jcscs vol 5 nr 2 october 2012
Transcript of jcscs vol 5 nr 2 october 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
1/45
-
7/21/2019 jcscs vol 5 nr 2 october 2012
2/45
Journal of
Computer Science and Control Systems
Vol. Nr. ctober201
University of Oradea Publisher
Academy of Romanian Scientists
University of Oradea Faculty of Electrical Engineering and Information Technology
S
C S
-
7/21/2019 jcscs vol 5 nr 2 october 2012
3/45
EDITOR IN-CHIEF
Eugen GERGELY - University of Oradea, Romania
EXECUTIVE EDITORS
Gianina GABOR - University of Oradea, Romania Daniela E. POPESCU - University of Oradea, RomaniaHelga SILAGHI - University of Oradea, Romania Viorica SPOIAL - University of Oradea, Romania
ASSOCIATE EDITORS
Mihail ABRUDEAN Technical University of Cluj-Napoca, RomaniaLorena ANGHEL I.N.P. Grenoble, FranceGheorghe Daniel ANDREESCU "Politehnica" University of Timisoara, RomaniaAngelica BACIVAROV University Politehnica of Bucharest, RomaniaValentina BALAS Aurel Vlaicu University of Arad, RomaniaBarnabas BEDE The University of Texas at El Paso, USADumitru Dan BURDESCU University of Craiova, RomaniaPetru CASCAVAL "Gheorghe Asachi" Technical University of Iasi, RomaniaHoria CIOCARLIE "Politehnica" University of Timisoara, RomaniaTom COFFEY University of Limerick, IrelandGeert DECONINCK Katholieke Universiteit Leuven, BelgiumIoan DESPI University of New England, Armidale, Australia
Jozsef DOMBI University of Szeged, HungaryToma Leonida DRAGOMIR "Politehnica" University of Timisoara, RomaniaIoan DZITAC Agora University of Oradea, RomaniaJnos FODOR Szent Istvan University, Budapest, HungaryVoicu GROZA University of Ottawa, CanadaKaoru HIROTA Tokyo Institute of Technology, Yokohama, JapanStefan HOLBAN "Politehnica" University of Timisoara, Romaniatefan HUDK Technical University of Kosice, SlovakiaGeza HUSI University of Debrecen, HungaryFerenc KALMAR University of Debrecen, HungaryJan KOLLAR Technical University of Kosice, SlovakiaTatjana LOMAN Technical University of Riga, LatviaMarin LUNGU University of Craiova, RomaniaAnatolij MAHNITKO Technical University of Riga, Latvia
Ioan Z. MIHU Lucian Blaga University of Sibiu, RomaniaShimon Y. NOF Purdue University, USAGeorge PAPAKONSTANTINOU National Technical University of Athens, GreeceDana PETCU Western University of Timisoara, RomaniaMircea PETRESCU University Politehnica of Bucharest, RomaniaEmil PETRIU University of Ottawa, CanadaMircea POPA "Politehnica" University of Timisoara, RomaniaConstantin POPESCU University of Oradea, RomaniaDumitru POPESCU University Politehnica of Bucharest, RomaniaAlin Dan POTORAC "Stefan cel Mare" University of Suceava, RomaniaDorina PURCARU University of Craiova, RomaniaNicolae ROBU "Politehnica" University of Timisoara, RomaniaHubert ROTH Universitt Siegen, GermanyEugene ROVENTA Glendon College, York University, CanadaIoan ROXIN Universite de Franche-Comte, FranceImre J. RUDAS Tech Polytechnical Institution, Budapest, HungaryRudolf SEISING European Centre for Soft Computing, Mieres (Asturias), SpainIoan SILEA "Politehnica" University of Timisoara, RomaniaLacramioara STOICU-TIVADAR "Politehnica" University of Timisoara, RomaniaAthanasios D. STYLIADIS Alexander Institute of Technology, GreeceLorand SZABO Technical University of Cluj Napoca, RomaniaJanos SZTRIK University of Debrecen, HungaryHonoriu VLEAN Technical University of Cluj-Napoca, RomaniaLucian VINTAN "Lucian Blaga" University of Sibiu, RomaniaMircea VLADUTIU "Politehnica" University of Timisoara, Romaniaahin YILDIRIM Erciyes University, Turkey
ISSN 1844 - 6043This 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 andcryptology, Computer networks, Modern control systems, Process control and task scheduling, Web design, Databases anddata mining, Computer graphics and virtual reality, Image processing.
__________________________________________________________________________________________________________2 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
4/45
-
7/21/2019 jcscs vol 5 nr 2 october 2012
5/45
__________________________________________________________________________________________________________4 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
6/45
Complexity Appreciation for BLDC Flat Top Sinus Implementation
DACHIN Tudor1, MEZA Serban2, NEMES Marian3 ,VODA Adriana4, BADILA Florin5
1Universitatea Lucian Blaga Sibiu, Bd. Victoriei Nr.10, Sibiu, [email protected] Tehnica Cluj-Napoca, Str. Constantin Daicoviciu Nr. 15, Cluj-Napoca, [email protected] Automotive Systems S.R.L., Str. Salzburg Nr.8, Sibiu, [email protected]
4iQuest Technologies, Str. Motilor Nr.6-8, Cluj-Napoca, [email protected] Electronic, Str. Caprioarelor nr. 2, Sibiu, [email protected]
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, simulations,
mathematical model, driving technique.
I. INTRODUCTION
Automotive applications that make use of BrushlessDirect Current (BLDC) motors have gained popularitydue to energy efficiency advantages, energy/volumeratio and increased diversity and availability of
application specific control chips. BLDC motors, incomparison 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 ortransmission gear oil), require smaller available volumeand can be driven with very high powers. A BLDCmotor has the similar efficiency as permanent magnetsynchronous machines. Especially due to energyconsumption concerns and the need to improve overallsystem efficiency, usage of BLDC motors is considereda viable alternative and/or an upgrade for more and moreautomotive control applications. In order to encouragethis growth, the manufacturers (motor manufacturers or
integrators) have to increase the knowledge andexperience of their customers, to help in decidingbetween various market available solutions.
One critical aspect of building a system is the designstage and over-designing some aspects (like speed ortorque) is the simple solution to ensure product success.Proper design must include calculations, simulations andmeasurements to choose an optimum solution that uses amotor that fits all requirements, without above describedover-design. Once a motor has been chosen, a controlscheme that matches the required speed and torquecontrol is needed.
The control scheme requires a minimum of hardwareprocessing power, which together with the desiredpurpose of the Electronic Control Unit sets themicrocontroller requirements. This paper presents amethod of mathematically describing the motor, applies
said model in a Pspice simulation and then proposes aless resource demanding implementation of the controlalgorithm. In order to build up a good motor model, theexisting ones have been studied and advantages anddisadvantages compared. The used mathematical modelis to be found in the existing literature. The usage of this
model in order to implement the flat top sinuscommutation 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 wereused 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. oin [A] is themagnetic flux in the star point of the equivalentmagnetic circuit. It is an operand in the magnetic modeland 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 magnetthrough the wiring a.b or c. RM[a,b,c]in [H
-1] = [A/Wb] is
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 5
mailto:[email protected]:[email protected] -
7/21/2019 jcscs vol 5 nr 2 october 2012
7/45
the magnetic resistance of the magnetic circuit throughthe 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 modelwas 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 issymmetrical, rigid and does not contain a zerocomponent; The magnetic flux density in the air gap of themachine is energized by a sinusoidal electric rotation be
(basic wave model);
During real measurements, it is not possible toseparately measure each magnetic resistance due tomotors properties influence on the other systemcomponents (L, R). Hence simulations cannot berelevant only by simulating which magnetic flux is beinggenerated at which moment in time and phasedisplacement. The first part of the simulation model(Figures 1 and 2) describes the magnetic loop withmagnetic 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 asfollowing: Theta Coil A, B, C - Magnetic flux of thewiring a, b, c Theta PM A, B, C - Magnetic flux of thepermanent magnet through the wiring a, b, c. PSI PM a,
b, c - Linked magnetic flux of the permanent magnetthrough the wiring a, b, c RM a, b, c - Magneticresistance of the magnetic circuit through the wiring a,
b, c. These resistances are dependent on the position andthe saturation of the coils.
An electronic control unit [6] and its attached motorare connected with real wires to the battery/alternatorsupply system, which have finite resistance and parasiticinductance/capacitance. The resulting model (Figure 3)is used to simulate real behavior (e.g. oscillations in thecable 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 notallow simultaneous switching of the high and low sidetransistors in the same inverter leg. This is implementedas a minimum dead time between transistor firing.
Figure 4. 6-transistor Inverter Bridge
When the signal PWM_123 has the value 1, themultiport switch will receive the value 0V towards theexit. This situation implements the opening of the lowside transistor and closing the high side one. When the
signal PWM_123 has the value 2, the multiportswitch will receive the maximum differential amplitudeof the 3-phase voltage system towards the exit. Thissituation implements the opening of the high sidetransistor and closing the low side one. When the signalPWM_123 has the value 3, the multiport switch willimplement a voltage dependent on the current and on thevoltage from the motor. The condition 3 presents thecondition of the H bridge with both transistors closed. Aslong 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 belimited at one of this values. This is caused by thevoltage drop on the internal/external diodes.
__________________________________________________________________________________________________________6 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
8/45
IV. CONTROL TECHNIQUES
The driving technique is being implemented throughsoftware. For the PSpice simulation, a repeating patterndefined as a digital stimuli file control the transistor
bridge.
The way of driving is being applied taking intoaccount the type of application.[7]
The signals for the H Bridge are the PWM signalsused to open the transistors. In real life and in case ofsimulations, the driver has to have implemented a dead
time sequence between the turn-on of the phases in orderto avoid damaging the transistors and in order to allowcorrect measurement of transferred power from thesupply 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, aswitching technique between firing the high side with
PWM and low side with full ON or OFF and inverse canbe implemented.
A. 6-step PWM control
One possibility is using a 6 step PWM controltechnique (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 rotationalcontrol 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, byhaving 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.
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 7
-
7/21/2019 jcscs vol 5 nr 2 october 2012
9/45
(11)
(12)
(13)
Figure 6. 12-step control technique.
All three legs of the inverter must have individualPWM duty cycles, dependent of the rotor angle. Theduty cycle can be calculated using the below equation:
(14)
A proposed method in order to have a good current
control is the Sinus Flat top modulation [5], [6], [7]. Thevoltages resulting by applying the duty cycle will have aquasi-sinus form. The final effect on the motor has totake into account the integration property due to themotor inductances. The phase voltages will be calculatedas follows:
The maximum values of the duty cycle [6] will bewritten in a look up table and the value of the rotor angleand two other values, with 120 shift, read. All thesevalues will be inputted in the above equation and theresulting PWM commands forwarded to themicrocontroller 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 asrunning a different initialization script.
V. CONCLUSION
(4)
(5)
(6)
Mathematical models and simulation models werebuilt to study, improve and implement a control strategyfor BLDC motors, with minimum hardware/softwareresources.
The supply voltage will be delivered by applying a
variable Duty cycle between 0
0% and 255
100%(with 8 bit PWM hardware units). The PWM stepnumber and the extreme values are chosen after takinginto account how fine the resulting control must be. Forthe next step we have to calculate the sinus voltagewithout the star null point:
By requiring two look up tables and minimal positioninformation, the implementation of each drivingtechnique is easy to use, relying on emulation ofadditional sensors and performing like more complicatedsolutions. The type of the application will allow decidingon the method to use.
VI. REFERENCES
[1] Michel, Robert. 2009. Kompensation vonsttigungsbedingten Harmonischen in den Strmen
feldorientiert geregelter Synchronmaschinen. Dresden :
Vieweg + Teubner, 2009.
(7)
(8)
(9)
[2] Schrder, Dierk. 2009. Elektrische Antriebe - Regelungvon Antriebssystemen. Berlin Heidelberg : Springer-
Verlag, 2009.
Where U[1,2,3] in [V] is the calculated voltage valuewithout the zero value from star point connection atsupply 1, 2 or 3. Also, we have to calculate the zero ofthe system:
[3] Tarmoom, Osama. 2006. Beitrag zur Auslegung vonPermanent-Magnet-Motoren fr spezielle Einsatzgebiete
dargestellt am Beispiel einer Versuchsmaschine. Cottbus :
s.n., 2006.[4] AVR928: Scalar sensorless methods to drive BLDC
motors.
(10)
[5] FCM8202 3-Phase Sinusoidal Brushless DC MotorControler.
Where U123Null in [V] is the offset on the 3 phasevoltages. It must be added to the voltages calculatedwithout the zero system:
[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 EC-
Motoren Serir MCBL2805
__________________________________________________________________________________________________________8 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
10/45
Use of Artificial Neural Network for Testing Effectiveness of
Intelligent Computing Models for Predicting Shelf Life ofProcessed Cheese
GOYAL Sumit, GOYAL Kumar Gyanendra
National Dairy Research Institute, Karnal -132001, IndiaE-mail:[email protected]; [email protected]
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; artificialintelligence; 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 doublejacketed 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 ofprocessed cheese stored at 7-8C.
The first Artificial Neural Network (ANN) wasinvented in 1958 by psychologist Frank Rosenblatt
called perceptron. It was intended to model how thehuman brain processed visual data and learned torecognize objects.
An artificial neural network operates by creatingconnections between many different processingelements, each analogous to a single neuron in a
biological brain.These neurons may be physically constructed or
simulated by a digital computer. Each neuron takesmany input signals, then, based on an internal weightingsystem, produces a single output signal that's typicallysent 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 functionnetwork is an ANN that uses radial basis functions asactivation functions. It is a linear combination of radialbasis functions.
They are used in function approximation, time seriesprediction, and control. Radial basis function network
consists of one layer of input nodes, one hidden radial-basis function layer and one output linear layer [2].
As an increasing number of new foods compete forspace on supermarket shelves, the words speed and
innovation have become the watchwords for foodcompanies 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 thespeed and innovation system. How the consumerperceives the product is the ultimate measure of foodquality.
Therefore, the quality built in during thedevelopment and production process must last throughthe distribution and consumption stages.
Shelf life studies can provide important informationto product developers and manufacturers enabling them
to ensure that the consumer will receive a high qualityproduct for a significant period of time after production.Since long time taking shelf life studies do not fit with
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 9
-
7/21/2019 jcscs vol 5 nr 2 october 2012
11/45
the speed requirement, hence new accelerated studieshave been developed [3]. The results of this researchconcerning development of radial basis (exact fit) ANNmodels for predicting the shelf life of processed cheesewould be very beneficial for consumers, dairy factoriesmanufacturing processed cheese, wholesalers, retailers,
food researchers, academicians and regulatoryauthorities.
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 ismeasured 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 1ststep 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 OxideSemiconductor 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 couldbe a useful tool for the characterization and control of
honey [5].
C. Beef
A series of partial least squares (PLS) models wereemployed to correlate spectral data from FTIR (Fourier
Transform Infrared Spectroscopy) analysis with beeffillet spoilage during aerobic storage at differenttemperatures (0,5,10,15,and20C).
The performance of the PLS models was comparedwith a three - layer feedforward ANN developed usingthe same dataset. FTIR spectra were collected from the
surface of meat samples in parallel with microbiologicalanalyses to enumerate total viable counts.
Sensory evaluation was based on a three-pointHedonic scale classifying meat samples as fresh, semi-fresh, and spoiled. The purpose of the modellingapproach employed in this work was to classify beefsamples in the respective quality class as well as topredict their total viable counts directly from TIR
spectra.The results obtained demonstrated that both
approaches showed good performance in discriminatingmeat samples in one of the three predefined sensoryclasses.
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 testingdataset.
The ANN classification model performed equallywell in discriminating meat samples, with correctclassification 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 andstatistical indices (bias factor, accuracy factor and rootmean 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 disjointsubsets, viz., training set having 30 observations, and
__________________________________________________________________________________________________________10 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
12/45
-
7/21/2019 jcscs vol 5 nr 2 october 2012
13/45
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
Selecting
minimum
error
Evaluation
of error and
weights
Training
ANN models
__________________________________________________________________________________________________________12 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
14/45
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 forpredicting the shelf life of processed cheese stored at 7-8o 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 thestudied experiments (Table 1).
Figure 3.Input and output parameters for ANN models.
The modeling results showed that there wasexcellent agreement between the experimental data andpredicted values, with a high determination coefficient(R
2= 0.999087741) and Nash - Sutcliffo Coefficient (E
2
= 0.999999168) showing that the developed model wasable to analyze nonlinear multivariate data with verygood performance. It is evident from the high R2and E2
that the ANN computing models are very effective inpredicting the shelf life of processed cheese.
V. CONCLUSIONS
An effective new radial basis (exact fit) method
based on artificial neural network is proposed forpredicting the shelf life of processed cheese stored at 7-8
oC. The results were verified by comparing them with
laboratory observations by employing mean squareerror, root mean square error, coefficient ofdetermination and Nash - sutcliffo coefficient. Verygood correlation was found between experimental dataand the developed mathematical models, thusconfirming the suitability of radial basis (exact fit)artificial neural networks for predicting the shelf life of
processed cheese.
REFERENCES
[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 Aaccumulation in grape juice based-medium by Aspergillus
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 13
-
7/21/2019 jcscs vol 5 nr 2 october 2012
15/45
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 ofFood 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 theclassification 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 neuralnetworks and partial least squares modelling for the rapiddetection of the microbial spoilage of beef fillets based on
Fourier transform infrared spectral fingerprints, FoodMicrobiology, vol.28, no.4, pp. 782790, 2011.
[7] Sumit Goyal and G.K. Goyal, Brain based artificialneural network scientific computing models for shelf lifeprediction 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 networkintelligent computing models for predicting shelf life of
soft cakes, Global Journal of Computer Science andTechnology, vol.11, no.14, version 1.0, pp. 29-33, 2011.
[9] Sumit Goyal and G.K. Goyal, Advanced computingresearch on cascade single and double hidden layers fordetecting shelf life of kalakand: An artificial neural
network approach, International Journal of ComputerScience & Emerging Technologies, vol.2, no.5, pp. 292-295, 2011.
[10]Sumit Goyal and G.K. Goyal, Application of artificialneural 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 forprediction of sensory quality of instant coffee flavouredsterilized 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 neuronbased artificial intelligent scientific computer engineering
models for estimating shelf life of instant coffee sterilized
drink, International Journal of Computational Intelligenceand 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 forpredicting 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 withalmonds, International Journal of Latest Trends inComputing, vol.2,no.3, pp. 434-438, 2011.
[15]Sumit Goyal and G.K. Goyal, Development ofintelligent computing expert system models for shelf lifeprediction 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 forshelf life prediction of post-harvest coffee sterilized milkdrink, 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 predictionof processed cheese, International Journal of u- and e-
Service, Science and Technology, vol.5, no.1, pp.63-69,2012.
[18]Sumit Goyal and G.K. Goyal, A novel method for shelflife detection of processed cheese using cascade single andmulti 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 ofprocessed cheese, International Journal of Intelligent
Systems and Applications, vol.4, no.5, pp.30-37, 2012.
[20]
Sumit Goyal and G.K. Goyal, Estimating processedcheese 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 forshelf life prediction of processed cheese, International
Journal of Artificial Intelligence and KnowledgeDiscovery, vol.1, no.4, pp.12-15, 2011.
__________________________________________________________________________________________________________14 Volume 5, Number 2, October 2012
http://www.medlabs.com/Downloads/food_product_shelf_http://www.medlabs.com/Downloads/food_product_shelf_ -
7/21/2019 jcscs vol 5 nr 2 october 2012
16/45
A Z-Source Inverter for an Integrated Starter Alternator
HANGIU Radu-Petru, FILIP Andrei-Toader, MARIClaudia 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: [email protected]
Abstract This paper presents an overview of an
integrated starter alternator system used in mild hybridelectric vehicles. Inverter configurations that are used
in HEV are assessed with an emphasis on the novel Z-
source inverter topology. The final part presents a
simulation model of a bi-directional Z-source inverter,
developed in AMESim, and the simulation results.
Keywords: ISA; model; simulation; Z-source,
AMESim.
I. INTRODUCTION
The conventional internal combustion engine (ICE)era is at its dawn because of the increase in fuel pricesand a more stringent legislation on greenhouse gasemissions. The obvious alternative for personaltransportation is the electric vehicle, but the technologyneeded to make this type of vehicles accessible to the
masses is still at an initial stage. The hybrid electricvehicle (HEV) is a viable compromise until moreefficient batteries or fuel cells are developed.
The HEV comes in different configurations.Depending on the power flow within the vehicle and onthe energy sources, they can be classified as: series,parallel, series-parallel or as mild HEVs. The mild HEVconfiguration represents the first step in the transitionfrom an ICE vehicle to a full HEV and onwards to a fullelectric vehicle. The integrated starter-alternator (ISA)unit represents the key difference between an ICEvehicle and a mild HEV.
The ISA is in fact an electric machine that has a two
quadrant operation thus combining in one unit the starterand the alternator of a conventional ICE vehicle. Besidesits main function of cranking the ICE and generatingelectric power, the ISA can be used to implement otherfunctionalities that may improve fuel efficiency and ridecomfort, such as: start/stop functionality, powerboosting and regenerative braking.
The ISA is driven by a power converter which, in aconventional system, is composed of an inverter and abuck DC-DC converter. The inverter steps up thevoltage of the battery and provides AC to the motor andthe buck converter steps down the rectified voltage inorder to charge the battery when the ISA is in generator
mode.The traditional inverter configurations used for HEVare the voltage source inverter (VSI) and the current
source inverter (CSI). The VSI has the advantage of alow cost and a simple control but has importantdrawbacks such as the necessity of a bulky dc buscapacitor, high electromagnetic noises, high frequencylosses and the necessity of a separate dc-dc converter inorder to boost the battery voltage. The CSI main
advantage comes from its capability to boost the dcvoltage of the battery without a separate boost converter.Other advantages come from the fact that it canwithstand short circuits across any two of its outputterminals, it doesnt need a bulky dc bus capacitor oranti-parallel diodes thus reducing the overall size of theinverter. CSI disadvantages are the inability to reversethe DC current in order to charge the battery, highconduction losses and the need for forced commutationwhich means that only switches capable of blockingvoltage in both directions can be used.
The Z-source (impedance source) inverter (ZSI) is anovel inverter configuration, first proposed by F. Z.Peng [1], that overcomes some of the limitations oftraditional inverters.This paper presents an application of a ZSI for apermanent magnet synchronous motor ISA to be used ina mild HEV. First the ISA architecture and ZSI arepresented in detail. The final part presents simulationresults of the ISA driven by a ZSI.
II. ISA SYSTEM OVERVIEW
As mentioned before, a vehicle equipped with anISA is in fact a mild HEV. The main difference between
a mild HEV and a full HEV comes from the fact that inthe case of the mild HEV, the ICE is always on and isdriving the wheels as long as the vehicle is moving.
A. Drive train configuration
There are several possible arrangements within thevehicle drive train for the electric machine acting as anISA [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 threepossibilities, one with the electric machine on thecrankshaft, one with the machine sandwiched between
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 15
-
7/21/2019 jcscs vol 5 nr 2 october 2012
17/45
two clutches and one with the machine mounted on theinput shaft of the transmission. Similar arrangements arepossible in the case of the non-coaxial solution wherethe electric machine that is now mounted on the side ofthe drive train will be connected via a drive.
The most widespread solution is that with the electric
machine mounted on the crankshaft. The rotor isattached directly to the ICEs crankshaft and replacesthe flywheel. The advantage of such an arrangement,besides its simplicity, consists in achieving a dual massflywheel effect by connecting the electric machine to thefriction plate for the clutch, with a torsion damper. Thiswill eliminate excessive transmission gear rattle, reducegear shift effort and further increase fuel economy.
B. Operation modes
The ISAs main function is to generate power whilethe ICE is on but it can also assist the engine when highloads are requested. The systems modes of operation
are:1) Internal combustion engine cranking2) Power generation mode3) Regenerative braking mode4) Power boost mode
C. Electric machines
For functioning as both a starter and an alternator theelectric machine must have a wide constant power speedrange. It must provide a high starting torque for enginecranking even at very low temperatures and it must havea high efficiency in generator mode at speeds rangingfrom 1500 rpm to 4000 rpm. At the moment the two
competing electric machines for ISA applications are thePermanent Magnet Synchronous Machine (PMSM) andthe Induction Machine (IM).
III. Z-SOURCE INVERTER
The Z-Source inverter, shown in Fig. 1, replaces thedc-link present in a conventional VSI with an impedancenetwork 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 boostthe dc-bus voltage. By varying the active state andshoot-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 pulsewidth modulation (PWM) control used for conventionalinverters needs to be modified. Several control methodshave been proposed:1) Simple boost control (SBC)
This method was first proposed in [1] and it utilizesan upper and a lower limit to control the shoot-through states. When the triangular carrier wave isgreater than the upper limit, or lower than the lowerlimit, the inverter is in shoot-through state, the rest
of the time the control is the same as a normalcarrier-based PWM. This methods disadvantage isthe high voltage stress on the switches.
Figure 1. Z-Source inverter.
2) Maximum boost control (MBC)
This method was presented in [3]. Its workingprinciple is to turn all of the inverters zero statesinto shoot-through states thus minimizing thevoltage stress on the switchesIts main draw back is that it produces low frequencyripples 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]. Thedrawbacks of the previous two methods aremitigated by keeping a constant shoot-through dutyratio while maximizing the boost factor. This isachieved either by utilizing a variable lower andupper limit for controlling the shoot-through states,either by injecting a third-harmonic component intothe reference signals. The envelopes used in the firstcase are periodic signals that have a frequency threetimes higher than that of the inverter output. Theinjected third harmonic component is 1/6 of the
fundamental component. When using third harmonicinjection the upper and lower limits for controllingthe shoot-through states are straight lines. Thedistance between this two limits is constant and
equal to M3 [4].
B. Component rating
The inverters boost factor B and gain G can bedetermined using the following [4]:
13
1
MB (1)
132/
MMMB
EVGS
m (2)
Where Vmis the peak phase output voltage,M is themodulation index andESis dc-bus voltage.
For rating the ZSIs inductors and capacitors, for anycontrol strategy, the following equations can be used [5]:
cos3
1(2
mi
SSS
MIk
dTdEL
)S (3)
)1(8 Sv
mSS
EkC
cos3
Sd
MITd (4)
Where dSis the shoot-through duty ratio, kvand kiaredesired voltage and current ripple factors, Imis the peakphase current and TSis the switching period.
__________________________________________________________________________________________________________16 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
18/45
Figure 2. Control model.
C. DC-DC converter
In order to accept a reverse power flow from the ISAto 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 secondswitch and an inductor placed between the Z network
and the 3 phase bridge and a smoothing capacitor. Thebi-directional switch replaces the Z-source networkdiode so that, depending on the desired operating mode,the converter can now accept a direct or a reverse powerflow. This converter can perform either as a buckconverter, either as a reversed polarity buck-boostconverter depending on the duty ratio.
By controlling the duty ratio of the switches theconverter output voltage can be regulated. The voltagetransfer ratio of the converter is determined with thefollowing equation [6]:
D
DG
12
(5)
Where Gis the voltage gain andDis the duty ratio.
IV. SYSTEM MODELING AND SIMULATIONRESULTS
The system model has been implemented inAMESim, which is a multi-domain simulation softwarefor the modeling and analysis of one-dimensional (1D)systems.
The chosen electric machine is a three-phase, 7.5kW, outer rotor PMSM presented in [7]. The outer rotorconfiguration allows for an easy integration of the
machine in the vehicle drive train and also helps withthe cooling of the permanent magnet rotor. The electricmachine specifications are:
Table 1. Electric machine specifications.
Active
power
7.5
kW
Voltage
72
V
Frequency 100Hz
Pole
pair
number
15
Rated
speed
400
rpm
Rated
torque
150
Nm
Statorcurrent 66A
Efficiency 86.8%
Power
factor
0.65
The battery dc voltage (ES) is set at 12 V, the inverterswitching frequency (TS) at 9 kHz and the voltage (kv)and current (ki) ripple factors are 5% and 1%,respectively. Considering the electric machines ratedvoltage and the batterys voltage the inverter mustprovide a voltage gain (G) of 6. The modulation indexM
is 0.621.Based on this values the Z-network inductance andcapacity are determined using (3) and (4). Thecalculated values are:
HL 1.54 (6)
mFC 9.3 (7)
The chosen control method is the constant boostcontrol method based on its advantages. The controlmodel of the system is presented in Fig. 2. It consists ofsine reference signal generators, a triangular carriergenerator and a series of signal comparison blocks usedto determine the control signals for the inverter switches.
The simulated reference signals and triangularcarrier, inverter output current and inverter outputvoltage are presented in Fig. 3, Fig. 4 and Fig. 5.
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 17
-
7/21/2019 jcscs vol 5 nr 2 october 2012
19/45
-
7/21/2019 jcscs vol 5 nr 2 october 2012
20/45
-
7/21/2019 jcscs vol 5 nr 2 october 2012
21/45
the location of the fault: inner race, outer race,balls, and cage [7].
the fault signature: single-point defects andgeneralized roughness [11].
A. SINGLE-POINT DEFECTS
A single-point defect can produce differentcharacteristic fault frequencies in the vibration spectrumof the machine. These frequencies are predictable anddepend on the surface of the bearing which contains thefault [12].
The single-point defects cause periodic impulses invibration signals. Amplitude and period of theseimpulses are determined by shaft rotational speed, faultlocation and bearing dimensions. Therefore, a specificfrequency can be attributed to each component of thebearing [13].
The fundamental cage frequency is given by:
cos1
2 D
dff rc
(1)
The ball defect, respectively the inner race defectfrequencies can be computed by using the followingequations:
2
2
2
cos1D
df
d
Df rbd
(2)
cos1
2 D
dnfffnf rcrid
(3)
The formula for computing the outer race defectfrequencies is the following:
cos1
2 D
d
d
nfnff rcod
(4)
where,fris the rotor speed,n the number of balls, dthediameter of the ball,Dthe pitch diameter of the bearingand the contact angle as shown in Fig. 1. The typicalvalue of the contact angle is 0.
For most bearings with six to twelve balls, thefrequencies given by (3) and (4) can be approximatedwith [13]:
rid fnf 6.0 (5)
rod fnf 4.0 (6)
It is known that any air-gap eccentricity producesanomalies in the air-gap flux density, which is reflectedon the stator current. In the case of a bearing fault thecharacteristic fault frequencies are modulated by theelectrical supply frequency at a predictable frequency[13].
Vsbng fmff (7)
wherefs is the electrical supply frequency,fVone of thefault frequencies defined by equations (1)(4) andm = 1, 2, 3...
B. GENERALIZED ROUGHNESS FAULT
Generalized roughness fault is the most frequentcause of bearing failure. It usually occurs in theindustrial environment due to various causes such as[12]:
Lack or loss of the lubricant, contamination oflubricant 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 toproduce generalized roughness fault, as well as single-point defects. A generalized roughness fault of a bearingcan be easily determined because it spins roughly or
with some difficulty.
III. CONDITION MONITORINGOF BEARING FAULTS
A significant part of the papers on the fault diagnosisof induction machines are dealing with on the faults ofrolling bearings.
Even though that vibration based conditionmonitoring techniques are usually applied for thediagnosis of the bearings, many papers use the statorcurrent analysis, due to its advantages.
The methods used for stator current analysisdecompose and analyze the signal using varioustechniques such as Fourier analysis, neural networks,wavelets, statistical analysis, etc.
In [14] the authors analyze two types of bearingfaults: a hole drilled into the outer raceway and anindentation produced in the inner and outer surface.Vibration and current analysis is applied to both faultyconditions.
The specific characteristic fault frequencies arehighlighted for both faults. The analysis of the first faultshows two components, fod and 2fod in the vibrationspectrum, and |fs fod| and |fs 2fod| in the currentspectrum. For the second type of fault the highlightedcharacteristics are fod, 2fod and fid for the vibrationspectrum, and |fs fod|, |fs 2fod| and |fs fid|, for thecurrent spectrum.
In [27] the authors introduce a new formulation forthe current spectral analysis for the detection of bearingfailures in induction motors driven by frequency powerconverters. The fault is an outer race defect and theauthors highlight an increase of the specific faultfrequencies components of the current spectrum.
In [15] two inner raceway faults (drilled hole andspalls) are studied using vibration and current analysis.
The results show that the fault frequencies are clearlyvisible only in the vibration spectrum. The authors statethat the assembling, disassembling, remounting and
__________________________________________________________________________________________________________20 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
22/45
realigning the test motor can alter the vibration andcurrent spectra.
Some authors studied the detection of faults in
electrical machines by using the stray flux around themotor [17], [18], [19], [20], [21], [22].In [20] a small active area sensor was used for the
detection of eccentricity and bearing faults. For thebearing fault analysis, a hole was drilled in the innerrace of the bearing. Current and stray flux measurementswere effectuated under different loading conditions. Thespectra of the signals were obtained using Fast FourierTransform (FFT). The characteristic fault frequencieswere almost the same in both spectra, current and flux,but the amplitudes of these components were low inboth cases.
In [16] the Park's Vector Approach is used for the
detection of broken bearings. The analysis is made on afour bearings with drilled holes in the inner race andouter race. One of the bearings has two drilled holes inthe outer race. The authors concluded that the diagnosisof the inner race faults is more difficult because ofvibration signal is weak and it is not fully transmitted tothe outer race. The results show a good detection of thefaulty conditions.
The Park's Vector Approach was also used in [23]for the diagnosis of three bearings with differentdiameter holes in the outer race. The proposed methodshowed good results in detecting even an incipient faultcan be detected using this method. This paper also
presents a new technology for artificially introducingbearing faults such as: pitting, flutting or falsebrinelling. The method consists in removing the pins ofthe cage, so all the bearing components can beaccessible.
In [24] the Continuous wavelet transform (CWT) isused for the extraction of characteristic features fromvibration signals measured for induction machinessubjected to bearing fluting. The faults of the bearingwere obtained artificially by using an ElectricalDischarge Machining (EDM) and thermal ageing. Theproposed method was compared with Short-TimeFourier Transform and it was highlighted that the CWT
has some advantages. By using the CWT the authorswere able extract small amplitudes that cannot beobserved along the frequency axis. Also, they found
extra amplitudes caused by the damage of the bearingbetween 24 kHz.
In [25, 26] the broken bar and bearing faults (innerrace defect) of several inverter-fed induction machinesare studied with a new hybrid algorithm that combinesthe analysis of the signal in time and frequency domain.
This new method uses a combination betweenIndependent Component Analysis (ICA) and FFT inorder to analyze features of the stator current. ICA is astatistical technique for decomposing a complex datasetinto independent subparts. The authors state thatproposed method detects and classifies correctly thecharacteristic fault frequency components. In the case ofbearing faults, the detection is more difficult. It is shownthat the predominant characteristic fault frequency isgiven byfid
For bearing fault diagnosis other authors have useddifferent methods such as: neural networks [28], hiddenMarkov modeling [29], instantaneous power factor [30],
etc.
Figure 2. Example of artificially drilled holesin the outer raceway of a bearing [16].
IV. CONCLUSIONS
The literature reviewed in this paper aims toinvestigate the possibility of employing the analysis ofthe stray flux and stator current for bearing faultdetection of induction machines in future papers.
ACKNOWLEDGMENT
This paper was supported by the project"Improvement of the doctoral studies quality inengineering science for development of the knowledgebased society-QDOC" contract no. POSDRU/107/1.5/S/78534, project co-funded by the European Social Fundthrough the Sectorial Operational Program HumanResources 2007-2013.
REFERENCES
[1] W.T. Thomson, "A review of on-line conditionmonitoring techniques for three-phase squirrel-cageinduction motors Past present and future," Proceedingsof the IEEE Symposium on Diagnostics for ElectricalMachines, Power Electronics and Drives(SDEMPED '99), Gijon (Spain), pp. 3-18, 1999.
[2] M.E.H. Benbouzid, "A review of induction motorssignature 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 motorreliability survey of industrial and commercialinstallations Part I and II," IEEE Transactions onIndustry Applications, vol. IA21, no. 4 (July 1985),pp. 853-872, 1985.
[4] F. Filippetti, G. Franceschini, C. Tassoni, "Neuralnetworks aided online diagnostics of induction motorrotor faults," IEEE Transactions on IndustryApplications, 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
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 21
-
7/21/2019 jcscs vol 5 nr 2 october 2012
23/45
-
7/21/2019 jcscs vol 5 nr 2 october 2012
24/45
Design and Implementation of a PWM Inverter for Reluctance
MotorsMARGINEAN Calin1, MARGINEAN Ana-Maria1, VESE Ioana1, TRIFA Viorel1, TRIFU
Emil2
1Technical 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:
[email protected], [email protected], [email protected], [email protected]
S.C. TRAMBUS S.R.L., Cluj-Napoca, Romania, E-Mail: [email protected]
Abstract The paper deals with aspects regarding the
simulation, implementation and testing of a PWMinverter 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.
I. INTRODUCTION
The advance of incremental motion control systems,where one usually uses stepping motors, has beenenforced by the multiplicity of their utilization in digitalcontrolled machine-tools drives, peripheral computer
equipments, telecommunications through laser andsatellites, nuclear techniques, industrial robots,
aeronautical and military equipments etc. In this context,the VRSMs promise the low-cost production andmotivate the comprehensive research and design
although they are not included in the classical treatmentof the DC or AC electrical drives.
Among reluctant motors, variable reluctance
stepping motors (VRSM) and switched reluctancemotors (SRM) are the most popular. VRSM, which isvery representative as reluctant motors, serves especiallyin digitally controlled open-loop positioning servo-systems and is very suitable for board instrumentations.SRM has imposed itself in the last years, especially in
variable speed applications, due to the simpletechnology involved. It is used especially in speedservo-systems and operates with self-commutation oftheir phases.
Despite of their excellent robustness as actuators inspecial applications, both types of motors are confronted
with a major problem of their supplying and driving
systems, whichfocused much interest among specialistsin 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 inPWM inverters.
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 arepresented for phase 1.
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 23
-
7/21/2019 jcscs vol 5 nr 2 october 2012
25/45
-
7/21/2019 jcscs vol 5 nr 2 october 2012
26/45
Figure 8. Signals from sequencer using SLPS interface.
Figure 9. Current and voltage for phase 1 with SLPS interface.
III. PWMINVERTER
Usually reluctant motors are supplied by voltage
source (series resistance switches or dual voltageschemes) 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-bridgeinverters.[2]
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 arepresented in figure 12
Figure 11. Orcad model of the inverter for one phase.
Figure 12. Pspice simulation results.
After the simulation, the pulse generator, thesequencer and the inverter was practically realized andtested.
IV. EXPERIMENTALRESULTS
In order to test the equipment, an experimental test
bench was made based on DSpace DS1104 controllerboard. The test bench is presented in figure 13.
Figure 10. Block diagram of the inverter.
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 25
-
7/21/2019 jcscs vol 5 nr 2 october 2012
27/45
Figure 14 and figure 15 presents the pulse generatorand sequencer, respectively a module of the PWMinverter.
Figure 13. Experimental test bench.
Figure 16. ControlDesk virtual control panel for current
acquisition.
V. CONCLUSIONS
An PWM inverter was designed and simulated using
two most comm in the industryon software products
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]
Figure 14. Pulse generator and sequencer.
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 andcould 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 15one can see that
shunt resistors Rsh13and Rsh24were 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 curreacquisition made with ControlDesk software fro
DSpace.
ntm
[1] V. Trifa, C. Marginean, . Zarnescu, Investigation of
variable relucta ors dynamics using
Matlab-Simulink ceedings of the 7th
07.
Lnce stepping mot
environment, ProInternational Conference on Development andApplication 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 vehicle-
motor 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 andComputer, 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.uide, March 20[8] ***Dspace ControlDesk Experiment G
__________________________________________________________________________________________________________26 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
28/45
Microstructure Development by Controlling Grain Size
RAMAKRISHNAN Sumathi1
, MAHALINGAM Usha2
1Department of Electronics and Communication Engineering
Pavai College of Technology, Namakkal-637 018, Tamil Nadu, India
Email ID: [email protected]
2Department of Computer Science and Engineering
Sona College of Technology, Salem-636 005, Tamil Nadu, India
Email ID: [email protected]
Abstract - I n thi s Paper, a classical PID control ler and
an adaptive fuzzy logic controller are employed for
controll ing the microstructur e development duri ng hot
working process. The strength of any material is
dependent on the grai n size of that materi al [ 4], [9].
The strength the materi al is increased when its grai n
size is reduced. I n this paper, the standard
Ar rehenious equation of 0.3% carbon steel i s uti li zed
to obtain an optimal deformation path such that the
grai n size of the product shoul d be 26m. The 0.3%
carbon steel improves in the machinability by heat
treatment [ 8]. I t must also be noted that th is steel is
especially adaptable for machining or forging and
where surf ace hardness is desirable. The plant model
is developed with grain size. The effect of processcontrol 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 initi al grai n size of 180m. The simulation is done
on vari ous grai n sizes using both the control lers by
MATLAB simulink toolbox. When comparing the
responses, the PID control ler 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 Control ler.
I.INTRODUCTION
The development of optimal design and controlmethods for manufacturing processes is needed foreffectively reducing part cost, improving part deliveryschedules, and producing specified part quality on arepeatable basis. Existing design methods are generallyad hoc and lack adequate capabilities for finding
effective process parameters such as deformation rate,die and work piece temperature, and tooling systemconfiguration. This situation presents major challenges
to process engineers who are faced with smaller lotsizes, higher yield requirements, and superior qualitystandards. Therefore, it is important to develop newsystematic methodologies for process design and controlbased upon scientific principles, which sufficientlyconsider the behavior of work piece material and themechanics of the manufacturing process. A new strategyfor systematically calculating near optimal controlparameters for control of microstructure during hotdeformation processes has been developed based onoptimal control theory [1]. This approach treats thedeforming material as a dynamical system explainedbelow.
II. STATICAND DYNAMICMODEL
The static model of 0.3% carbon steel [3] is,
Where,
Q
The dynamic model of 0.3% carbon steel is obtainedby using the Arrehenius equation for changes intemperature during hot extrusion is given below.
Where,
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 27
-
7/21/2019 jcscs vol 5 nr 2 october 2012
29/45
The dynamic equation for grain size can be obtainedby differentiating the equation (1) with respect to
temperature and then multiplied by change intemperature T, which follows that,
III. OPEN LOOP MODEL
It is proposed to optimize the grain size of 26mfrom the initial grain size of 180m. TheMatlab/Simulink simulation model for open loop system
is obtained from the equation (3). The steady stateoperating ranges for the control parameters temperature,strain, strain rate and the grain size are considered as,
IV. PIDCONTROLLER
PID controllers are used extensively in the industryas an all-in-all controller, mostly because it is anintuitive control algorithm. A theoretical PID controller
[6] is
Whereu(t) = the input signal to the plant modele(t) = the error signal is defined as e(t) = r(t)y(t)
andr(t) = the reference input signal.y(t) = plant outputKp, Tiand Tdare the proportional gain, integral time
and derivative time respectively.The coefficientsKp, Ti, Td andP,I,D are related by:
P = KpI = Kp/TiD = KpTd
The controller has three parts: The proportional termis providing an overall control action proportional to theerror signal through the all-pass gain factor. The integralterm is reducing steady-state errors through low-frequency compensation by an integrator. Thederivative term is improving transient response throughhigh-frequency compensation by a differentiator.
Figure 1. Block Diagram to Optimize Grain Size by PIDController
V. FUZZYLOGICCONTROLSYSTEM
It is a classical logic system which provides analternative way of thinking, which allows modeling ofcomplex systems using knowledge and experience ofoperating a control system. It provides a simple way todraw definite conclusions from vague or imprecise
information and resembles human decision-making inits ability to work with of approximate data and findprecise control system solution. The concept of fuzzylogic is not presented as a control methodology, but as away of processing data by allowing partial setmembership rather than crisp set membership or non-membership. If feedback controllers programmed toaccept noisy, imprecise input, they would be much moreeffective and perhaps easier to implement.
It lends itself to implementation in system rangingfrom simple, small, embedded microcontrollers to large,networked, multi-channel PC or workstation based dataacquisition and control system. It can be implemented in
software, hardware or a combination of both. Fuzzylogic provides a simple way to arrive at a definiteconclusion based on vagueness, ambiguity, imprecision,noise or missing input information. Approach to controlproblems, using fuzzy logic, mimics how a human beingwould 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 fourmajor blocks, which are Fuzzification, KnowledgeBase, Inference Mechanism and Defuzzification.
The following simulations are done in order tosee the performance of the proposed PIDcontroller. The controller parameters are alldetermined using trial and error method. TheMatlab/Simulink simulation model of the proposedPID controller is shown in Figure 1.After several
trial and error runs, the controller parameters of theclassical PID controller are set to Kp=1, Ki=0.2,and Kd=0.01 to provided the desired response.
__________________________________________________________________________________________________________28 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
30/45
Figure 2. Functional Block Diagram of Fuzzy Logic Control System
A.
Fuzzification:
The grain size control of the 0.3% carbon steel isdesigned based on 2 inputs and 1 output. Inputs forthis controller are error and change in error and the
output is grain size. Fuzzification is where we definethe quantization and membership function for theinput and output variables in universe of discourse. Itinvolves the conversion of the input and outputsignals into a number of fuzzy represented values(fuzzy sets).
5.1 Error (e): It is quantized into three fuzzy sets asnegative error (N), zero error (Z) and positive error(P). Figure 3 shows the error input variable,
Figure 3. Error Range
5.2 Change in Error (ce): It is quantized into threefuzzy sets as negative change in error(N), zerochange in error(Z) and positive change in error(P).Figure 4 shows the change in error input variable,
Figure 4. Change in Error Range
5.3 Output: It is quantized into three fuzzy sets asnegative output (N), zero output(Z) and positiveoutput(P). Figure 5 shows the change in 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:
Fuzzy logic uses a set of rules to define itsbehavior. The rules define the conditions expectedand outcomes desired with if/then statements. Theserules replace formulas. They must cover all situationsthat may occur but are not to be written for everypossible combination. The rules are expressed interms of linguistic or fuzzy variables which are
adjectives like large positive error, small positiveerror, zero error, small negative error and largenegative error, which modify the variable. Forminimum of variables it can simply have positive,zero and negative as variables for each of the controlparameters. For microstructure development, stain,strain rate and temperature are the controlparameters. The fuzzy rule table for the triangularmembership function for grain size is formed asshown in table 1.
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 29
-
7/21/2019 jcscs vol 5 nr 2 october 2012
31/45
Figure 6. Block Diagram to Optimize Grain Size by FLC
Table 1. Fuzzy Rule Table
C. Inferencing:
The inference mechanism provides the mechanismfor invoking or referring to the rule base such that theappropriate rules are fired. There are severalmethodologies to derive the inferencing method.Two most common methods used in FLC are themax-min composition and max(algebraic) productcomposition. The inference or firing with this fuzzyrelation is performed via the operations between thefuzzified crisp input and fuzzy relation representingthe meaning of the overall set of rules. As a result ofcomposition, one obtains the fuzzy set describing thefuzzy value of the overall control output. For thissystem, max-min composition is used for theinferencing.
D. Defuzzification:
The function of defuzzification is scale mappingwhich converts the range of values of output variableinto corresponding universe of discourse and it yieldsa non-fuzzy (crisp) control action. For this system,centroid method is used for defuzzification. It isgiven by the algebraic expression,
dzz
dzzz
c
c
).(
.).(*
The Matlab/Simulink simulation model of theproposed fuzzy logic controller is shown in Figure
6.After several trial and error runs, the controller hasthe followingparameter = 1.49, Ke=150 and Kp= 0.1 to provided the desired response.
VI. SIMULATIONAND ERRORCALCULATION
The process control parameters strain, strain rateand temperature are optimized for a required grainsize of 26m from an initial grain size of 180m andits corresponding trajectories are shown. The timetaken is in seconds.
A.
PID Controller Result:
The PID controller output for 26m, 30m and35m are given in figure 7, figure 8 and figure 9respectively.
Figure 7. Response for Grain Size of 26m
Figure 8. Response for Grain Size of 30m
__________________________________________________________________________________________________________30 Volume 5, Number 2, October 2012
-
7/21/2019 jcscs vol 5 nr 2 october 2012
32/45
Figure 9. Response for Grain Size of 35m
B.Fuzzy Logic Controller Result:
The fuzzy logic controller output for 26m, 30mand 35m are given in fig. 10, fig. 11 and fig.12respectively.
Figure 10. Response for Grain Size of 26m
Figure 11. Response for Grain Size of 30m
Figure 12. Response for Grain Size of 35m
C. Error:
The following Integral Square Error and IntegralAbsolute Error values have been obtained using PIDand FLC.
Table 2. Error Comparison
Setpoint
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
VII.CONCLUSION
The dynamic model for 0.3% carbon steel formicrostructure control is developed. The steady statevalue for strain, strain rate and temperature to obtaingrain size from 180m to 26m are selected as 1, 1and 1100 respectively. The dynamic model issimulated by PID controller and fuzzy logiccontroller separately to optimize grain size from
180m to 26m. In both the case, a simulation timeof 10 seconds is considered and the optimization isdone. The integral square error and integral absoluteerror are also calculated. It is observed that thesettling time is less and also the ISE and IAE arecomparatively less in the case of PID.
The PID controller has only three parameters toadjust. It is commonly used to regulate the timedomain behavior of dynamic system. Controlledsystem shows good results in terms of response timeand precision when these parameters are adjustedwell.
A fuzzy logic controller has a lot of parameters to
adjust. The most important thing is to make a goodchoice of rule base and parameters of membershipfunctions. It is sensitive to the distribution ofmembership functions but not to the shape of themembership functions. It doesnt have much bettercharacteristics in time domain. One of the mostimportant problem with FLC is that the computingtime much longer than the PID controller, because ofthe complex operations as fuzzification andparticularly in defuzzification. In addition, with thegrowing requirement of the system performance, thefuzzy membership functions and the interferencerules becomes more and are complicated.
From these results, PID seems to be better choicefor optimization of process control parameters.
__________________________________________________________________________________________________________Journal of Computer Science and Control Systems 31
-
7/21/2019 jcscs vol 5 nr 2 october 2012
33/45
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 theoryprinciples to optimization of grain size during hot
extrusion,Materials Science and Technology, Vol14, pp.25-31, Jan 1998.
[3] James C Malas et.al.,Optimization of microstructuredevelopment during hot working using controltheory,Metullurgical and Materials TransactionsA, Vol 28 A, pp.1921-1930, Sep 1997.
[4]. L.W. Ma, X. Wu and K. Xia,Microstructure andproperty of a medium carbon steel processed by equalchannel angular pressing, materials forum volume32 2008, Edited by J.M. Cairney, S.P. Ringer and R.Wuhrer Institute of Materials EngineeringAustralasia Ltd.
[5] S.Venugopal, Optimization of MicrostructureDevelopment during Deformation processingUnder Dynamic Conditions, Journal of Proc.of theWorkshop on Dynamic Processing of materials,(2000).
[6] Katsuhiko ogata, Modern Control Engineering: Thirdedition, 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 texturedevelopment during hot deformation of magnesium
alloy AZ31,Volume 19, Issue 1,February 2009.[9] Monika HRADILOVa,b, Frank MONTHEILLET,
Anna FRACZKIEWICZ, Christophe DESRAYAUDand Pavel LEJEK, Microstructure development ofca-doped mgzn alloy during hot deformation , 23.25. 5. 2012, Brno, Czech Republic, EU.
__________________________________________________________________________________________________________32 Volume 5, Number