causes such as
[12]:
Lack or loss of the lubricant, contamination of
lubricant
Misalignment
Shaft currents
Environmental conditions (dust, water, acid and
humidity).
Bearing corrosion, produced by the presence of
water and acids.
These causes lead to a faster wear of components of
the bearing, especially raceways and balls. They to
produce generalized roughness fault, as well as singlepoint defects. A generalized roughness fault of a bearing
can be easily determined because it spins roughly or
with some difficulty.
(1)
The ball defect, respectively the inner race defect

frequencies can be computed by using the following
equations:
f bd
B. GENERALIZED ROUGHNESS FAULT
(7)
where fs is the electrical supply frequency, fV one of the

fault frequencies defined by equations (1)(4) and
m = 1, 2, 3...
III. CONDITION MONITORING

OF BEARING FAULTS
A significant part of the papers on the fault diagnosis
of induction machines are dealing with on the faults of
rolling bearings.
Even though that vibration based condition
monitoring techniques are usually applied for the
diagnosis of the bearings, many papers use the stator
current analysis, due to its advantages.
The methods used for stator current analysis
decompose and analyze the signal using various
techniques such as Fourier analysis, neural networks,
wavelets, statistical analysis, etc.
In [14] the authors analyze two types of bearing
faults: a hole drilled into the outer raceway and an
indentation produced in the inner and outer surface.
Vibration and current analysis is applied to both faulty
conditions.
The specific characteristic fault frequencies are
highlighted for both faults. The analysis of the first fault
shows two components, fod and 2fod in the vibration
spectrum, and |fs fod| and |fs 2fod| in the current
spectrum. For the second type of fault the highlighted
characteristics are fod, 2fod and fid for the vibration
spectrum, and |fs fod|, |fs 2fod| and |fs fid|, for the
current spectrum.
In [27] the authors introduce a new formulation for

the current spectral analysis for the detection of bearing
failures in induction motors driven by frequency power
converters. The fault is an outer race defect and the
authors highlight an increase of the specific fault
frequencies components of the current spectrum.
In [15] two inner raceway faults (drilled hole and
spalls) are studied using vibration and current analysis.
The results show that the fault frequencies are clearly
visible only in the vibration spectrum. The authors state
that the assembling, disassembling, remounting and

21
__________________________________________________________________________________________________________
realigning the test motor can alter the vibration and

current spectra.
Figure 2. Example of artificially drilled holes

in the outer raceway of a bearing [16].
Some authors studied the detection of faults in

electrical machines by using the stray flux around the
motor [17], [18], [19], [20], [21], [22].
In [20] a small active area sensor was used for the
detection of eccentricity and bearing faults. For the
bearing fault analysis, a hole was drilled in the inner
race of the bearing. Current and stray flux measurements
were effectuated under different loading conditions. The
spectra of the signals were obtained using Fast Fourier
Transform (FFT). The characteristic fault frequencies
were almost the same in both spectra, current and flux,
but the amplitudes of these components were low in
both cases.
In [16] the Park's Vector Approach is used for the
detection of broken bearings. The analysis is made on a
four bearings with drilled holes in the inner race and
outer race. One of the bearings has two drilled holes in
the outer race. The authors concluded that the diagnosis
of the inner race faults is more difficult because of
vibration signal is weak and it is not fully transmitted to
the outer race. The results show a good detection of the
faulty conditions.
The Park's Vector Approach was also used in [23]
for the diagnosis of three bearings with different
diameter holes in the outer race. The proposed method
showed good results in detecting even an incipient fault
can be detected using this method. This paper also
presents a new technology for artificially introducing
bearing faults such as: pitting, flutting or false
brinelling. The method consists in removing the pins of
the cage, so all the bearing components can be
accessible.
In [24] the Continuous wavelet transform (CWT) is
used for the extraction of characteristic features from
vibration signals measured for induction machines
subjected to bearing fluting. The faults of the bearing
were obtained artificially by using an Electrical
Discharge Machining (EDM) and thermal ageing. The
proposed method was compared with Short-Time
Fourier Transform and it was highlighted that the CWT
has some advantages. By using the CWT the authors
were able extract small amplitudes that cannot be
observed along the frequency axis. Also, they found
extra amplitudes caused by the damage of the bearing

between 24 kHz.
In [25, 26] the broken bar and bearing faults (inner
race defect) of several inverter-fed induction machines
are studied with a new hybrid algorithm that combines
the analysis of the signal in time and frequency domain.
This new method uses a combination between
Independent Component Analysis (ICA) and FFT in
order to analyze features of the stator current. ICA is a
statistical technique for decomposing a complex dataset
into independent subparts. The authors state that
proposed method detects and classifies correctly the
characteristic fault frequency components. In the case of
bearing faults, the detection is more difficult. It is shown
that the predominant characteristic fault frequency is
given by fid
For bearing fault diagnosis other authors have used
different methods such as: neural networks [28], hidden
Markov modeling [29], instantaneous power factor [30],
etc.
IV. CONCLUSIONS
The literature reviewed in this paper aims to
investigate the possibility of employing the analysis of
the stray flux and stator current for bearing fault
detection of induction machines in future papers.
ACKNOWLEDGMENT
This paper was supported by the project
"Improvement of the doctoral studies quality in
engineering science for development of the knowledge
based society-QDOC" contract no. POSDRU/107/1.5/
S/78534, project co-funded by the European Social Fund
through the Sectorial Operational Program Human
Resources 2007-2013.
REFERENCES
[1] W.T. Thomson, "A review of on-line condition
monitoring techniques for three-phase squirrel-cage
induction motors Past present and future," Proceedings
of the IEEE Symposium on Diagnostics for Electrical
Machines,
Power
Electronics
and
Drives
(SDEMPED '99), Gijon (Spain), pp. 3-18, 1999.
[2] M.E.H. Benbouzid, "A review of induction motors
signature analysis as a medium for faults detection,"
IEEE Transactions on Industrial Electronics, vol. 47, no. 5
(October 2000), pp. 984-993, 2000.
[3] Motor Reliability Working Group, "Report of large motor
reliability survey of industrial and commercial
installations Part I and II," IEEE Transactions on
Industry Applications, vol. IA21, no. 4 (July 1985),
pp. 853-872, 1985.
[4] F. Filippetti, G. Franceschini, C. Tassoni, "Neural
networks aided online diagnostics of induction motor
rotor faults," IEEE Transactions on Industry
Applications, vol. 31, no. 4 (July-August 1995),
pp. 892-899, 1995.
[5] A. Bellini, C. Concari, G. Franceschini, E. Lorenzani,
C. Tassoni, A. Toscani, "Thorough understanding and
22
__________________________________________________________________________________________________________
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
experimental validation of current sideband components

in induction machines rotor monitoring," Proceedings of
the 32nd Annual IEEE Conference on Industrial
Electronics
(IECON '2006),
Paris
(France),
pp. 4957-4962, 2006.
D.G. Dorrell, W.T. Thomson, "Analysis of airgap flux,
current, and vibration signals as a function of the
combination of static and dynamic airgap eccentricity in
3-phase induction motors," IEEE Transactions on
Industry Applications, vol. 33, no. 1 (January-February
1997), pp. 24-34, 1997.
R.R. Schoen, T.G. Habetler, F. Kamran, R.G. Bartheld,
"Motor bearing damage detection using stator current
monitoring," IEEE Transactions on Industry Applications,
vol. 31,
no. 6
(November-December
1995),
pp. 1274-1279, 1995.
H.A. Toliyat, T.A. Lipo, J.C. White, "Analysis of
concentrated winding induction machine for adjustable
speed drive applications- Part 1 (Motor Analysis)," IEEE
Transactions on Energy Conversion, vol. 6, no. 4
(December 1991), pp. 679683, 1991.
H.A. Toliyat, T.A. Lipo, "Transient analysis of cage
induction machines under stator, rotor bar and end ring
faults," IEEE Transactions on Energy Conversion,
vol. 10, no. 2 (June 1994), pp. 241247, 1994.
J. Antoni, R.B. Randall, "On the use of the cyclic power
spectrum in rolling element bearings diagnostics,"
Journal of Sound and Vibration, vol. 281, no. 1-2 (March
2005), pp. 463-468, 2005.
J.R. Stack,
T.G. Habetler,
R.G. Harley,
"Fault
classification and fault signature production for rolling
element bearings in electric machines," IEEE
Transactions on Industry Applications, vol. 40, no. 3
(May-June 2004), pp. 735-739, 2004.
M.D. Prieto, J.C. Roura, J.L. Romeral Martnez,
"Bearings Fault Detection Using Inference Tools," in
"Vibration Analysis and Control New Trends and
Developments," ed. Beltrn-Carbajal, F., InTech, Rijeka
(Croatia), 2011.
N. Tandon, A. Choudhury, "A review of vibration and
acoustic measurement methods for detection of defects in
rolling element bearings," Journal of Tribology, vol. 32,
no. 8 (August 1999), pp. 469-480, 1999.
R.R. Schoen, T.G. Habetler, F. Kamran, R.G. Bartheld,
"Motor bearing damage detection using stator current
monitoring," IEEE Transactions on Industry Applications,
vol. 31,
no. 6
(November-December
1995),
pp. 1274-1279, 1995.
R.R. Obaid, T.G. Habetler, J.R. Stack, "Stator current
analysis for bearing damage detection in induction
motors," Proceedings of the 4th IEEE International
Symposium on Diagnostics for Electric Machines, Power
Electronics and Drives (SDEMPED '2003), Atlanta
(USA), pp. 182-187, 2003.
J. Zarei, J Poshtan, "An advanced Park's Vector approach
for bearing fault detection," Proceedings of the IEEE
International Conference on Industrial Technology
(ICIT '2006), Mumbai (India), pp. 1472-1479, 2006.
H. Henao, C. Demian, G.A. Capolino, "A frequencydomain detection of stator winding faults in induction
machines using an external flux sensor," IEEE
Transactions on Industry Applications, vol. 39, no. 5
(September-October 2003), pp. 1511-1516, 2002.
H. Henao, G.A. Capolino, C. Mari, "On the stray flux
analysis for the detection of the three-phase induction
machine faults," Conference Record of the Industry
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
Applications Conference IAS '2003 (38th IAS Annual

Meeting), vol. 2, pp. 13681373, 2003.
A. Yazidi,
H. Henao,
G.A. Capolino,
M. Artioli,
F. Filippetti, D. Casadei, "Flux signature analysis: An
alternative method for the fault diagnosis of induction
machines," Proceedings of the 2005 IEEE Russia Power
Tech, St. Petersburg (Russia), pp. 1-6, 2005.
S.M.J. Rastegar Fatemi, H. Henao, G.A. Capolino,
"Gearbox monitoring by using the stray flux in induction
machine based electromechanical system," Proceedings
of the14th IEEE Mediterranean Electrotechnical
Conference (MELECON '2008), Ajaccio (France),
pp. 484-489, 2008.
S.-B. Han, D.-H. Hwang, S.-H. Yi, D.-S. Kang,
"Development of diagnosis algorithm for induction motor
using flux sensor," Proceedings of the International
Conference on Condition Monitoring and Diagnosis
(CMD '2008), Beijing (China), pp 140-142, 2008.
O. Vitek, M. Janda, V. Hajek, P. Bauer, "Detection of
eccentricity and bearing faults using stray flux
monitoring," Proceedings of the IEEE International
Symposium on Diagnosis for Electrical Machines, Power
Electronics & Drives (SDEMPED '2011), Bologna (Italy),
pp. 456-461, 2011.
J.L.H. Silva, A.J.M. Cardoso, "Bearing failures diagnosis
in three-phase induction motors by extended Park's
Vector approach," Proceedings of the 31st Annual
Conference of IEEE Industrial Electronics Society
(IECON '2005), Raleigh (USA), 2005.
E. Ayaz, A. Ozturk, S. Seker, "Continuous Wavelet
Transform for bearing damage detection in electric
motors," Proceedings of the IEEE Mediterranean
Electrotechnical Conference (MELECON '2006), Malaga
(Spain), pp. 1130-1133, 2006.
T.W. Chua, W.W. Tan, Z.-X. Wang, C.S. Chang, "Hybrid
time-frequency domain analysis for inverter fed induction
motor fault detection," Proceedings of the IEEE
International Symposium on Industrial Electronics
(ISIE '2010), Bari (Italy), pp. 1633-1638, 2010.
Z. Wang, C.S. Chang, Z. Yifan, "A feature based
frequency domain analysis for fault detection of induction
motors," Proceedings of the 6th IEEE Conference on
Industrial Electronics and Applications (ICIEA '2011),
Beijing (China), pp. 27-32, 2011.
O. Duque, M. Prez, D. Morigo, "Detection of bearing
faults in cage induction motors fed by frequency
converter using spectral analysis of line current,"
Proceedings of the IEEE International Conference on
Electric Machines and Drives (IEMDC "2005), San
Antonio (USA), pp. 17-22, 2005.
M. Torbatian, M.H. Kahaei, J. Poshtan, "Bearing fault
detection using level-dependent noise reduction
technique," Proceedings of the 3rd IEEE Symposium on
Signal Processing and Information Technology (ISSPIT),
pp. 146-149, 2003
H. Ocak, K.A. Loparo, "A new bearing fault detection
and diagnosis scheme based on hidden Markov modeling
of viibration signal," Proceedings of the IEEE
International Conference on Acoustics, Speech and Signal
Processing (ICASSP '2001), Salt Lake City (USA), vol. 5,
pp. 3141-3144, 2001.
A Ibrahim, M. El Badaoui, F. Guillet, F. Bonnardot, "A
new bearing detection method in induction machines
based on Instantaneous Power Factor," IEEE
Transactions on Industrial Electronics, vol. 55, no. 12
(December 2008), pp. 4252-4259, 2008.

23
__________________________________________________________________________________________________________
Design and Implementation of a PWM Inverter for Reluctance

Motors
MARGINEAN Calin1, MARGINEAN Ana-Maria1, VESE Ioana1, TRIFA Viorel1, TRIFU
Emil2
1
Technical University of Cluj Napoca, Romania,
Department of Electrical Machines and Drives, Faculty of Electrical Engineering,
str. Memorandumului nr.28, 400114 Cluj Napoca, Romania, E-Mail:
ignatc@edr.utcluj.ro, anamaria.marginean@gmail.com, ioana.vese@edr.utcluj.ro, trifa@edr.utcluj.ro
2
S.C. TRAMBUS S.R.L., Cluj-Napoca, Romania, E-Mail: office@trambus.ro
Abstract The paper deals with aspects regarding the

simulation, implementation and testing of a PWM
inverter for reluctance motors. The main objectives is
to present the simulation stage using the SLPS
interface between Orcad PSpice electric circuit
simulator and Matlab Simulink system simulator.
SLPS interface developed by Cybernet Systems offer
the possibility to integrate the real outputs of an
circuit obtained with PSpice simulator and the ideal
model or mathematical Simulink model, thus enabling
the designers to identify and correct integration issues
of electronics within a system[3].
Keywords: reluctance motors, PWM inverter, Orcad
PSpice, Matlab Simulink, SLPS interface.
with a major problem of their supplying and driving

systems, which focused much interest among specialists
in the last decade.
The VRSM used in our study is a common 8 pole, 4
phase motor with the following main characteristics:
- electromagnetic peak torque = 2Nm;
- phase current = 5 A;
- phase voltage = 60 V;
- step angle = 2.650 (136 steps/rot).
Figure 1 shows the electromagnetic structure of the
used VRSM stator.
Each phase is built from two diametrically opposite
poles windings, in such a manner that each phase has
two ends, available for various connecting techniques in
PWM inverters.
I. INTRODUCTION
The advance of incremental motion control systems,
where one usually uses stepping motors, has been
enforced by the multiplicity of their utilization in digital
controlled machine-tools drives, peripheral computer
equipments, telecommunications through laser and
satellites, nuclear techniques, industrial robots,
aeronautical and military equipments etc. In this context,
the VRSMs promise the low-cost production and
motivate the comprehensive research and design
although they are not included in the classical treatment
of the DC or AC electrical drives.
Among reluctant motors, variable reluctance
stepping motors (VRSM) and switched reluctance
motors (SRM) are the most popular. VRSM, which is
very representative as reluctant motors, serves especially
in digitally controlled open-loop positioning servosystems and is very suitable for board instrumentations.
SRM has imposed itself in the last years, especially in
variable speed applications, due to the simple
technology involved. It is used especially in speed
servo-systems and operates with self-commutation of
their phases.
Despite of their excellent robustness as actuators in
special applications, both types of motors are confronted
Figure 1. Structure of the VRSM stator.
The first step in PWM inverter design for the VRSM

motor was the simulation using Matlab Simulink system
simulator. The Simulink model for reluctance motor
which also contain the PWM inverter is presented in
figure 2[1].
The signals from the sequencer are presented in figure
3 and in figure 4 waveforms for current and voltage are
presented for phase 1.
24
__________________________________________________________________________________________________________
The Orcad model for the pulse generator and

sequencer is presented in figure 5.
Figure 2. Matlab/Simulink model for system simulation.
Figure 5. Orcad model of pulse generator and sequencer.
For pulse generation, an 555 timer IC was used in

astable mode, and the sequencer is based on 7473 dual
master-slave J-K flip-flop.
Signals from the pulse generator and sequencer
obtain with Pspice simulator are presented in figure 6.
Figure 3. Signals from sequencer.
Figure 6. Signals from pulse generator and sequencer in PSpice.
Figure 4. Current and voltage for phase1.
Starting from this Simulink model first step was to

replace the subsystem composed of pulse generator and
the sequencer with an Orcad model.
The pulse generator and the sequencer simulated in

Pspice were integrated in Simulink model using the
SLPS interface between Orcad PSpice electric circuit
simulator and Matlab Simulink system simulator.
SLPS interface developed by Cybernet Systems
offer the possibility to integrate the real outputs of an
circuit obtained with PSpice simulator and the ideal
model or mathematical Simulink model, thus enabling
the designers to identify and correct integration issues of
electronics within a system[3].
The Simulink model with SLPS interface is
presented in figure 7. The signal from the sequencer and
the current and voltage waveform for phase 1 are
presented in figure 8 and figure 9.
For pulse generation, an 555 timer IC was used in

astable mode, and the sequencer is based on 7473 dual
master-slave J-K flip-flop.
equation used in the bioreactor model represents the
Monod modified equations based on Michaelis-Menten
kinetics, proposed by Aiba et al., and described by Z.K.
Nagy[2].
The mass balances for the biomass is expressed by
equation (1) as:
Figure 7. Simulink model with SLPS interface.

25
__________________________________________________________________________________________________________
As phase currents are unipolar in case of the

proposed motor, MOS half-bridge inverter schemes is
chosen. Each bridge is controlled by four high/low side
drivers for MOS transistors. As MOS drivers IR2110
from International Rectifier has been used.
Figure 11 presents the Orcad Capture model for one
phase, and the simulated results with PSpice are
presented in figure 12
Figure 8. Signals from sequencer using SLPS interface.
Figure 11. Orcad model of the inverter for one phase.
Figure 9. Current and voltage for phase 1 with SLPS interface.
III. PWM INVERTER

Usually reluctant motors are supplied by voltage
source (series resistance switches or dual voltage
schemes) or by current sources (PWM schemes). PWM
schemes are the most popular inverters in case of
reluctant motors [8] due to their adaptability to various
techniques based on voltage and current processing.
Figure 10 shows the block diagram of the designed
inverter [2].
PWM inverters for motor electronic supply are
dependent on the motor type. Usually, in case of PM
synchronous motors and Brushless DC motors, bipolar
current inverters are needed, so full-bridge inverters
have to be used. In case of switched reluctance motors,
unipolar inverters lead to cheaper half-bridge
inverters.[2]
Figure 12. Pspice simulation results.
After the simulation, the pulse generator, the

sequencer and the inverter was practically realized and
tested.
IV. EXPERIMENTAL RESULTS
In order to test the equipment, an experimental test
bench was made based on DSpace DS1104 controller
board. The test bench is presented in figure 13.
Figure 10. Block diagram of the inverter.
26
__________________________________________________________________________________________________________
Figure 14 and figure 15 presents the pulse generator

and sequencer, respectively a module of the PWM
inverter.
Figure 13. Experimental test bench.

Figure 16. ControlDesk virtual control panel for current
acquisition.
V. CONCLUSIONS
Figure 14. Pulse generator and sequencer.
An PWM inverter was designed and simulated using

two most common software products in the industry
Matlab/Simulink and Orcad/PSpice. The SLPS interface
between this two well known software environments
offers the possibility to integrate the real outputs of an
circuit with the ideal model provided by Matlab
Simulink.
The interface actually supports the substitution of a
Simulink block with an equivalent PSpice electrical
circuit.[3]
Also an experimental test bench was made in order
to test the real inverter. The results are very good and so
we can conclude that the inverter works properly and
could be used as a premise to develop new research for
upgrading motor performance up to general purpose
electrical drives requirements.
REFERENCES
Figure 15. A module of the made PWM inverter.
Comparing figure 10 and figure 15 one can see that

shunt resistors Rsh13 and Rsh24 were replaced with LTSR
6-NP, closed loop (compensated) multirange current
transducer using the Hall effect, produced by LEM.
The current feedback loop is based on dedicated
LM555 modules, which are highly stable devices for
generating accurate time delays or oscillation and
incorporate Trigger Schmitt circuits.
The signals provided by the current transducers LTSR
6-NP are used by this current feedback loop board in
order to perform the PWM operation.
Figure 16 presents the virtual control panel for current
acquisition made with ControlDesk software from
DSpace.
[1] V. Trifa, C. Marginean, L. Zarnescu, Investigation of

variable reluctance stepping motors dynamics using
Matlab-Simulink environment, Proceedings of the 7th
International Conference on Development and
Application Systems, ISBN 973-666-106-7, 27-29 May
2004, Suceava, pp. 164-166.
[2] V. Trifa, C. Marginean, E. Trifu, Contributions regarding
the development of a light urban transportation vehiclemotor and PWM inverter design, Proceedings of OPTIM
2008 Conference, Brasov, 22-24 may 2008, vol. II-A,
ISBN 978-973-131-030-5, pp. 307-312.
[3] Saied Moslehpour, Ercan K. Kulcu and Hisham Alnajjar,
Model-Based Control Design Using SLPS Simulink
PSpice Interface, Journal of Communication and
Computer, May 2010, Volume 7, No.5, ISSN 1548-7709,
USA
[4] PSpice SLPS Interface version 2.5 Users Guide,
Cybernet Systems Co Ltd., 2004-2004.
[5] PSpice User Guide, Cadence Design Systems, 2003.
[6] ***IR2110 Datasheet, International Rectifier
[7] ***Dspace DS1104 R&D Controller Board Hardware
installation and configuration, December 2006.
[8] ***Dspace ControlDesk Experiment Guide, March 2007.

27
__________________________________________________________________________________________________________
Microstructure Development by Controlling Grain Size

RAMAKRISHNAN Sumathi1, MAHALINGAM Usha2
1
Department of Electronics and Communication Engineering

Pavai College of Technology, Namakkal-637 018, Tamil Nadu, India
Email ID: elakkiyasumi@gmail.com
2
Department of Computer Science and Engineering

Sona College of Technology, Salem-636 005, Tamil Nadu, India
Email ID: ushaanu@ieee.org
Abstract - In this Paper, a classical PID controller and

an adaptive fuzzy logic controller are employed for
controlling the microstructure development during hot
working process. The strength of any material is
dependent on the grain size of that material [4], [9].
The strength the material is increased when its grain
size is reduced. In this paper, the standard
Arrehenious equation of 0.3% carbon steel is utilized
to obtain an optimal deformation path such that the
grain size of the product should be 26m. The 0.3%
carbon steel improves in the machinability by heat
treatment [8]. It must also be noted that this steel is
especially adaptable for machining or forging and
where surface hardness is desirable. The plant model
is developed with grain size. The effect of process
control parameters such as strain, strain rate, and
temperature on important microstructural features can
be systematically formulated and then solved as an
optimal control problem. These approaches are
applied to obtain the desired grain size of 26m from
an initial grain size of 180m. The simulation is done
on various grain sizes using both the controllers by
MATLAB simulink toolbox. When comparing the
responses, the PID controller provides better
performance compared with fuzzy controller.
Resulting tabulated performance indices showed a
considerable improvement in settling time besides
reducing steady state error.
Keywords Carbon steel, strain, strain rate,
Temperature, PID Controller, Fuzzy Logic Controller.
to process engineers who are faced with smaller lot

sizes, higher yield requirements, and superior quality
standards. Therefore, it is important to develop new
systematic methodologies for process design and control
based upon scientific principles,
which sufficiently
consider the behavior of work piece material and the
mechanics of the manufacturing process. A new strategy
for systematically calculating near optimal control
parameters for control of microstructure during hot
deformation processes has been developed based on
optimal control theory [1]. This approach treats the
deforming material as a dynamical system explained
below.
II. STATIC AND DYNAMIC MODEL
The static model of 0.3% carbon steel [3] is,
Where,
The dynamic model of 0.3% carbon steel is obtained

by using the Arrehenius equation for changes in
temperature during hot extrusion is given below.
I. INTRODUCTION
The development of optimal design and control
methods for manufacturing processes is needed for
effectively reducing part cost, improving part delivery
schedules, and producing specified part quality on a
repeatable basis. Existing design methods are generally
ad hoc and lack adequate capabilities for finding
effective process parameters such as deformation rate,
die and work piece temperature, and tooling system
configuration. This situation presents major challenges
Where,
28
__________________________________________________________________________________________________________
The dynamic equation for grain size can be obtained

by differentiating the equation (1) with respect to
temperature and then multiplied by change in
temperature T, which follows that,
III.
The following simulations are done in order to

see the performance of the proposed PID
controller. The controller parameters are all
determined using trial and error method. The
Matlab/Simulink simulation model of the proposed
PID controller is shown in Figure 1.After several
trial and error runs, the controller parameters of the
classical PID controller are set to Kp=1, Ki=0.2,
and Kd=0.01 to provided the desired response.
OPEN LOOP MODEL
It is proposed to optimize the grain size of 26m

from the initial grain size of 180m. The
Matlab/Simulink simulation model for open loop system
is obtained from the equation (3). The steady state
operating ranges for the control parameters temperature,
strain, strain rate and the grain size are considered as,
Figure 1. Block Diagram to Optimize Grain Size by PID

Controller
V. FUZZY LOGIC CONTROL SYSTEM

IV. PID CONTROLLER
PID controllers are used extensively in the industry
as an all-in-all controller, mostly because it is an
intuitive control algorithm. A theoretical PID controller
[6] is
Where
u(t) = the input signal to the plant model
e(t) = the error signal is defined as e(t) = r(t) y(t)
and
r(t) = the reference input signal.
y(t) = plant output
Kp, Ti and Td are the proportional gain, integral time
and derivative time respectively.
The coefficients Kp, Ti, Td and P, I, D are related by:
P = Kp
I = Kp/Ti
D = KpTd
The controller has three parts: The proportional term
is providing an overall control action proportional to the
error signal through the all-pass gain factor. The integral
term is reducing steady-state errors through lowfrequency compensation by an integrator. The
derivative term is improving transient response through
high-frequency compensation by a differentiator.
It is a classical logic system which provides an

alternative way of thinking, which allows modeling of
complex systems using knowledge and experience of
operating a control system. It provides a simple way to
draw definite conclusions from vague or imprecise
information and resembles human decision-making in
its ability to work with of approximate data and find
precise control system solution. The concept of fuzzy
logic is not presented as a control methodology, but as a
way of processing data by allowing partial set
membership rather than crisp set membership or nonmembership. If feedback controllers programmed to
accept noisy, imprecise input, they would be much more
effective and perhaps easier to implement.
It lends itself to implementation in system ranging
from simple, small, embedded microcontrollers to large,
networked, multi-channel PC or workstation based data
acquisition and control system. It can be implemented in
software, hardware or a combination of both. Fuzzy
logic provides a simple way to arrive at a definite
conclusion based on vagueness, ambiguity, imprecision,
noise or missing input information. Approach to control
problems, using fuzzy logic, mimics how a human being
would take a decision, but at a faster rate.
The functional block diagram of fuzzy logic control
[7] system is illustrated in Figure 2. It includes four
major blocks, which are Fuzzification, Knowledge
Base, Inference Mechanism and Defuzzification.

29
__________________________________________________________________________________________________________
Figure 2. Functional Block Diagram of Fuzzy Logic Control System
A. Fuzzification:
The grain size control of the 0.3% carbon steel is
designed based on 2 inputs and 1 output. Inputs for
this controller are error and change in error and the
output is grain size. Fuzzification is where we define
the quantization and membership function for the
input and output variables in universe of discourse. It
involves the conversion of the input and output
signals into a number of fuzzy represented values
(fuzzy sets).
5.3 Output: It is quantized into three fuzzy sets as

negative output (N), zero output(Z) and positive
output(P). Figure 5 shows the change in error input
variable,
5.1 Error (e): It is quantized into three fuzzy sets as

negative error (N), zero error (Z) and positive error
(P). Figure 3 shows the error input variable,
Figure 5. Output Range
For grain size the range is chosen as,

(i) Error range = -83 to 83
(ii) Change in error range = -4 to 4
(iii) Output range = -1 to 1
B. Knowledge Based rule:
Figure 3. Error Range
5.2 Change in Error (ce): It is quantized into three

fuzzy sets as negative change in error(N), zero
change in error(Z) and positive change in error(P).
Figure 4 shows the change in error input variable,
Figure 4. Change in Error Range
Fuzzy logic uses a set of rules to define its

behavior. The rules define the conditions expected
and outcomes desired with if/then statements. These
rules replace formulas. They must cover all situations
that may occur but are not to be written for every
possible combination. The rules are expressed in
terms of linguistic or fuzzy variables which are
adjectives like large positive error, small positive
error, zero error, small negative error and large
negative error, which modify the variable. For
minimum of variables it can simply have positive,
zero and negative as variables for each of the control
parameters. For microstructure development, stain,
strain rate and temperature are the control
parameters. The fuzzy rule table for the triangular
membership function for grain size is formed as
shown in table 1.
30
__________________________________________________________________________________________________________
Figure 6. Block Diagram to Optimize Grain Size by FLC
Table 1. Fuzzy Rule Table
The Matlab/Simulink simulation model of the

proposed fuzzy logic controller is shown in Figure
6.After several trial and error runs, the controller has
the following parameter = 1.49, Ke=150 and Kp
= 0.1 to provided the desired response.
VI. SIMULATION AND ERROR CALCULATION
The process control parameters strain, strain rate
and temperature are optimized for a required grain
size of 26m from an initial grain size of 180m and
its corresponding trajectories are shown. The time
taken is in seconds.
A. PID Controller Result:
C. Inferencing:
The inference mechanism provides the mechanism
for invoking or referring to the rule base such that the
appropriate rules are fired. There are several
methodologies to derive the inferencing method.
Two most common methods used in FLC are the
max-min composition and max(algebraic) product
composition. The inference or firing with this fuzzy
relation is performed via the operations between the
fuzzified crisp input and fuzzy relation representing
the meaning of the overall set of rules. As a result of
composition, one obtains the fuzzy set describing the
fuzzy value of the overall control output. For this
system, max-min composition is used for the
inferencing.
The PID controller output for 26m, 30m and

35m are given in figure 7, figure 8 and figure 9
respectively.
Figure 7. Response for Grain Size of 26m
D. Defuzzification:
The function of defuzzification is scale mapping
which converts the range of values of output variable
into corresponding universe of discourse and it yields
a non-fuzzy (crisp) control action. For this system,
centroid method is used for defuzzification. It is
given by the algebraic expression,
z*
( z).z.dz
( z).dz
c

31
__________________________________________________________________________________________________________
C. Error:
The following Integral Square Error and Integral
Absolute Error values have been obtained using PID
and FLC.
Table 2. Error Comparison
Set
point
ISE
IAE
PID
FLC
PID
FLC
26m
301.7
27060
5.069
321.3
30m
282.7
24610
4.972
297.4
35m
260.6
21800
4.93
269.9
B. Fuzzy Logic Controller Result:

The fuzzy logic controller output for 26m, 30m
and 35m are given in fig. 10, fig. 11 and fig.12
respectively.
VII. CONCLUSION
The dynamic model for 0.3% carbon steel for

microstructure control is developed. The steady state
value for strain, strain rate and temperature to obtain
grain size from 180m to 26m are selected as 1, 1
and 1100 respectively. The dynamic model is
simulated by PID controller and fuzzy logic
controller separately to optimize grain size from
180m to 26m. In both the case, a simulation time
of 10 seconds is considered and the optimization is
done. The integral square error and integral absolute
error are also calculated. It is observed that the
settling time is less and also the ISE and IAE are
comparatively less in the case of PID.
The PID controller has only three parameters to
adjust. It is commonly used to regulate the time
domain behavior of dynamic system. Controlled
system shows good results in terms of response time
and precision when these parameters are adjusted
well.
A fuzzy logic controller has a lot of parameters to
adjust. The most important thing is to make a good
choice of rule base and parameters of membership
functions. It is sensitive to the distribution of
membership functions but not to the shape of the
membership functions. It doesnt have much better
characteristics in time domain. One of the most
important problem with FLC is that the computing
time much longer than the PID controller, because of
the complex operations as fuzzification and
particularly in defuzzification. In addition, with the
growing requirement of the system performance, the
fuzzy membership functions and the interference
rules becomes more and are complicated.
From these results, PID seems to be better choice
for optimization of process control parameters.
32
__________________________________________________________________________________________________________
REFERENCES
[1] D.E. Kirk: Optimal Control Theory: An Introduction,
Prentice-Hall
Inc., Englewood Cliffs, NJ, 1970,
pp. 29-46 and pp. 184-309.
[2] W.G.Frazier et.al.,Application of control theory
principles to optimization of grain size during hot
extrusion,Materials Science
and Technology, Vol
14, pp.25-31, Jan 1998.
[3] James C Malas et.al.,Optimization of microstructure
development
during hot working using control
theory,Metullurgical and Materials
Transactions
A, Vol 28 A, pp.1921-1930, Sep 1997.
[4]. L.W. Ma, X. Wu and K. Xia, Microstructure and
property of a medium carbon steel processed by equal
channel angular pressing, materials forum volume
32 2008, Edited by J.M. Cairney, S.P. Ringer and R.
Wuhrer Institute of
Materials Engineering
Australasia Ltd.
[5] S.Venugopal, Optimization of Microstructure

Development during
Deformation processing
Under Dynamic Conditions, Journal of Proc.of the
Workshop on Dynamic Processing of
materials,
(2000).
[6] Katsuhiko ogata, Modern Control Engineering: Third
edition, 1999,
Prentice-Hall of India Private
limited, New Delhi.
[7] Timothy J Ross, Fuzzy Logic with engineering
applications.
[8] Xu-yue YANG, Ze-sheng JI, H. MIURA and T.
SAKAI, Dynamic recrystallization and texture
development during hot deformation of magnesium
alloy AZ31, Volume 19, Issue 1, February 2009.
[9] Monika HRADILOVa,b, Frank MONTHEILLET,
Anna FRACZKIEWICZ, Christophe DESRAYAUD
and Pavel LEJEK, Microstructure development of
ca-doped mgzn alloy during hot deformation , 23.
25. 5. 2012, Brno, Czech Republic, EU.

33
__________________________________________________________________________________________________________
Progressive Wavelet Correlation as a Tool

for Recognition of the Images
STOJANOVIC Igor1, ZDRAVEV Zoran2, TASEVSKI Angel3
Goce Delcev University - Stip, Macedonia,
Department of Signal Processing and Communication Systems Group, Faculty of Computer Science,
2
Information Systems Department, Faculty of Computer Science,
Toso Arsov 14, 2000 Stip, Macedonia, E-Mail1: igor.stojanovik@ugd.edu.mk, E-Mail2: zoran,zdravev@ugd.edu.mk
1
Goce Delcev University - Stip, Macedonia,

Faculty of Mechanical Engineering,
St. Partizantska No. 1, Vinica, Macedonia, E-Mail: angel.tasevski@ugd.edu.mk
Abstract An algorithm for recognition and retrieval

of image from image collection is developed. Basis of
the algorithm is the progressive wavelet correlation.
The final result is the recognition and retrieval of the
wanted image, if it is in the image collection.
Instructions for the choice of correlation threshold
value for obtaining desired results are defined.
Keywords: discrete cosine transform; multiresolution;
progressive wavelet correlation; recognition; wavelets.
I. INTRODUCTION
Images, drawings, photographs as means of
communication among people, for sending and receiving
messages have been part of everyday life for a long a
time. The easy to use World Wide Web, the reduced
price of storage devices as well as the increased
calculating power allow an essential and efficient
management of large quantity of digital information. All
of these factors offer a number of possibilities to the
designers of real image-browsing and retrieval systems.
However, image digitalization systems do not allow
for an easy management of collections of images. A
particular form of cataloguing and indexing is still
required. The need for efficient storage and retrieval of
images was recognized by managers of large collections
of images long time ago and was studied at a workshop
sponsored by the American National Science
Foundation in 1992 [1]. Data representation, feature
extractions and indexing, image query matching and
user interfacing were identified as areas with a potential
for improvement. One of the issues identified was the
problem of locating a picture in a large and diverse
collection of images.
The earliest and the most sophisticated descriptorbased image recognition engine is IBM QBIC [2].
Another set of content-based tools for image recognition
and retrieving have also improved throughout the years.
Examples for such tools are VisualSEEk[3], WebSEEk
[4] and ImageRover [5]. ImageRover uses low
resolution for image representation in six regions in
order to cover particular information along with the

region based descriptors. The present commercially
available engines for image recognition, based on
descriptors, provide no assurance that the required
information from the libraries can be found. For some
applications such as for example, collections of medical
images or satellite images the even the smallest details
can be important. Descriptor based retrieval engines
cannot
meet
satisfactorily
such
requirement.
Reconstruction of details here happens only to a
particular level of resolution. In addition large images
can contain so many details which cannot be comprised
in any compact description.
An alternative approach to the problem set forth
above is pixel (elements of digitalized image) based
recognition and retrieval. This type of recognition
involves analysis of the image. Still this requires intense
computing especially when the image contains many
subtle details. Despite this fact, the existence of large
number of operations per image doesnt seriously
restrict the application of pixel-based recognition and
retrieval techniques, especially not when it comes to
research and experimental purposes.
Pixel-based techniques work by locating a particular
pattern in a given image library. Popular criteria for
matching are the normalized correlation coefficients [6],
which measure the differences between images and
patterns from the library. The particular strength of these
criteria is that they are insensitive to uniform differences
in brightness.
Some of the work done in the area of PWC
(Progressive Wavelet Correlation) [6] are outlined in
Section 2. Our proposal about applying of PWC for
recognition images stored in a database is presented in
Section 3. Results of experiments are presented in
Section 4.
II. PROGRESSIVE WAVELET CORRELATION
In this section we summarize the technique described
in [6], [7]. The fundamental operation for recognition is
34
__________________________________________________________________________________________________________
the circular correlation x y . The jth entry of the

circular correlation is defined as:
N 1
x y j xi j mod N yi ,
j = 0, 1, , N-1 (1)
i 0
FN x TN ,M , R H 1 FM , R Hx
x1 x N 1
x0
x
x2 x0
.
X 1

x N 1 x0 x N 2
FM , R
FM I R , that is it has R interlaced copies of transform

of size M. The matrix TN ,M , R is a Fourier update matrix
that transforms FM , R into FN : FN TN ,M ,R FM ,R .
Fourier-Wavelet Multiresolution Theorem:
(2)
FN / R, R H1 x
(7)
TN / R, N / R 2 , R I R U 2,11 FN / R 2 , R 2 H 2 x
The notation P i R denotes subsampling of P by

taking components whose indices are equal to i modulo
R. For example, if R = 4 and i = 2, the subsampling
operation extracts the elements 2, 6, 10, 14, from P.
Progressive wavelet correlation using Fourier methods is
based on four theorems: the Wavelet-Correlation
Theorem, the Fourier-Wavelet Correlation Theorem, the
Fourier-Wavelet Subband Theorem and the FourierWavelet Multiresolution Theorem. To simplify the
discussion all data are assumed to be one-dimensional
vectors.
Wavelet-Correlation Theorem:
R 1
x y 0R Hx k R Hy k R
(3)
k 0
where H is wavelet-packet transform. is the Kronecker product of I M and W, H I M W , where I M is

M M identity matrix and W is an R R matrix with
property W T W I R . The wavelet transform packet
matrix H has a special structure. H is block diagonal
with block size R. For instance, W can be 2 2 Haar
matrix:
1 1 1

W
.
2 1 1
(4)
Fourier-Wavelet Correlation Theorem:
x y 0R
R 1
(5)
FM1 FM Hx k R . * FM Hy k R
k 0
(6)
where N = MR. The matrix FM , R is an interlaced

Fourier transform matrix with structure
where x and y are column vectors of length N. The

matrix form is x y Xy , where X is left circulant
matrix generated by x:
Fourier-Wavelet Subband Theorem:
where FM is the Fourier transform matrix of dimension

M and FM is the complex conjugate of FM .
TN , N / R, R H11
FN x
N / R, N / R 2 , R
I R U 2,11 FN / R 2 , R 2 H 2 x
(8)
where N = MR2. H2 is a coarse transform matrix that is

block diagonal with block of size R2 with the structure
H 2 I N / R2 W1 W1 and operates on R2 subbands,
each of length N/R2. W1 is an R R wavelet filter matrix
with property W1T W1 I R . H1 is a fine transform matrix
that is block diagonal with block size N/R with structure
H1 I N / R W1 . There is an update matrix U2,1 that
refines H1 into H2, H2=U2,1H1. The matrix U2,1 is block
diagonal with block of size R2 with the following
structure:
U 2,1 I N / R2 W1 I R
(9)
III. METHOD FOR IMAGE RECOGNITION

JPEG compression is based on the discrete-cosine
transform (DCT) [6]. The matrix C8 is an 8 8 DCT
matrix that is used to create transforms of 8 8
subimages in a JPEG representation of an image.
The multiresolution recognition process relies on the
factorization of the DCT matrix C8 = V8,4V4,2V2, where
V2 and V4,2 are matrices built from Kronecker products
of W and the identity matrix.
The matrix V2 I 4 W consists of 4 interlaced
copies of W and is of size 8 8 . The matrix V4,2 has a
structure V4,2 I 2 W I 2 .
If we write C8 = V8,4V4,2V2 where V8,4 is a matrix
whose coefficients we want to compute, then
V8,4 C8V21V4,21
(10)
obtain the last expression by multiplying both sides by

V21 V4,21 . The matrix V8,4 satisfies equation V8, 4

35
__________________________________________________________________________________________________________
V W I 4 . The inverse of V is given by (11), where

m cos2m / 32 .
1
0
0
V 1
0
0
0
0
0
0
2 7 0 5 6
6 1 0 3 2
0
0
0
2 1 0 3 6
6
0
0
2
0
0
6 7 0 5 2
0
0
0
0
2 1
0 3 6
0
0 5 2 0 6 7
1
0
0
0
0 5 6
0 2 7
2
0
0
0
6
0
0
0
0 3 2
0 6 1
(11)
The matrix H is an N N matrix with the structure

I M C8 where N = 8M. It produces the JPEG transform
of a vectors of length N. Let x be image stored as a
JPEG transform of a vector Hx with an instance of a
pattern y with JPEG transform Hy. The algorithm
consists of three incremental steps, each of which
quadruples the number of correlation points. The three
incremental steps are:
1. Coarse correlation Generate the Fourier
transforms FM ,8 Hx and FM ,8 Hy . Multiply the
transforms point by point and partition them into
eight subbands of length M. Add these eights
vectors and take the inverse Fourier transform of the
sum. Every eighth point of the correlation is
generated.
2. Medium correlation Multiply FM ,8 Hx by
T2M ,M ,2 I 4 I M W I 4 V 1
I 4 I M W I 4 V 1 . Multiply the resulting vectors point by point and partition them in

four subband of length 2M. Add the subbands and
you will create a single vector of length 2M. Taking
the inverse Fourier transform of size 2M yields the
correlation at indices that are multiples of 4 mod 8
of the full correlation.
Fine correlation Multiply the x and y transform
from the preceding step by T4M ,2 M ,2 I 2
2 M ,M , 2
3.
and FM ,8 Hy by
I M V4,2 and T4M ,2M ,2 I 2 I M V4,2 , respec-
4.
tively. Multiply the resulting vectors point by point

and partition them in two subbands of length 4M.
Add the subbands and you will create a single
vector of length 4M. Take the inverse Fourier
transform of size 4M to obtain the correlation at
indices that are multiples of 2 mod 8 and 6 mod 8 of
the full correlation.
Full correlation Multiply the x and y transform
from the last step by I 8M ,4M ,2 I M V2 and
T8M ,4M ,2 I M V2 , respectively. Multiply the resulting vectors point by point and take the inverse
Fourier transform of size 8M to obtain the correlation at odd indices.
Fig. 1 is a flow diagram showing the steps performed

for an image recognition according to the PWC method
and it is in the Test block where one checks whether a
particular set of pixels of the given pattern y and image x
have similar values. The algorithm can be described as
follow:
Step 1: A candidate image is coarsely correlated
with the pattern. Every eighth point of the
correlation is generated.
Step 2: It is determined whether the pattern
suitably matches the candidate image. If not,
then another candidate image may be chosen or
the process abandoned.
Step 3: If the match was suitable, then the
candidate image is medium correlated with the
pattern. We obtain the correlation at indices that
are multiples of 4 mod 8 of the full correlation.
Step 4: Another similar match test is performed.
Step 5: A candidate image is fine correlated with
the pattern. Fine correlation means to obtain the
correlation at indices that are multiples of 2 mod
8 and 6 mod 8 of the full correlation.
Step 6: Another similar match test is performed.
Step 7: Full correlation: obtain the correlation at
odd indices.
Step 8: If a suitable match is found for the fully
correlated image, then the image recognised for
has been found.
When multiresolution analysis technique is applied
to JPEG representation of data, the lowest resolution
correlation can be performed directly on the JPEG
images to produce 1/64 of the correlation points. The
regonition consists of three incremental steps. The
process can be halted at any stage if the intermediate
results indicate that the correlation will not result in a
match.
We investigate what happens in the two-dimensional
case. Let the image size be N by N. In step 1, we have
64 subbands of length N2/64. We perform one step of the
inverse 2D JPEG transfer function, and one 2D step of
the forward Fourier transform function. The next step
includes adding the 64 subbands point by point to create
a 2D array of size N/8 by N/8. Taking the inverse
Fourier transform, we obtain the correlations at points
that lie on a grid that is coarser than the original pixel
grid by a factor of 8 in each dimension. In step 2, we
obtain 16 subbands of size N2/16 by adding the 16
subbands point by point, and taking the Fourier inverse.
We will obtain the correlation values on a grid that is
coarser than the original grid by a factor of 4 in each
dimension. In step 3, we obtain 4 subbands of size N2/4.
Finally, in step 4, the full resolution is obtained.
Formulas for calculating normalized correlation
coefficients that measure differences between images
and patterns are given in [6]. Normalized correlation
coefficients can be computed from the correlations
described above. The normalization is very important
because it allows for a threshold to be set. Such a
threshold is independent of the encoding of the images.
36
__________________________________________________________________________________________________________
Coarse-Correlated
Candidate
Match?
No
Medium-Correlated
Candidate
Match?
No
Fine-Correlated
Candidate
Match?
Abandon Search
No
Fully-Correlated
Candidate
Match?
No
Image Found
Fig. 1. Flow diagram of PWC method
The normalized correlation coefficient has a

maximum absolute value of 1. Correlations that have
absolute values above 0.9 are excellent, and almost
always indicate a match found. Correlations of 0.7 are
good matches. Correlations of 0.5 are usually fair or
poor. Correlations of 0.3 or less are very poor.
There is a tradeoff between the value of the threshold
and the likelihood of finding a relevant match. Higher
thresholds reduce the probability of finding something
that is of interest, but they also reduce the probability of
falsely matching something that is not of interest.
The progressive wavelet correlation provides
guidelines on how to locate an image in the image
library. To make this method practical, we must first
decide how to store the images. The initial choice is to
store them in a disk file system. This can be seen as the
quickest and simplest approach. A better alternative that
should be considered is to store those images in a
database.
Databases offer several strengths over traditional file
system storage, including manageability, security,
backup/recovery, extensibility, and flexibility.
We use the Oracle Database for investigation
purposes. There are two ways of storing an image into
the Oracle Database. The first one is the use of Large
Objects LOB, and the second one is the use of Oracle
interMedia. To store images into the database we use the
BLOB datatype. After creation of one BLOB column

defined table we also create a PL/SQL package with a
procedure for loading images (named load). This
procedure is used to store images into the database. The
implementation of the progressive wavelet correlation in
Matlab and the connection between the algorithm with
the database are the next steps. The Database Toolbox is
part of an extensive collection of toolboxes for use with
Matlab.
Before the Database Toolbox is connected to a
database, a data source must be set. A data source
consists of data for the toolbox to access, and
information about how to find the data, such as driver,
directory, server, or network names. Instructions for
setting up a data source depend on the type of database
driver, ODBC or JDBC. For testing purposes JDBC
drivers were usually used [8].
After setting up the data source for connecting to and
importing data from a database we have used several
standard functions of the Matlab Database Toolbox. We
can retrieve BINARY or OTHER Java SQL data types.
However, the data might require additional processing
once retrieved. For example, data can be retrieved from
a MAT-file or from an image file. Matlab cannot
process these data types directly. One needs knowledge
of the content and might need to massage the data in
order to work with it in Matlab, such as stripping off
leading entries added by the driver during data retrieval.
The last step in the adaptation is to create Matlab
applications that use the capabilities of the World Wide
Web to send data to Matlab for computation and to
display the results in a Web browser. In the simplest
configuration, a Web browser runs on a client
workstation, while Matlab, the Matlab Web Server
(matlabserver), and the Web server daemon (httpd) run
on another machine. In a more complex network, the
Web server daemon can run on a separate machine
[9,10].
The practical implementation of progressive wavelet
correlation includes two main subsystems: the server
sub-system and the client subsystem.
The server subsystem handles the processes of image
storing in a database and similarity measure. The client
subsystem handles the process of querying. In addition,
we present here the system architecture of the system
(Fig. 2).
Image
Query Interface
PL/SQL
Similarity Measure
Image Size
Coarse
Correlation
JDBC
Image Database
Medium
Correlation
Internet
HTTP
Intranet
Fine
Correlation
Full
Correlation
Threshold
Server
Extranet
Name of the image
Matlab Server
Server
Client
Fig. 2. System Architecture of PWC system

37
__________________________________________________________________________________________________________
The following query image procedure is proposed:
Step 1: The user presents the input parameters

of the query interface and sends the query
message.
Step2: HTTP Server receives the query
message and forwards it to the Matlab Server.
Step3: The similarity measure mechanism
measures the similarity between the query
image and the database images comparing the
different levels of correlation with the
introduced correlation threshold.
Step 4: The Matlab Server passes to the HTTP
Server all the images candidates whose
correlations along with the query image are
larger than the adopted correlation threshold.
Step5: HTTP Server sends the candidate
images of the query interface to the clients
site.
Step6: The user receives the candidate images
by means of the query interface.
IV. NUMERICAL RESULTS
This section represents experimental results obtained

by means of image retrieval through an algorithm of
progressive wavelet correlation.
Different experiments were set up as follows:
a) The required image is included several times in
the database with different names;
b) The image is included only once in the database;
c) Aside from the required image, the database also
contains an image very similar to the required
one (smudged in some parts or an image
generally slightly different);
d) The required image is not present in the database.
The experiments are carried out upon databases
including between 250 and 1000 store images. Oracle
10g version 10.1.0.2.0, served as our database, while we
used Matlab version is 7.0.4.365 (R14) Service Pack 2
for image recognition.
Two images, called flower01.jpg and flower10.jpg,
served as recognized targets. Image flower01.jpg
appears eight times under different names in the
database, while the image flower10.jpg appears six
times. The database contains images that are more
visually similar to the image flower10.jpg.
Evaluation of the quality of the system concerning its
precision p is estimated using the following the
definition:
| A(q) R(q) |
| A(q) |
In the following tables are given precision for

different correlation threshold values ranging from 0,2
to 0,7 with step 0,1. The number of images in the
database is 1000.
TABLE 1. flower01.jpg.
Threshold
0,2
0,3
0,4
0,5
0,6
0,7
-1
Retrieved
images
761
435
113
19
Precision
0,01
0,02
0,07
0,42
TABLE 2. flower10.jpg.
Threshold
0,2
0,3
0,4
0,5
0,6
0,7
-1
Retrieved
images
911
640
285
79
18
Precision
0,007
0,009
0,02
0,08
0,33
Considering the Tables 1 and 2, it can be concluded

that a high accuracy value can be obtained for
correlation threshold values greater than or equal to 0,7.
The results presented in the Table 3 and 4 refer to
two different images 21.jpg and 40.jpg. Each of these
images is included only once in the database.
TABLE 3. 21.jpg.
Threshold
0,2
0,3
0,4
0,5
0,6
0,7
-1
Retrieved
images
879
500
200
37
Precision
0,001
0,002
0,005
0,27
0,5
TABLE 4. 40.jpg.
Threshold
0,2
0,3
0,4
0,5
0,6
0,7
-1
Retrieved
images
854
541
174
27
Precision
0,001
0,002
0,006
0,37
0,5
The next example refers to image 181.jpg. This

example is specific because the database contains a
similar image 183.jpg with its vertical sides slightly
smudged. The similar images are shown in Fig. 3.
(12)
where q stands for query, R(q) signifies a set of relevant

images for the query in the database, while A(q) stands
for the set of images returned as a response to the set
query q.
181.jpg
183.jpg
Fig. 3. Two very similar images
38
__________________________________________________________________________________________________________
TABLE 5. 181.jpg.
Threshold
0,4
0,5
0,6
0,7
0,8
0,9
-1
Retrieved
images
332
77
Precision
0,006
0,03
0,67
It is evident from table that when correlation

threshold values are 0,7 and 0,8, both the images 181.jpg
and 183.jpg are retrieved. If the correlation threshold is
equal to or greater than 0,9 only image 181.jpg is
retrieved.
The table bellow gives the number of retrieved
images for different value of the correlation threshold
for image 50.jpg, which is not present in the database.
systems. We believe that in the coming years PWC

based methods will be able to achieve detailed analysis
of thousands of images per second.
As far as computing progressive wavelet correlation
is concerned, the question still to be answered is whether
there is an alternative factorization of DCT matrix which
will meet the structural requirements and will transfer as
many computations as possible from the first into the
second and third step in the algorithm.
According to the future researches, focused on the
construction of image recognition and retrieval system,
regardless of the technique that the system uses, only the
issue concerning system quality assessment from the
aspect of efficiency and applicability is to be tackled.
Retrieval systems should be comparable for the purpose
of identifying the good techniques [11].
REFERENCES
TABLE 6. 50.jpg.
Threshold
0,3
0,4
0,5
0,6
0,7
0,8
-1
Retrieved
images
719
412
119
14
For correlation threshold values greater than or equal

to 0,8 there arent any images retrieved from the
database. Therefore if the correlation threshold is set
sufficiently high, the system correctly predicts the
absence of the image.
IV. CONCLUSIONS
Based on our experience and experimental evidence
we conclude that PWC is a useful tool for image
recognition and retrieving from databases.
The main feature of PWC is its high accuracy. With
the choice of an adequate correlation threshold it is
possible to conclude whether the given image is present
in the database, whether there are images similar to the
required one with different names, whether there are
images slightly different from the required one, and
whether the required image is actually included in the
database.
The examples studied imply that with the selection
of a threshold greater than or equal to 0,7 it can be
established that the required image is included in the
database. Such selection of threshold gives the accurate
number of images identical to the required one
comprised in the database. With a correlation threshold
greater than or equal to 0,9 a slight difference between
two very similar images can be ascertained. In our
examples, when the minimal threshold value is 0,8 it is
established that the required image is not included in the
database.
Owing to the large number of operations, the pixelbased methods for image recognition and retrieval are
slow compared to commercially available content-based
[1] R. Jain, Workshop report: NSF Workshop on Visual

Information Management Systems in Storage and
Retrieval for Image and Video Databases (Niblack, W R
and Jain, R C, eds), Proc SPIE 1908, 198-218, 1993.
[2] M. Flickner, H. Sawhney, W. Niblack, et al, Query by
image and video content: The QBIC system, IEEE
Comp., vol. 28, pp. 23-32, Sept. 1995.
[3] J. R. Smith and S. F. Chang, Querying by color regions
using the VisualSEEk content-based visual query system,
Intelligent Multimedia Information Retrieval (Maybury,
MT, ed). AAAI Press, Menlo Park, CA, 23-41, 1997a.
[4] J. R. Smith and S. F. Chang, An image and video search
engine for the World-Wide Web in Storage and Retrieval
for Image and Video Databases V (Sethi, I K and Jain, R
C, eds), Proc SPIE 3022, 84-95, 1997b.
[5] S. Sclaroff, L. Taycher, and M. La Cascia, Imagerover:
A content-based image browser for the world wide web,
IEEE Wksp. Content-Based Access of Image and Video
Libraries, pp. 29, June 1997.
[6] H. S. Stone, Progressive Wavelet Correlation Using
Fourier Methods, IEEE Trans Signal Processing, vol.
47, pp. 97-107, Jan. 1999.
[7] H. S. Stone, Image Libraries and the Interne, IEEE
Commun. Magazine, pp. 99-106, Jan. 1999.
[8] I. Stojanovic, S. Bogdanova, and M. Bogdanov,
Retrieving Images Using Content-Based Followed by
Pixel-Based Search, 15th Int. Conf. on Systems, Signals
and Image Processing, pp. 271-274, Bratislava, Slovakia,
June 25-28, 2008.
[9] I. Stojanovic, S. Bogdanova and M. Bogdanov, ContentBased Image Retrieving Improved by Pixel-Based
Search, 14th IWSSIP 2007 and 6th EC-SIPMCS 2007,
pp. 395-398, Slovenia, Maribor, 27-30 June, 2007.
[10] I. Stojanovic, S. Bogdanova and M. Bogdanov,
Retrieving Images Using Content-Based Search and
Progressive Wavelet Correlation, ICEST2008, pp. 121124, Nis, Serbia, June 25-27, 2008.
[11] Ritendra Datta, Dhiraj Joshi, Jia Li and James Z. Wang,
Image Retrieval: Ideas, Influences, and Trends of the
New Age, ACM Computing Surveys, vol. 40, no. 2,
article 5, pp. 5:1-60, April 2008.

39
__________________________________________________________________________________________________________
Detecting Outliers in Interval-Valued Data Using Heuristic

Possibilistic Clustering
VIATTCHENIN Dmitri
National Academy of Sciences of Belarus,
United Institute of Informatics Problems, Laboratory of Information Protection,
Surganov Str. 6, 220012 Minsk, The Republic of Belarus, E-Mail1: viattchenin@mail.ru
Abstract The paper deals with the problem of outlier

detection in the interval-valued data. The
corresponding technique based on a heuristic method
of possibilistic clustering. The description of basic
concepts of the heuristic method of possibilistic
clustering based on the allotment concept is provided.
A technique of the interval-valued data preprocessing
is also given. A methodology of outlier detection in the
interval-valued data is proposed. An illustrative
example of the methods application to the artificial
interval-valued data set is carried out. An analysis of
the experimental results is given and some conclusions
are forwarded.
Keywords:
data mining, clustering; outlier;
uncertainty; interval-valued data; clustering structure.
I. INTRODUCTION
Some preliminary remarks are considered in the first
subsection. The second subsection of this introduction
includes a brief review of robust clustering methods.
A.
Preliminary Remarks
Data mining is the application of specific algorithms

for extracting patterns from the data. Cluster analysis
can be considered as an important method of data
mining.
Clustering is a process aiming at grouping a set of
objects into classes according to the characteristics of
data so that objects within a cluster have high mutual
similarity while objects in different clusters are
dissimilar. In other words, cluster analysis refers to a
spectrum of methods, which try to divide a set of objects
X {x1 ,..., x n } into subsets, called clusters, which are
pair wise disjoint, all non empty and reproduce X via
union.
Fuzzy sets theory, which was proposed by Zadeh [1],
gives an idea of uncertainty of belonging to a cluster,
which is described by a membership function. Fuzzy
clustering methods have been applied effectively in
image processing, data analysis, symbol recognition and
modeling. Heuristic methods of fuzzy clustering,
hierarchical methods of fuzzy clustering and
optimization methods of fuzzy clustering were proposed

by different researchers.
The best known optimization approach to fuzzy
clustering is the method of fuzzy c -means, developed
by Bezdek [2]. The fuzzy c -means algorithm is the
basis of the family of fuzzy clustering algorithms. The
family of objective function-based fuzzy clustering
algorithms is described in Hppner, Klawonn, Kruse and
Runkler [3] in detail.
A possibilistic approach to clustering was proposed
by Krishnapuram and Keller in [4]. The approach was
developed by other researchers. However, algorithms of
possibilistic clustering are also objective functions-based
algorithms [3]. On the other hand, a heuristic approach
to possibilistic clustering was outlined in [5] and the
approach was developed in other publications.
One of the basic problems of data mining is the
outlier detection. An outlier is an observation of the data
that deviates from other observations so much that it
arouses suspicions that it was generated by a different
mechanism from the most part of data [6]. Of course,
outliers are often bad data points. On the other hand, in
many applications outliers contain important
information and their correct identification is crucial. An
illustrative example is a computer security intrusion
detection system, which finds outlier patterns as a
possible intrusion attempts. Intrusion detection
corresponds to a suite of techniques that are used to
identify attacks against computers and network
infrastructures. Anomaly detection is a key element of
intrusion detection in which perturbations of normal
behavior suggest the presence of intentionally or
unintentionally induced attacks, faults and defects.
Detecting outliers has practical application in wide
areas: defense, medicine, geology, economy. So, outliers
should be investigated carefully.
Many data mining techniques find outliers as a sideproduct of clustering algorithms. However, these
techniques define outliers as points, which do not lie in
clusters. Thus, the techniques implicitly define outliers
as the background noise in which the clusters are
embedded. Another class of techniques defines outliers
as points, which are neither not a part of a cluster nor a
part of the background noise; rather they are specifically
points which behave very differently from the norm.
40
__________________________________________________________________________________________________________
Most fuzzy clustering techniques are designed for

handling crisp data with their class membership
functions. However, the data can be uncertain. Different
types of uncertainty can be characterizing the initial data
which must be processed by clustering algorithms. For
example, a brief review of uncertain data clustering
methods is given in [7]. An interval uncertainty of the
initial data is a basic type of uncertainty in clustering
problems. That is why outlier detection in the intervalvalued data is very important problem.
B.
A Brief Review of Fuzzy Clustering Techniques for

the Interval-Valued Data
Let us consider in brief some fuzzy clustering

methods for the interval-valued data. The intervalvalued data should be described in detail in the first
place [8].
Let X {x1 ,, x n } be a set of n objects in an m1 dimensional feature space with coordinate axis labels
( x1 , , x t , , x m } . Each object x i is represented as
1
vector
t1
i
x [ x
of intervals
t1 (min)
i
, x
t1 (max)
i
xi ( x 1i ,, x it ,, xim )
1
where
] . So, the interval-valued data table
t1
i
X nm [ x ] is made up of n rows representing the n

1
objects to be clustered, and m1 columns representing the

m1 interval variables. In other words, each cell of this
table contains an interval x it [ xit (min) , x it (max) ] ,
i {1, , n} , t1 {1, , m1 } .
A number of authors have addressed the problem of
fuzzy clustering algorithms for the interval-valued data.
For example, the interval fuzzy c -means algorithm is
proposed in [8]. In the IFCM-algorithm, the FCMalgorithm [2] is extended to deal with the intervalvalued data. In [9], an interval competitive
agglomeration clustering procedure is proposed to
overcome the problems of the unknown number c of
fuzzy clusters and the initialization of the prototypes in
the ICA-algorithm for the interval-valued data. A robust
version of the ICA-algorithm was proposed in [10] and
the version was called the RICA-algorithm. A fuzzy
clustering method for the interval-valued data with an
adaptable variable selection is proposed in [11].
The major drawbacks of the above approaches are
the difficulty in detecting outliers. Moreover, some few
publications deal with the problem of outlier detection in
interval-valued data sets [12].
A novel methodology of outlier detection in intervalvalued data sets is proposed in the presented paper. The
methodology is based on the heuristic approach to
possibilistic clustering. So, the contents of this paper is
as follows: in the second section basic concepts of the
heuristic method of possibilistic clustering and
techniques of the interval-valued data preprocessing are
presented, in the third section types of clustering
structures are considered and the methodology of
outliers detection is proposed, in the fourth section an
example of application of the proposed methodology to
1
the artificial interval-valued data set is given, in the fifth

section some final remarks are stated.
II. HEURISTIC POSSIBILISTIC CLUSTERING AND
INTERVAL-VALUED DATA PREPROCESSING
Basic concepts of the heuristic method of
possibilistic clustering are defined in the first subsection.
Some techniques of the interval-valued data
preprocessing are considered in the second subsection of
the section.
A.
Basic Concepts of the Heuristic Method of

Possibilistic Clustering
Heuristic algorithms of fuzzy clustering display low

level of complexity and high level of essential clarity.
Some heuristic clustering algorithms are based on a
definition of the cluster concept and the aim of these
algorithms is cluster detection conform to a given
definition. Such algorithms are called algorithms of
direct classification or direct clustering algorithms.
An outline for a new heuristic method of fuzzy
clustering was presented in [5], where a basic version of
direct clustering algorithm was described and the
version of the algorithm that is called the D-AFC(c)algorithm was given also in [13]. The D-AFC(c)algorithm can be considered as a direct algorithm of
possibilistic clustering and the algorithm is the basis of
the family of other heuristic algorithms of possibilistic
clustering.
Let us remind the basic concepts of the heuristic
method of possibilistic clustering. Let X {x1 ,..., x n }
be the initial set of elements and T : X X [0,1] some
binary fuzzy relation on X with T ( xi , x j ) [0,1] ,
xi , x j X being its membership function. Fuzzy
tolerance is the fuzzy binary intransitive relation which
possesses the symmetry property and the reflexivity
property. In this context the classical fuzzy tolerance in
the sense of given definition was called usual fuzzy
tolerance in [5]. In general, some fuzzy tolerance cannot
satisfy the reflexivity condition sometimes [5]. So, the
corresponding fuzzy relation will be subnormal fuzzy
relation. However, the essence of the method here
considered does not depend on the kind of fuzzy
tolerance. That is why the method herein is described for
any fuzzy tolerance T .
Let be the -level value of the fuzzy tolerance
T , (0,1] . Columns or rows of the fuzzy tolerance
matrix are fuzzy sets { A1 ,..., A n } on X . Let Al ,

l {1, , n} be a fuzzy set on X with A ( x i ) [0,1] ,
l
x i X being its membership function. The -level
fuzzy set A(l ) ( x i , Al ( xi )) | Al ( x i ) , xi X is
-cluster.
fuzzy
l
A(l ) A l ,
So,
(0,1] ,
A { A , , A } and A ( x i ) is the membership

l
degree of the element x i X for some fuzzy -cluster

41
__________________________________________________________________________________________________________
A(l ) , (0,1] , l {1, , n} . The membership degree

will be denoted li in further considerations. Value of
is the tolerance threshold of fuzzy -cluster
elements. The membership degree of the element
x i X for some fuzzy -cluster A(l ) , (0,1] ,
l {1, , n} can be defined as a
( x ), xi Al
li A i
,
0,
otherwise
l
where
the
(1)
Al {xi X | A ( xi ) } ,
-level
(0,1] of a fuzzy set A is the support of the fuzzy
-cluster A(l ) , Al Supp( A(l ) ) . The membership

degree defines a possibility distribution function for
some fuzzy -cluster A(l ) , (0,1] , and this
possibility distribution function is denoted l ( xi ) .
Let { A(1 ) ,..., A(n ) } be the family of fuzzy -clusters
for some . The point el Al , for which
xi
(2)
is called a typical point of the fuzzy -cluster A(l ) ,
(0,1] , l [1, n] . Obviously, a fuzzy -cluster can

have several typical points. That is why symbol e is the
index of the typical point.
Let R z ( X ) { A(l ) | l 1, c, 2 c n} be a family of
fuzzy -clusters for some value of tolerance threshold
, which are generated by a fuzzy tolerance T on the
initial set of elements X {x1 ,..., x n } . If condition
where is a set of elements x j , j {1,, n} in all

intersection areas of different fuzzy clusters. The density
D( A(l ) ) of fuzzy -cluster A(l ) Rc ( X ) was defined
in [14] as follows:
1
D( A(l ) )
(5)
li ,
nl x A
i
where nl card ( Al ) , and membership degree li is

defined by formula (1).
B.
Notes on the Interval-Valued Data Preprocessing
Relational heuristic algorithms of possibilistic

clustering can be applied directly to the data given as a
matrix of some fuzzy tolerance T [ T ( x i , x j )] ,
i, j 1, , n . This means that it can be used with the
objects by attributes data by choosing a suitable metric
to measure similarity. The interval-valued data can be
normalized as follows [15]:
xit1 ( t 2 ) min xit1 ( t2 )
i ,t 2
t1 ( t 2 )
xi
,
(6)
t1 ( t 2 )
max xi
min xit1 ( t2 )
i ,t 2
i ,t 2
where t1 1,, m1 , t 2 {min, max} . So, each object

xi , i 1, , n can be considered as an interval-valued
fuzzy
set
x ( x t ) [ x ( x t (min) ), x ( x t (max) )] ,
and
i 1, , n , t 1, , m is its membership function,
l
( )
el arg max li , xi Al
particular, the measure of separation and compactness of

the allotment was defined in [14] as follows:
D( A(l ) )
A
c
R ( X )
(4)
VMSC ( R ( X ); c)
lj ,
c
n x
li
0 , x i X
(3)
l 1
is met for all A(l ) , l 1, c , c n , then the family is the

allotment of elements of the set X {x1 ,..., x n } among
fuzzy -clusters { A(l ) , l 1, c, 2 c n}
where x ( x t (min) ) [0,1] , x ( x t (max) ) [0,1] .
for some
value of the tolerance threshold . It should be noted

that several allotments Rz (X ) can exist for some
tolerance threshold . So, symbol z is the index of an
allotment.
The allotment among fuzzy -clusters can be
considered as the possibilistic partition [4] and fuzzy clusters in the sense of the expression (1) are elements
of the possibilistic partition [13].
Thus, the problem of cluster analysis can be defined
as the problem of discovering the unique allotment
Rc ( X ) , resulting from the classification process and
detection of fixed or unknown number c of fuzzy clusters can be considered as the aim of classification.
Direct heuristic algorithms of possibilistic clustering can
be divided into two types: relational versus prototypebased. These algorithms were described in other
publications. Moreover, different validity measures for
the D-AFC(c)-algorithm were proposed in [14]. In
Different distances for interval-valued fuzzy sets

were proposed in some publications. Dissimilarity
coefficients between the objects can be constructed on a
basis of generalizations of distances between fuzzy sets
which were proposed in [15]. A generalization of the
normalized Euclidean distance for interval-valued fuzzy
sets can be described by the expression [15]:
I ( xi , x j )
,
(7)
1 m 1
t (t )
t (t ) 2
) x (x
)
2 x (x
m1 t 1 2 t {min,
max}
for all i , j 1,, n .

The normalized Euclidean distance between intervalvalued fuzzy sets based on Hausdorff metric was
defined by Grzegorzewski in [16] as follows:
e I ( xi , x j )
( x t1 (min) ) ( x t1 (min) ) 2 , ,
xj
1 m1
xi
max
2
t
(max)
t
(max)
m1 t1 1
xi ( x 1
) xj (x 1
)
(8)
42
__________________________________________________________________________________________________________
Burillo and Bustince [17] suggested some methods

for measuring distances between interval-valued fuzzy
sets.In particular, the normalized Euclidean distance was
defined in [17] as follows:
d I ( xi , x j )
2
t (min)
) x ( x t (min) ) ,
(9)
1 m x (x
2
t
t
(max)
(max)
2m1 t 1 ( x
) x (x
)
x

A similarity measure was defined by Ju and Yuan in
[18] as follows:
s I ( xi , x j )
1
x ( x t (min) ) x ( x t (max) )
1
1
m1
m1
t1 1
2
x ( x t (min) ) x ( x t (max) )
i
(10)
to the matrix of fuzzy intolerance I [ I ( xi , x j )] .

III. AN OUTLINE OF THE APPROACH
Types of clustering structures are described in the
first subsection. A generalized methodology for
detecting outliers in the interval-valued data set is
proposed in the second subsection of the section.
Types of Clustering Structures
The initial matrix of the interval-valued data can be

represented as a set of two matrices X nt m [ xit (t ) ] ,
2
i 1,, n , t 2 {min, max} , and the actual number c of

fuzzy clusters can be different for each matrix
X nt m [ xit ( t ) ] . A clustering structure of the data set
1
clustering structure called unstable.

Detection of most plausible number of fuzzy clusters in the sought clustering structure for the
uncertain data set X can be considered as a final aim of
classification and the construction of the set of values of
most possible number of fuzzy -clusters with
corresponding possibility degrees is the important step
in this way. The corresponding technique of
constructing the stable clustering structure, which
corresponds to most natural allocation of objects among
fuzzy -clusters for the interval-valued data set is
proposed in [20].
B.
A Methodology of Outlier Detection in the IntervalValued Data
tolerance T [ T ( xi , x j )] , i , j 1,, n can be obtained

after application of complement operation [19]
T ( xi , x j ) 1 I ( xi , x j ) , i, j 1, , n ,
(11)
2
for all i , j 1,, n and 1 .
The matrix of fuzzy intolerance relation
I [ I ( xi , x j )] , i , j 1,, n is a result of application
of formulae (7), (8) or (9) to the family of intervalvalued fuzzy sets {x1 ,..., x n } . The matrix of fuzzy
A.
stable. Thirdly, if the number of clusters c is different

for matrices X nt m [ xit (t ) ] , t 2 {min, max} then the
In this work, we consider outliers defined as objects,

which are far from most of other data. The purpose of
proposed approach is first to apply the technique of
constructing the stable clustering structure and then find
otliers from the resulting fuzzy -clusters of the
obtained allotment Rc (X ) . So, the methodology for
detecting outliers in the interval-valued data set can be
summarized as follows:
1. The allotment Rc (X ) among unknown number
c of fuzzy -clusters should be constructed
according to the technique of constructing the
stable clustering structure;
2. The support Supp( A(l ) ) Al should be detected
for each fuzzy -cluster A(l ) Rc ( X ) ;
3. The following condition is checked:
if for some fuzzy -cluster A(l ) Rc ( X ) the
condition card ( Al ) 1 is met
then the corresponding object
xi Al ,
i {1,, n} should be identified as an outlier.
In other words, if a fuzzy -cluster includes only
one typical point then the fuzzy -cluster can be
interpreted as an outlier.
IV. AN ILLUSTRATIVE EXAMPLE
depends on the type of the initial data.

Three types of the clustering structure were defined
in [20]. Firstly, if the number of clusters c is some
constant
for
each
matrix
X nt m [ xit ( t ) ] ,
The artificial interval-valued data set is described in

the first subsection and results of the data processing are
presented in the second subsection of the section.
t 2 {min, max} and the coordinates of prototypes
A.
{ 1 ,, } of the clusters { A1 ,, A } are constant,

then the clustering structure called stable. Secondly, if
the actual number of clusters c is some constant for
each matrix X nt m [ xit (t ) ] , t 2 {min, max} and the
The proposed methodology for detecting outliers in

the interval-valued data should be illustrated by an
example. Let us consider the set of objects
X {x1 ,, x16 } in two-dimensional feature space. Each
object xi X is represented as vector of intervals. The
data set is presented in Table 1.
coordinates of prototypes of the clusters are not

constant, then the clustering structure called quasi-
The Data Set

43
__________________________________________________________________________________________________________
TABLE 1. The artificial interval-valued data set

Numbers of
objects, i
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
x 1
x 2
[30, 80]
[40, 50]
[60, 70]
[80, 90]
[50, 90]
[80, 100]
[180, 220]
[190, 200]
[180, 220]
[210, 230]
[210, 240]
[240, 250]
[10, 20]
[10, 20]
[270, 300]
[280, 300]
[140, 150]
[160, 170]
[130, 190]
[170, 180]
[120, 160]
[110, 130]
[160, 190]
[110, 180]
[140, 150]
[130, 150]
[170, 180]
[180, 190]
[10, 20]
[280, 290]
[270, 300]
[10,20]
Attributes of objects are denoted by x 1 and x 2 . The

interval-valued data are shown also in Fig. 1.
Fig. 1. The interval-valued data set for numerical experiments
In Fig. 1, there are two clusters with outlying objects

x13 , x14 , x15 , and x16 .
B.
Experimental Results
Let us consider an application of the proposed

methodology to the problem of outlier detection in the
interval-valued data. The data were normalized using the
formula (6). The proposed methodology was applied to
the artificial data using formulae (7), (8), (9) and (10).
The measure of separation and compactness of the
allotment (4) was used in the process of constructing
stable clustering structure.
By executing the methodology using the function of
dissimilarity (7), we obtain that the actual number of
fuzzy clusters is equal 6. The allotment Rc (X ) among
six fully separated fuzzy -clusters was obtained for
0.8208 . Membership functions of six classes of the
allotment R (X ) are presented in Fig. 2. Fuzzy -
clusters of the obtained allotment are subnormal fuzzy

sets because a condition of antireflexivity is not met for
the function of dissimilarity (7).
Fig. 2. Membership functions of six classes obtained from the

proposed methodology using the function of dissimilarity (7)
The distance (8) was used in the second experiment.

So, the allotment Rc (X ) among six fuzzy -clusters
was obtained for the tolerance threshold 0.7902 .
Membership functions of six classes are shown in Fig. 3.
Fig. 3. Membership functions of six classes obtained from the

proposed methodology using the distance (8)
Membership values of the first class are represented

in Fig. 2 and Fig. 3 by , membership values of the
second class are represented by , membership values of
the third class are represented by , membership values
of the fourth class are represented by , membership
values of the fifth class are represented by , and
membership values of the sixth class are represented by
. So, objects x13 , x14 , x15 , and x16 should be
identified as outliers in both experiments.
The proposed methodology for detecting outliers was
applied to the data set using the distance (9). The
allotment Rc (X ) among seven fully separated fuzzy
-clusters was obtained for the tolerance threshold
0.8856 . Membership functions of seven classes of
the allotment R (X ) are presented in Fig. 4, where
membership values of the first class are represented by
, membership values of the second class are
represented by , membership values of the third class
are represented by , membership values of the fourth
44
__________________________________________________________________________________________________________
class are represented by , membership values of the

fifth class are represented by , membership values of
the sixth class are represented by , and membership
values of the seventh class are represented by .
Fig. 4. Membership functions of seven classes obtained from

the proposed methodology using the distance (9)
By executing the methodology using the similarity

measure (10) for 2 , we obtain the results which are
similar to the results obtained for the distance (9). The
allotment Rc (X ) among seven fuzzy -clusters was
obtained for the tolerance threshold 0.8994 .
Corresponding membership functions are presented in
Fig. 5.
Fig. 5. Membership functions of seven classes obtained from

the proposed methodology using the similarity measure (10)
A clustering result depends on the selected distance

or similarity measure. However, outliers were detected
correctly in all four experiments. That is why results of
numerical experiments seem satisfactory.
V. CONCLUSIONS
In this paper, a clustering-based methodology to
identify outliers in the interval-valued data is proposed.
The outlier detection process at the same time is
effective for extracting fuzzy clusters and very efficient
in finding outliers.
REFERENCES
[1] L.A. Zadeh, Fuzzy sets, Information and Control, vol.
8, no. 3, pp. 338-353, 1965.
[2] J.C. Bezdek, Pattern Recognition with Fuzzy Objective

Function Algorithms. New York, Plenum Press, 1981.
[3] F. Hppner, F. Klawonn, R. Kruse, and T. Runkler, Fuzzy
Cluster Analysis: Methods for Classification, Data
Analysis and Image Recognition. Chichester, John Wiley
& Sons, 1999.
[4] R. Krishnapuram and J.M. Keller, A possibilistic
approach to clustering, IEEE Transactions on Fuzzy
Systems, vol. 1, no. 2, pp. 98110, 1993.
[5] D.A. Viattchenin, A new heuristic algorithm of fuzzy
clustering, Control and Cybernetics, vol. 33, no. 2, pp.
323-340, 2004.
[6] F. Angiulli and C. Pizzuti, Outlier mining in large highdimensional data sets, IEEE Transactions on Knowledge
and Data Engineering, vol. 17, no. 2, pp. 203215, 2005.
[7] V. Kreinovich and O. Kosheleva, Towards dynamical
systems approach to fuzzy clustering, in Developments in
Fuzzy Clustering, D.A. Viattchenin, ed., Minsk, VEVER
Publishing House, pp. 10-35, 2009.
[8] F.A.T. de Carvalho, Fuzzy c-means clustering methods
for symbolic interval data, Pattern Recognition Letters,
vol. 28, no. 4, pp. 423-437, 2007.
[9] C.-C. Chuang, J.-T. Jeng, and C.W. Tao, Interval
competitive agglomeration clustering algorithm, Expert
System with Applications, vol. 37, no. 9, pp. 6567-6578,
2010.
[10] J.-T. Jeng, C.-C. Chuang, C.-C. Tseng, and C.-J. Juan,
Robust interval competitive agglomeration clustering
algorithm with outliers, International Journal of Fuzzy
Systems, vol. 12, no. 3, pp. 227-236, 2010.
[11] M. Sato-Ilic, Symbolic clustering with interval-valued
data, Procedia Computer Science, vol. 6, pp.358-363,
2011.
[12] S. Li, R. Lee, S.-D. Lang, Detecting outliers in interval
data, in Proceedings of the 44st Annual Southeast
Regional Conference ACM2006 (Melbourne, Florida,
USA, March 10-12, 2006), R. Menezes, ed., New York,
ACM Press, pp. 290-295, 2006.
[13] D.A. Viattchenin, A direct algorithm of possibilistic
clustering with partial supervision, Journal of.
Automation, Mobile Robotics and Intelligent Systems, vol.
1, no. 3, pp. 29-38, 2007.
[14] D.A. Viattchenin, Validity measures for heuristic
possibilistic clustering, Information Technology and
Control, vol. 39, no. 4, pp. 321-332, 2010.
[15] D.A. Viattchenin, An outline for a heuristic approach to
possibilistic clustering of the three-way data, Journal of
Uncertain Systems, vol. 3, no. 1, pp. 64-80, 2009.
[16] P. Grzegorzewski, Distances between intuitionistic fuzzy
sets and/or interval-valued fuzzy sets based on Hausdorff
metric, Fuzzy Sets and Systems, vol. 148, no. 2, pp. 319328, 2004.
[17] P. Burillo and H. Bustince, Entropy on intuitionistic
fuzzy sets and on interval-valued fuzzy sets, Fuzzy Sets
and Systems, vol. 78, no.3, pp. 305-316, 1996.
[18] H. Ju and X.H. Yuan, Similarity measures on intervalvalued fuzzy sets and application to pattern recognition,
in Fuzzy Information and Engineering, D.Y. Cao, ed.,
Berlin, Springer-Verlag, pp. 875-883, 2007.
[19] A. Kaufmann, Introduction to the Theory of Fuzzy
Subsets. New York, Academic Press, 1975.
[20] D.A. Viattchenin, Constructing stable clustering
structure for uncertain data set, Acta Electrotechnica et
Informatica, vol. 11, no. 3, pp. 42-50, 2011.

Jcscs Vol 5 NR 2 October 2012

Uploaded by

Document Information

Original Description:

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

Jcscs Vol 5 NR 2 October 2012

Uploaded by

Copyright:

Available Formats

Academy of Romanian Scientists

University of Oradea, Faculty of Electrical Engineering and Information Technology

Vol. 5, Nr. 2, October 2012

University of Oradea Publisher

- University of Oradea, Romania

Daniela E. POPESCU - University of Oradea, Romania

ISSN 1844 - 6043

Technical University of Cluj-Napoca, Romania

Journal of Computer Science and Control Systems

Journal of Computer Science and Control Systems

Complexity Appreciation for BLDC Flat Top Sinus Implementation

Abstract - The paper presents the usage of a better

Automotive applications that make use of Brushless

The control scheme requires a minimum of hardware

In order to build a simulation model [1][2], the

the magnetic resistance of the magnetic circuit through

The simulation was built up in Cadence Orcad

No skin effect losses in conductors;

No dependency between the inductance and the

The driving 3 phased voltage system is

The magnetic flux density in the air gap of the

Figure 1. Model of magnetic loop with electric flux from

Figure 2. Model of magnetic loop with magnetic flux from

In the above figures the naming of elements is as

b, c. These resistances are dependent on the position and

Figure 3. Battery and connections model

The controlling element is a 6 transistor bridge

Figure 4. 6-transistor Inverter Bridge

When the signal PWM_123 has the value 1, the

Journal of Computer Science and Control Systems

The driving technique is being implemented through

Figure 5. 6-step control technique.

The switching pattern is the same for each of the 3

Bridge is inactive (tri-state, high impedance)

Terminal connected to ground

Terminal connected to supply

High Side PWM (If the duty cycle is equal to

330 -30 : bridge inactive

30 -90 : high-side PWM

90 -150 : phase to supply

150 -210 : bridge inactive

210 -270 : low-side PWM

270 -330 : phase to ground

Figure 6. 12-step control technique.

A proposed method in order to have a good current

The maximum values of the duty cycle [6] will be

Mathematical models and simulation models were

Journal of Computer Science and Control Systems

Use of Artificial Neural Network for Testing Effectiveness of

Abstract This paper presents the suitability of

intelligent models for predicting the shelf life of

the speed requirement, hence new accelerated studies

Fried Potato Chips

Journal of Computer Science and Control Systems

validation set 6. Mean Square Error : MSE (1), Root

: E2 (4) were applied in order to compare the prediction

Figure1. Input and output parameters for ANN models.

Q exp = Observed value;

Qcal = Predicted value;

Figure 2. Training pattern for ANN network.

IV. RESULTS AND DISCUSSION

Table 1. Results of Radial Basis (Exact Fit) model.