2
The Spot 2009 Proceedings
Theme:
The Spot is a conference dedicated to final year undergraduates, graduates and master students in
engineering fields.
Organizers staff and reviewers –final year master students, PHD students, recently (1-2 years)
employed engineers.
A minimum number of experienced consultants will participate (to keep the student oriented profile
of the event).
Objectives:







to encourage students
to give a good starting point for tomorrow’s scientists
an opportunity for young engineers to write papers
an opportunity for young engineers to organise a scientific conference
to prepare students for future conferences
low cost event
flexible event – people who will not people who will not be able to come on location, they have
the possibility of presenting their work online
Conference Orientation:
Papers from all fields of engineering.
Conference Location:


Aalborg University, Institute of Energy technology, Pontoppidanstræde 101, 9220 Denmark
Video broadcast on the internet
Conference Dates:




Announcement and call for papers: march the 1st 2009
Submitting of abstracts: April the 3rd 2009
Deadline for announcement of acceptance July the 3rd 2009
conference: October the 2nd 2009
Conference Proceedings Published: ISBN 978-87-89179-84-1
The SPOT 2009 – Student Conference
3
Advisors:
Name
Title
Origin
Field
Ewen Ritchie
Assistant
Professor
Institute of Energy Technology,
Denmark
Electrical Energy
Argeseanu Alin
Lecturer
Polytechnic University
of Timisoara, Romania
Electrical Energy
Name
Title
Origin
Field
Ana Moldovan
PHD Student
Polytechnic University
of Timisoara, Romania
Electrical Energy
Veronica Panaite
R&D Engineer
kk-electronic, Denmark
Electrical Energy
Krisztina Leban
Master Student
Institute of Energy Technology,
Denmark
Electrical Energy
Milosz Miskiewicz
Master Student
Institute of Energy Technology,
Denmark
Electrical Energy
Luminita Barote
PHD Student
Transilvania University of Brasov,
Romania
Electrical Energy
and Computer
Science
Suzan de Goede
Master student
Department of health sciences,
Denmark
Biomedical
Engineering
Name
Title
Origin
Field
Irina Stan
Master Student
Institute of Energy Technology,
Denmark
Electrical Energy
Reviewers:
Secretary:
Contact:
Website: http://grou.ps/the_spot_2009
Email: The_Spot_2009@yahoo.com
The SPOT 2009 – Student Conference
4
The SPOT 2009 – Student Conference
5
Choosing an electric motor for a bicycle,
performance prediction
Cristian Busca,
Master student at Aalborg
University, Department of Energy
Technology,
Pontoppidanstræde 101, 9220
Aalborg East, Denmark Aalborg,
bkcristi2000@gmail.com
ABSTRACT
This paper focuses on the power use analysis of an electric
bicycle. The power requirements of an electric bicycle depends
on various factors such as: bicycle type, bicycle and rider
weight, road conditions, desired cruising speed and acceleration.
By knowing which factors influence the power requirements of
an electric bicycle, one can choose an appropriate electric motor
available on the market for a conventional bicycle. By using the
data given by the motor manufacturer it is also possible to
predict the performance of the electric bicycle. Performance
prediction means estimating the hill climbing ability, maximum
cruising speed and acceleration. This paper addresses those who
wish to choose an electric drive for their bicycle or who want to
design an electric bicycle from scratch.
General Terms
Documentation, Performance, Design, Economics, Reliability.
Keywords
electric bicycle, electric bicycle performance prediction, electric
motor for a conventional bicycle, bicycle power use analysis,
electric drive of a bicycle.
Adaptive algorithm for SRM drives torque
optimization
Boian Daniela
Dpt. of Electrical Engineering,
Polytechnic Univ. of Timisoara,
Romania
ABSTRACT
The paper shows a new SRM drives control algorithm. The
basic idea of the algorithm is to take in consideration the
sequential movement of the SRM and to use the torque equation
of consecutive steps. The first step offers the information
necessary to compute the optimal turnoff angle and that
information will be used to determine the second step.
The SPOT 2009 – Student Conference
General Terms
Algorithms, Performance, Design, Reliability,
Keywords
Switch reluctance motor, control argorithm
6
Modular Multilevel Inverter Arm for MW
Power Rating
Cristian Sandu, Nicoleta Carnu, Valentin Costea
Master students at Aalborg University, Department of
Energy Technology,Pontoppidanstræde 101, 9220
Aalborg East, Denmark Aalborg
sanducristian@gmail.com ; carnunicoleta@gmail.com ;
vali12002@yahoo.com
ABSTRACT
This paper presents the design and control of a modular multilevel
inverter leg. The proposed topology of the inverter arm has sixteen
full H-bridge modules. The design aspects regarding the power
semiconductors ratings and attainable efficiencies are examined. The
dedicated control algorithm for this inverter configuration is staircase
modulation. Its implementation is facilitated by the inverter’s
structure. The inverter modules can be independently controlled thus
lowering the commutation frequency and so the commutation losses.
In order to maintain a constant voltage across the capacitors, voltage
Stig-Munk Nielsen
Aalborg University, Department of Energy Technology,
Pontoppidanstræde 101, 9220 Aalborg East, Denmark
Aalborg
smn@iet.aau.dk
feedback is required. The even power distribution among modules is
achieved by using voltage balancing. Further the inverter is analyzed
and simulated. The experimental setup is described with clear
reference to hardware and to the main control.
General Terms
Algorithms, Management, Design
Keywords
Design, multilevel inverter, control
Autonomous mobile robots avoiding
obstacles using ultrasonic sensor or video
camera
Dan Novischi
Department of Electrical Engineering,
Polytechnic University of Bucharest
Splaiul Independentei Nr. 313,
Bucharest, Romania
dan.novischi@gmail.com
ABSTRACT
This paper presents two mobile robots that have the ability to
avoid obstacles when going from a start point to an end point.
The robots were built using the NXT Lego Mindstorms kit: one
uses an ultrasonic standard sensor (US) and the other a custom
vision subsystem MindSensors NXT camera. The software was
written in two languages: NXC (NXC/NBC1.7 firmware) and
Java (Lejos0.7 firmware). The main algorithm, implemented
both with ultrasonic sensor and camera, computes the minimum
The SPOT 2009 – Student Conference
trajectory for avoiding the obstacle. Experiments were
performed to evaluate the robots performance.
General Terms
Algorithms, Management, Measurement,
Performance, Design, Experimentation.
Documentation,
Keywords
mobile robots, ultrasonic sensor, video camera, Bluetooth
communication, obstacle avoidance, detection, travel,
processing, experiments
7
Design of Inverter-Fed SPMSM-Motor Drive
Line in a FC Truck System
Jorge Varela Barreras
Institute of Energy Technology
Aalborg University
Pontoppidanstræde 101, 9220
Aalborg East, Denmark Aalborg,
ABSTRACT
The purpose of this paper is to design and implement an
efficient control for SPMSM, between their upper and lower
speed limits, under the requirements of a FC truck system. The
simulation of the SPMSM, truck load, VSI, modulation strategy
and control structure is performed in MATLAB/Simulink. For
the simulations, real parameters of the provided SPMSM
obtained in the laboratory are used. A Field Oriented Control
strategy is implemented and tested on several situations.
General Terms
Algorithms, Measurement, Performance, Design
Keywords
SPMSM, VSI, SVM PWM, FOC, Ackerman steering geometry.
Doubly Fed Induction Generator Fault
Simulation
Krisztina Leban, Ewen Ritchie,
Aalborg University,
Pontoppidanstræde 101, 9220
Aalborg, Denmark
Alin Argeseanu
Ileana Torac
Politehnica University Timisoara, Bl. V. Romanian Academy – Timisoara
Parvan 2, 300223, Romania
Branch. Bl. M. Viteazu 24, 300223,
Romania
alin_argeseanu@yahoo.com
krisztina_leban@yahoo.com.au
ABSTRACT
This paper focuses on validating a simplified wind turbine
system having a fault rid through protection. The machine used
was a doubly fed induction generator (DFIG). As protection
against short circuit transients, the crowbar protection was
employed in the simulation. An equivalent model was
constructed. Simplifications were made so as to have a system
composed of grid, transformer, line and generator represented by
elementary circuit elements (R, L, C and voltage sources).
Equivalent circuit models were simplified so that the fault
models may be used for synchronous machine parameters. The
assumption that the mechanical system cannot respond during
The SPOT 2009 – Student Conference
ileana_torac@yahoo.com
the short time of a three phased short circuit was made. The
simplified simulation model was compared to simulations
constructed with complex library subsystems. The SimPower
Systems Simulink library was used for this purpose.
General Terms
Documentation, Performance, Verification.
Keywords
wind energy plant, variable speed generator, induction
generator, double fed induction generator, DFIG simplified
model, Simulink model
8
Magnetic Field Enhancement using
Ferrofluid and Iron Powder
Krisztina Leban, Ewen Ritchie,
Alin Argeseanu
Aalborg University,
Pontoppidanstræde 101, 9220
Aalborg, Denmark
Politehnica University Timisoara, Bl. V.
Parvan 2, 300223, Romania
alin_argeseanu@yahoo.com
krisztina_leban@yahoo.com.au
ABSTRACT
This paper focuses on studying the variation the magnetic field
of a PM when covered with ferrofluid and iron powder. For
testing purposes, four neodymium magnets were used. One was
covered with ferrofluid; one was left ‘clean’; one covered with
magnetic powder and one with ferrofluid and iron powder
combined. The magnetic field growth was quantified through
measurements and the results presented in the paper
Keywords
General Terms
Ferrofluid, iron powder, magnetic field
Documentation, Performance, Verification.
Magneto rheological (MR) Fluid
Alin Argeseanu
Krisztina Leban, Ewen Ritchie,
Aalborg University, Pontoppidanstræde 101, 9220
Aalborg, Denmark
krisztina_leban@yahoo.com.au
ABSTRACT
This paper presents a novel soft actuator using magneto
rheological fluid enclosed in flexible membranes [1]. The goal
was to obtain the largest amount of force with a given excitation
and layered membranes. The amount of force resulted from the
interaction between the multilayered membranes filled with the
magnetic field produced by the excitation was measured. The
The SPOT 2009 – Student Conference
Politehnica University Timisoara, Bl. V. Parvan 2,
300223, Romania
alin_argeseanu@yahoo.com
phenomena were described and resulting forces are presented
and explained herewith..
General Terms
Documentation, Design, Performance, Verification.
Keywords
Actuator, magneto rheological (MR) fluid
9
Partial Discharge Data Denoising and
Feature Extraction
S.Vivekananthan, S.D.R.Suresh
Department of Electrical &Electronics
Engineering, College of Engineering,
Guindy, Anna University, Chennai-25
ABSTRACT
One of the major challenges of modern-day on-line partial
discharge (PD) measurement in High voltage electrical
equipments is the recovery of PD signals from a noisy
environment. The different sources of noise include thermal or
resistor noise added by the measuring circuit and high-frequency
sinusoidal signals that electromagnetically couple from radio
broad casts and/or carrier wave communications. Sophisticated
methods are required to detect PD signals correctly. Fortunately,
advances in Analog-to-Digital conversion (ADC) technology,
and recent developments in Digital Signal Processing (DSP)
enable easy extraction of PD signals. This paper deals with the
analysis of noise and denoising of PD signals using wavelet
method. This denoising method is employed on both simulated
as well as real PD data using MATLAB. Also statistical analysis
is done on different kind of partial discharges for feature
extraction.
General Terms
Documentation, Security, Standardization
Keywords
Partial discharge, wavelet, Threshold, Corona discharge, Denoise, Coupling capacitor, High voltage
Brain Computer Interfaces: to optimize or to
commercialize?
Suzan de Goede
Center for Sensory-Motor Interaction,
Department of Health Science and
Technology, Aalborg University,
Aalborg, Denmark;
ABSTRACT
Brain computer interfaces (BCI) provide a mean of
communication for severely paralysed patients. These systems
bypass the brain’s usual output channels, such as the muscles or
speech and use the brain signals to control an external device.
Currently, there are no BCI systems commercially available for
paralysed patients. A review of the literature has been performed
to assess what the current state of art in BCI technology is. An
analysis is made of the capabilities and limitations of clinically
applied BCI. The performance and sophistication of the
currently available BCI technology is at a high level. However,
The SPOT 2009 – Student Conference
for a commercial system to be marketed, some gaps need to be
bridged. Some essential requirements for a BCI system to be
used on day-to-day basis will be discussed.
General Terms
Algorithms, Management, Measurement, Documentation,
Performance, Design, Economics, Reliability, Experimentation,
Security, Human Factors, Standardization,
Keywords
Brain computer interfaces, brain signals, sensorimotor rhythm
10
Identification of biomarkers of renal failure
and renal dialysis using metabolomics
Munsoor Hanifa,
Allan Stensballe, Reinhard
Dpt. of Biotechnology, Chemistry and
Wimmer, Kim Esbensen
Environmental Engineering
Aalborg University, Denmark, ACABS
Research Group, Aalborg University
ABSTRACT
This paper focuses on validating a simplified wind turbine
system having a fault rid through protection. The machine used
was a doubly fed induction generator (DFIG). As protection
against short circuit transients, the crowbar protection was
employed in the simulation. An equivalent model was
constructed. Simplifications were made so as to have a system
composed of grid, transformer, line and generator represented by
elementary circuit elements (R, L, C and voltage sources).
Equivalent circuit models were simplified so that the fault
models may be used for synchronous machine parameters. The
assumption that the mechanical system cannot respond during
Helmut Meyer-Hofmann, Troels
Ring
Nyremedicinsk Afdeling, Aalborg
Hospital Aalborg, Denmark
the short time of a three phased short circuit was made. The
simplified simulation model was compared to simulations
constructed with complex library subsystems. The SimPower
Systems Simulink library was used for this purpose.
General Terms
Documentation, Performance, Verification.
Keywords
wind energy plant, variable speed generator, induction
generator, double fed induction generator, DFIG simplified
model, Simulink model
Switch Mode Power Supply for Inverter
Application
Cristian Sandu, Nicoleta Carnu,
Valentin Costea
Master students at Aalborg University,
Department of Energy Technology,
Pontoppidanstræde 101, 9220 Aalborg ,
Denmark vali12002@yahoo.com
Stig-Munk Nielsen
Aalborg University, Department of Energy
Technology, Pontoppidanstræde 101, 9220
Aalborg East, Denmark Aalborg
smn@iet.aau.dk
General Terms Design, Verification
ABSTRACT
The aim of this paper is to design a switch mode power supply
for auxiliary systems belonging to the inverter such as: control
and monitoring systems. The switch mode power supply
provides multiple voltage outputs necessary for the control or
other auxiliary systems. This kind of supply is a step down or a
buck converter.
The SPOT 2009 – Student Conference
Keywords
Inverter, switch mode power supply
11
Experimental Analysis of the Magnetic
Behavior of NdFeb Permanent Magnets
Veronica Panaite
Ewen Ritchie,
Aalborg University,
Pontoppidanstræde 101, 9220
Aalborg, Denmark
Aalborg University,
Pontoppidanstræde 101, 9220
Aalborg, Denmark
ABSTRACT
This paper presents a possible method to analyse the NdFeB
permanent magnets. They are powerful magnets, but they have a
low Curie temperature. This parameter influences their magnetic
properties, mainly modifying the magnets’ hysteresis. This
means both the magnetization and the demagnetization curves
can be modified. Two experiments have been performed on
some samples of NdFeB magnets in order to analyze the effect
of applied magnetic field to them.
General Terms
Documentation, Performance
Keywords
NdFeB, Neodymium-Iron-Boron, Rare-Earth Permanent
Magnet, Magnetic Properties
Types of NdFeB Permanent Magnets and
Application Possibilities
Veronica Panaite
Ewen Ritchie,
Aalborg University,
Pontoppidanstræde 101, 9220
Aalborg, Denmark
Aalborg University,
Pontoppidanstræde 101, 9220
Aalborg, Denmark
aer@iet.aau.dk
ABSTRACT
Keywords
This paper presents the main types of neodymium-iron-boron
(NdFeB) permanent magnets and their application. They are
strong magnets, but susceptible to corrosion. Therefore
protective coatings are important for NdFeB magnets. Since
they are powerful magnets, they are sought to be implemented in
magnetic related applications. The NdFeB magnets can be used
in all magnet-related fields, as table 4 shows.
NdFeB, neodymium-iron-boron, sinterred magnet, bonded
magnet, applied permanent magnet
General Terms
Documentation, Performance
The SPOT 2009 – Student Conference
1
Choosing an electric motor for a bicycle,
performance prediction
Cristian Busca, Master student at Aalborg University, Department of Energy Technology,
Pontoppidanstræde 101, 9220 Aalborg East, Denmark Aalborg, bkcristi2000@gmail.com

Abstract—This paper focuses on the power use analysis of an
electric bicycle. The power requirements of an electric bicycle
depends on various factors such as: bicycle type, bicycle and
rider weight, road conditions, desired cruising speed and
acceleration. By knowing which factors influence the power
requirements of an electric bicycle, one can choose an
appropriate electric motor available on the market for a
conventional bicycle. By using the data given by the motor
manufacturer it is also possible to predict the performance of the
electric bicycle. Performance prediction means estimating the hill
climbing ability, maximum cruising speed and acceleration. This
paper addresses those who wish to choose an electric drive for
their bicycle or who want to design an electric bicycle from
scratch.
Index Terms—electric bicycle, electric bicycle performance
prediction, electric motor for a conventional bicycle, bicycle
power use analysis, electric drive of a bicycle.
I. INTRODUCTION
N
owadays transportation is an important issue. It is well
known that the fossil fuel reserves become smaller as time
passes. This is also shown by the continuous price rise of gas.
On the other side, the need for transportation increases more
and more. Most of today’s means of transportation like cars,
trains, ships, busses and airplanes use fossil fuels as a source
of energy. There are only a few exceptions like some electric
cars and electric busses.
Even if the efficiency of today’s internal combustion engine
has improved considerably in comparison with the first
combustion engines, they are still polluting and releasing CO2
into the atmosphere.
A good alternative to all these polluting transportation
vehicles would be an electric vehicle. An electric vehicle
doesn’t use fossil fuel as s source of energy. It uses electricity
as energy source. The electricity stored in batteries powers an
electric motor which propels the electric vehicle. Electric
vehicles have many benefits over classical internal
combustion engine vehicles. Some of the benefits are: electric
vehicles do not release CO2 into the atmosphere, they produce
much less noise than conventional vehicles, and electric
vehicles last much longer than conventional vehicles because
they don’t have as many moving parts as conventional
vehicles due to the internal combustion engine. The main
drawback of an electric vehicle is the limited range due to the
limited energy storage capacity of the battery.
In order to overcome this problem it would be necessary to
develop new battery technologies which would allow much
higher energy densities than today’s batteries.
Compared to other forms of transportation the classical
bicycle can be considered one of the most efficient means of
transportation. To travel 1km by bicycle requires
approximately 5…15Wh of energy. To travel the same
distance by foot requires 15…20Wh of energy, 30…40Wh by
train and over 400Wh in a car.[3]
A car is a very inefficient mean of transportation because it
is heavy while a bicycle is light. A typical car can weigh
between 600…2000kg while a typical bicycle weights
between 10…25kg. This reduced weight is what makes the
bicycle one of the most efficient means of transportation.
The problem with a conventional bicycle is that it becomes
impractical for traveling long distances especially on hilly
terrain. This issue could be solved by the use of an electric
motor which would provide extra power or provide all the
power necessary to propel the bicycle. This way the bicycle
would become practical for traveling longer distances in
shorter time than pedal power alone.
The first electric bicycles appeared in the 1970s following
the energy crysis.[4]
There are several models of electric bicycle on the market
today with different motor and battery specifications. It is still
expensive to buy an electric bicycle and not everyone can
afford it.
For those who already own a bicycle and do not want to
spend money on a new electric bicycle a possibility would be
to buy an electric motor for their bicycle (electric bicycle
conversion kit). The hard part comes when one has to decide
which motor to buy.
This paper attempts to explain which factors influence the
specifications of the required electric motor and what will be
the expected performance of the electric bicycle. A detailed
analysis of the bicycle’s power use breakdown will be
presented. This way the reader will be able to choose the best
electric motor available to suit his needs.
Recent European Union regulations regarding electric
bicycles limit the maximum motor assisted speed of an electric
2
bicycle to 25km/h, the maximum continuous output power of
the electric motor to 250W and the maximum weight of the
electric bicycle to 40kg.[6]
II. BICYCLE POWER USE ANALYSIS
It is important to analyze the bicycle from the power use
point of view because it helps to understand what the input
power is used for.[2]
The equations used to do the calculations will be presented.
The below presented equations are necessary in order to
calculate the power needed to propel the electric bicycle.
A. Bicycle power equations
The power generated by a bicycle rider is transmitted
through the pedals and chain to the back wheel. This input
power is spent to overcome different forward movement
opposing forces.[2]
The total power needed to propel the bicycle may be
calculated using equation (1).
- is the total mass of the bicycle and rider [kg];
- is the gravitational acceleration [m/s2];
- is the angle between horizontal and the road on which the
bicycle is going [degrees].
As it may be seen in equation (3), the force due to rolling
resistance is directly related to the rolling resistance
coefficient, total weight of the bicycle and the cosine of the
angle between the road surface and horizontal.
depends on weather,
The rolling resistance coefficient
road surface type and tire type and pressure. The value of the
rolling resistance coefficient can range from 0.0022 to 0.0039.
For example a mountain bikes with knobby tires has a much
higher rolling resistance coefficient than a racing bicycle.
The force needed to overcome the air drag may be
calculated using equation (4).
(4)
Where:
-is the density of the air [kg/m^3];
(1)
Where: - is the total needed power [W];
-is the total forward movement opposing force [N];
- is the air drag coefficient;
- is the frontal area [m^2];
- is equal to the speed of the bicycle in the case of no
wind [m/s].
- is the desired velocity of the bicycle (road speed) [m/s].
The total force is calculated using equation (2).
(2)
Where:
- is the total necessary force needed to move the
bicycle forward [N];
- is the force needed to overcome rolling resistance
[N];
- is the force needed to overcome air drag [N];
- is the force needed to accelerate the bicycle [N];
- is the force needed for elevation change [N];
- is the efficiency of the bicycle’s drive train.
The force needed to overcome rolling resistance is
calculated using equation (3).
If there is a head wind, than it will be added to the bicycle’s
speed . In the case of a back wind, it will be subtracted from
the bicycle speed.
As shown in equation (4), the air drag force is directly
related to air density, air drag coefficient, frontal area and the
square of the speed.
The interesting thing about the air drag equation is that the
force needed to overcome air drag increases with the square of
the speed. As a result the power needed to overcome air drag
rises with the cube of the speed.
This means that for a doubling of cruising speed the input
power has to be increased around 8 times.
The values of
and depends on the aerodynamics of
the bicycle and rider. The better is the aerodynamics of the
bicycle and rider the lower the values of these coefficients are.
Typical values for the air drag coefficient can range from
0.1 to 0.78. The frontal area can have values from 0.3 to 0.5
m2.
The force needed to accelerate the bicycle may be
calculated using equation (5).
(5)
(3)
Where:
Where:
- is the acceleration [m/s2].
- is the rolling resistance coefficient;
As shown in equation (5), for a quicker acceleration a greater
3
force is needed.
The force needed for elevation change may be calculated
using equation (6).
- is the mechanical power at the traction wheel [W];
- is the torque at the traction wheel [Nm];
- is the angular velocity of the traction wheel [rad/s].
(6)
Where:
- is the slope of the road.
The slope is defined as being the rise with distance. For
example if you go forward 100m and the elevation change is
5m then you have a slope of s=5/100=0.05 (5%).
The force needed for elevation change is zero on a flat
surface, positive when climbing a hill and negative on
downhill.
Negative force means that it will actually help the bicycle to
move forward.
So far the equations which are used to calculate the
necessary power to maintain a constant speed were
presented.[2]
The formula used to calculate the amount of necessary
energy for acceleration is shown in equation (7).[5]
B. Constant speed calculations and results
Using the equations presented at section A, several
calculations have been made. The calculations were done
assuming that the bicycle has a constant speed. This way the
force needed to accelerate is eliminated from the total force
equation.
The issue of acceleration will be discussed after the constant
speed analysis.
The constants and coefficients used for calculation are the
following:[1]
m/s2
kg/m3
2
m
The efficiency of the bicycle’s drive train is considered to
be 95% for all the calculations in this paper, which leads to .
(7)
Where:
- is the amount of energy required for acceleration
from standstill to the desired speed [J];
- is the desired speed [m/s].
As shown in equation (7) the amount of energy needed for
acceleration depends only on the mass of the bicycle and final
speed.
The power needed for acceleration is shown in equation
(8).[5]
The constant speed analysis reveals how much power is
necessary to maintain speed and what is that power spent on.
By doing this kind of analysis it will be possible to determine
the power requirements for riders with different weight,
different road conditions and different bicycle types.
Different cases will be discussed. The first case is shown in
Figure 1 for a total weight of 80kg (rider and bicycle) and a
flat road which means 0% slope.
(8)
Where:
- is the average power needed for acceleration [W];
- is the time needed for acceleration [s].
As it may be observed in equation (8) the average power
needed for acceleration increases as the acceleration time
decreases. With other words said for a quick acceleration a
high average power is needed while for a slow acceleration
low average power is needed.
Equation (9) shows the mechanical equation of the
bicycle.[5]
(9)
Where:
Figure 1. Required power as a function of speed (case 1).
4
As shown in Figure 1, at low speeds all the components of
the forward movement opposing forces contribute almost
equally to the total force, while at higher speeds the air drag’s
contribution becomes predominant. This means that at high
speed the rolling resistance and drive train losses become
negligible.
Figure 2 shows the power use breakdown for the case when
the total weight is 50kg and the slope is 0%.
Figure 3. Required power as a function of speed (case 3).
As shown in Figure 3 the power needed for elevation
change has increased considerably in comparison with the
cases from figures 1 and 2 due to the fact that now the bicycle
is climbing a on a 3% hill.
As it may be seen in Figure 3, the power needed for
elevation change is predominant over the air drag power,
rolling resistance power and drive train losses power. A
considerable amount of energy is necessary for elevation
change. The power needed for elevation change is even
greater that the power needed to overcome air drag up until
the bicycle reaches 36km/h.
The last analyzed situation is that of a downhill run shown
in Figure 4. The total weight is 80kg and the bicycle is
descending with the slope of -3%.
Figure 2. Required power as a function of speed (case 2).
The difference that may be observed in Figures 1 and 2 is
that now the rolling resistance power has increased due to the
increased weight. The air drag power is the same. The power
needed for elevation change is zero in both cases as the
bicycle is going on a flat road.
Figure 3 shows the case when the total mass is 80kg and the
bicycle is climbing a 3% hill.
Figure 4. Required power as a function of speed (case 4).
As it may be seen in Figure 4 the total power need is very
small compared to cases 1, 2 and 3 because in this case the
bicycle is descending on a -3% slope. Thus potential energy is
converted into kinetic energy. The bicycle will slowly
accelerate without any input power from the electric motor or
rider up until it reaches a speed of about 31km/h.
A summary of the four cases shown in figure 1, 2, 3 and 4
is presented in the table from Figure 5. The values were read
for the speed of 25km/h as the regulations regarding electric
bicycles in the European Union limit the maximum motor
assisted speed of an electric bicycle to 25km/h.
5
Figure 5. Summary of the results presented in figures 1, 2, 3
and 4.
As it may be seen in the table from Figure 5 for a total
weight of 80kg on a flat road a total power of 105.97W is
needed to keep the bicycle cruising with 25km/h. The limit of
25km/h has been set due to legislative limitations.[6]
When the bicycle is climbing a 3% hill the total needed
power is 277.64W. The weight has little impact on the total
needed power if the bicycle runs on a flat road.
C. Calculation of the power required for acceleration
As it may be seen in equation ()
III. CONCLUSION
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
Biomechanics and biology of movement Af Benno Maurus Nigg,Brian
R. MacIntosh,Joachim Mester
http://www.sheldonbrown.com/rinard/aero/aerodynamics.htm
The Energy Cost of Electric and Human-Powered BicyclesJ.
http://www.varsitybike.com/electric-bike-history-facts/
http://www.cameronsoftware.com/ev/EV_CalculateMotorSize.html
http://www.ultramotor.com/uk/electric_bikes_legislation
Adaptive algorithm for SRM drives
torque optimization
Abstract
The paper shows a new SRM drives control algorithm. The basic idea of the algorithm is to take
in consideration the sequential movement of the SRM and to use the torque equation of
consecutive steps. The first step offers the information necessary to compute the optimal turnoff angle and that information will be used to determine the second step.
Keywords: switched reluctance machine, torque maximization model, triangular inductance
model, trapezoidal inductance model
1. Introduction
The Switched Reluctance Motor is the simplest of all electrical machines. The switched
reluctance motor is an electric motor with salient poles, in which torque is produced by the
tendency of the rotor to move in position where the inductance of the excited windings is
maximized and the reluctance is minimized. Excitation is a sequence of current pulses to each
phase in turn. Figure 1 show a SRM with three phases, six stator poles and four rotor poles.
The strong nonlinearity in the flux / current / position and torque / current / position curves
makes in principle the SRM apparently less attractive for servos. However the motor simplicity
and ruggedness in such applications may not be overlooked.
Figure 1. A 6/4 SRM section
2. Torque maximization model
The control of SRM drives depends on the absolute rotor position, phase current, rotor speed.
Many papers deal with the torque computation and torque ripple minimization for the
Switched Reluctance Drives (SRD). The phase inductance shape may be triangular or
1
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
trapezoidal. Both situations have been treated in this paper. In order to calculate the maximum
torque, we must make some simplifying considerations. We suppose that once the maximum
current obtained, it remains constant until the phase is switched off. Another supposition is
that the decreasing current shape is a straight line. And the slope of the decreasing current
shape remains constant.
The instantaneous SRM torque is given by:
q
M
(1)
M j
j 1
Where q is the number of SRM phases.
The medium torque developed by the SRM is given by [1][3]:
q
M

p
 p  1
(2)
 M jd
 1
2.1 Optimal turn-off angle calculus for triangular inductance model –infinite
permeability
The model is presented in Figure 2.
The instantaneous torque in this case is [2]

0
 1    0

M i   0.5  K i  i 2
0 
 0.5  K  i 2       
i
p
1

(3)
The instantaneous inductance is:

Lu
 1    0

Li   Lu  K
0   
L  K (  2  )       
p
1
 u
(4)
The instantaneous current is:
i10   [ 0 ,0]


ii  i11   [0, c ]

i12   [ c , e ]
(5)
The medium torque developed by the SRM, considering all three intervals is:
0

p1
q
Mm  ( Mid Mid  Mid)
p 1
0

(6)
2
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
For the first interval the torque developed is zero.
For the second and third intervals, the developed torques expressions are:
  c3

 (   a) c2 
2 

Ki
3

M 2  112 
2a 
3
2
2
2
 ( 2a   ) c  3  a   a 
  c3
2
(  (    a ) c ) 
 3
Ki112 
3
M 3   2  ( a 2  2a   2 ) c 
2a
3

3
a
 2
2
 a   a  3

(7)









(8)
By (6),(7),(8) the torque development becomes:
Mm 

Ki112  2 3
2 3
a3
2
2
2



2
(


a
)



(
4
a


2


a
)




 2a 2   2a 2  (9)
c
c
c

2
3
3
2a  3

where: a   e   c
(10)
For a triangular shape of phase inductance, the maximum motor torque is
produced when the current phase turn-off angle is given by:
c    a 
2
a
2
(11)
where:  is the maximum inductance angle and  e is the zero-current angle.
Figure 2. Current and phase inductance shapes for infinite permeability
3
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
Starting from the measured  off angle of the previous phase and considering this recursive
equation, an optimal  off angle for each phase, in real time, can be calculated.
2.2 Optimal turn-off angle calculus for trapezoidal inductance model –finite
permeability
The model is presented in Figure.3:
Figure 3. Current and phase inductance shapes for finite permeability
The instantaneous torque in this case is:

0
 1    0

2
0    s
 0.5  K i  ii
Mi  
0
 s    r

 0.5  K i  ii2  r     r   s

(12
The instantaneous inductance is:

Lu
 1    0

Lu  K  
0    s

Li  
La
s    r

 Lu  K  (   r   s )  r     r   s

and K 
La  Lu
s
(13)
(14)
Using the same procedure, the medium torque developed by the SRM, considering all intervals,
is given by:
4
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
 2 3
 c  ( s   r  2a)c2  (2a s   s2   r2  2a r  a 2 )c
2 
Ki
3
M m  112 
3
2a   s   r3 a 3
  a( s2  r2 )  a 2 ( s   r )

3
3



 (15)



For a trapezoidal shape of inductance, the off optimal angle is given by:
c 
s  r
a 2 ( s   r ) 2
a

2
2
4
(16)
Where:  s and  r are the two maximum inductance angles  s <  r
3. The algorithm implementation
The important observations are the simple form of the turn-off angles and the expression
generalizations. The mathematical model uses a general model of SRM and the model don’t
works with SRM geometry, dimensional parameters, materials performance, load or speed. In
the same time, the algorithm needs only few general information  e ,  c ,  or  e ,  c ,  s  r . All
these data are able from both strategies: with sensors or sensor less.
The essential idea of the algorithm is to consider the sequential movement of the SRM and to
use couple of consecutive steps. The first step offers the information necessary to compute the
optimal turn-off angle and that information works in the second step.
If a step sequence is Sn-1, Sn, Sn+1…the algorithm works:
-the step Sn-1 is the source of information for the step Sn and uses the computed optimal angle
with the data from Sn-2 step
-the step Sn offers data for the next step Sn+1 and uses data computed in the Sn-1 step
The accuracy of the algorithm is sufficient for high-performance applications because two
consecutive steps are quasi-identically and that presumption is acceptable and works well in
low-speed region or in high-speed region.
For the initial step, the algorithm accepts an ordinary value for the turn-off angle  . That value
is not an optimal one, because  is not computed. For each step  is the poles alignment
position.
Another important observation regards the algorithm necessary data. In the first SRM model
(model of infinite permeability, triangular model) the algorithm data are:
-  -the pole alignment position can be obtained with sensors or sensor less strategies
-  e -the zero current point
On the second model (model of finite permeability, trapezoidal model) the measured data are:
s and r, the two maximum inductance angles, s < r
5
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
-  e -the zero current point
All these data can be obtained with sensors methods or sensor less methods
The start steps encompass the new algorithm. With these observations briefly introduced here
it is now possible to describe the algorithm:
The start sequence:
-it is possible to use a sensor less estimation scheme to determine the exact rotor position
and to choose correctly the phase who must to feed. The start sequence has 2-3 steps with
c  
-a low cost solution uses only a start sequence, 2-3 steps with  c   ,without the correct
estimation of the rotor position (usual, SRM works better with long impulses for the start
conditions). The first phase is selected arbitrary.
-in both options, the direct start technique doesn’t use the optimized turn-off angle
-The optimal phases feed sequence:
-the last step from the start sequence becomes the initial step S0 of the algorithm
-the initial step offers the measurement data for the computation of the optimal angle  c
used on the next step S1
-the S1 is the first complete step of the algorithm: S1 uses the computed optimal angle and, in
the same time, offers the data set for the next computation of the optimal angle
-a certain steps sequence is Sn-1, Sn, Sn+1…the algorithm works according to that rule
3. The experimental set-up
The experimental set-up is presented in Figure 4:
Figure 4. The experimental set-up
6
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
The elements of the experimental set-up are: an SRM (the topology of the motor
Is: 3 phases, 6 stator poles, 4 rotor poles), a electromagnetic brake type FRAT120, a torque
transducer type FAST TM-HR-RD-7.5 Nm, an incremental encoder type XCC-1510T and a
special position sensor in accord which SRM topology.
The SRM control is made using Matlab/Simulink and dSpace. The phases currents control and
the switching signals are presented in the Figure 5
Figure 5. The phase currents control
The global control-currents, positions, speeds use the graphical; interface
presented in Figure 6:
7
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
Figure 6. The parameters control
4. The analysis of SRM dynamics
The analysis of SRM dynamics is described by the calculation of successive steps for 100 steps.
For this it will be analyzed the expression of optimal energized angle, in the SRM unsaturated
case model (the triangular model of inductivity). The expression is:

2
    0,3a
 c    a1 
2 

Where:  -constant (for a given SRM)
a   e   c  f L, i , w 
a –depends of the SRM type, of speed charge (of parameters which describe the dynamic
SRM behavior)
For realizing a simple command it will be accepted next approximation:

It will be considered two time moments which describe the sequential SRM
function, noted with n and (n+1).
8
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)

The two situations noted with n and (n+1) describe the SRM at steps n and (n+1).

In the two moments we will consider that the dynamics parameters of
electromechanical conditions constants:
I n  I n 1  L ; in  in 1  I

;
wn  wn 1  w
Possible differences (especially during the transitional arrangements: power,
braking, acceleration, load torque) at consecutive steps are negligible.

In constant regimes (speed, torque) previous assertions are obvious.

Based on these allegations can be concluded that the level of two consecutive
steps, electrical and mechanical parameters of the SRM is considered constant.
It was done three measurements (at 100 ms) at different charges:
9
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
From the graphics can be observed that the error at different charges is negligible. Otherwise,
two successive steps are not very different one of each other, and with some approximation
can be considered identical.
With the conclusion above it can be applied a predictive algorithm, which is measurement in
step n and command in step (n+1).
5. Conclusion
The operating performance of the algorithm in steady mode is comparable with
standard control SRM system. The analytical forms of the optimization
predictive algorithm are very simple (see the equations 11 and 16) and offer an
interesting opportunity in a low cost control strategy of the SRM drives.
References
[1] R.C. Becerra, M. Ehsani and T.J.E. Miller, “Commutation of SR Motors”, IEEE Transactions on
Power Electronics, vol.8, no.3, July 1993.
[2] A Argeseanu, C.Şorândaru “New Self Adaptive Optimal Control for general SRM ”, Proceedings
12th National Conference of Electrical Drives CNAE '2004, Cluj-Napoca (Romania):
[3] T.J.E. Miller, “Switched Reluctance Motors and their Control”, Magna Physics & Clarendon Press, Oxford,
1993.
10
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
Page |1
Abstract - This paper presents the design and
control of a modular multilevel inverter leg. The
proposed topology of the inverter arm has sixteen full Hbridge modules. The design aspects regarding the power
semiconductors ratings and attainable efficiencies are
examined. The dedicated control algorithm for this
inverter configuration is staircase modulation. Its
implementation is facilitated by the inverter’s distinctive
structure thus. The inverter modules can be
independently
controlled
thus
lowering
the
commutation frequency and so the commutation losses.
In order to maintain a constant voltage across the
capacitors, voltage feedback is required. The even power
distribution among modules is achieved by using voltage
balancing. Further the inverter is analyzed and
simulated.
The experimental setup is described with clear
reference to hardware and to the main control
algorithm.
I.
inverter leg. Its configuration is presented in
figure 1. The inverter arm has sixteen full H-bridge
modules. Each module contains four IGBT’s and
one capacitor. All modules have in parallel with
the capacitors, voltage sensors.
INTRODUCTION
The concept of multilevel converter originates
from the idea of step approximation of the
sinusoidal voltage waveform [1]. In recent years
an increasing research interest has been
orientated towards this area due to these
converters capability to increase the output
voltage magnitude at a relatively lower switching
frequency than that of a conventional two level
converter. This results in a reduced voltage stress
on each semiconductor [2]. The existing
converters used in HVDC applications are either
two or three level technology. Recent research [3]
[4], revealed that multilevel cascaded converter
configurations are most effective out of the three
aforementioned, especially for wind energy
applications. Therefore the aim is to design, build
and control a bidirectional modularized multilevel
Figure 1. Proposed configuration for the bidirectional DC/DC power
inverter
There is a specific application in wind power
system field that can benefit from the use of
modular bidirectional H-bridge converters. That is
the standard DC based wind farms solution which
is fitted with a shared DC bus connection of all the
turbines [5]. Further the DC bus is connected to a
DC/DC converter (fig. 2) in order to raise the
voltage for transport. This configuration no longer
requires the use of the secondary inverter for
each wind turbine thus reducing the cost and
weight of the wind turbines.
Page |2
II.
INVERTER DESIGN
This paper only refers to a single inverter leg;
therefore 16 IGBT units are connected into a
serial connection as depicted in figure 1.
Due to the system complexity the high power
section and the command section are presented
separately.
Figure 3. A principle schematic of the system
The high power section consists of the
following:
•
•
•
•
16 IGBT units;
4 contactors;
charging resistor;
DC power supply.
A basic representation of the high voltage
application is represented in figure 3.
The command section consists of:
•
FPGA
•
•
•
•
•
ADCs
Comparators
Gate drivers
CPLDs
Line drivers
The components that form the low power section
communicate directly or indirectly with the FPGA
through analog lines (in case of sensor
measurements) or by digital lines (in case of
command, data and logic signals). The
communication between the components is
mostly serial. For the gate drivers the
communication is parallel through differential
lines. An overview of the command part is
presented in figure 4.
The FPGA is the main processing unit of the
system. It communicates with the ADCs,
comparators, gate drivers and also provides a
user interface to ease monitoring and data input.
The FPGA interface with the components uses 2
different protocols: parallel and serial. The
parallel communication is only done when
connected to a DSP. The serial communication is
used for the rest of the units allowing speeds up
to 32 Mbps. The FPGA does not have any ADC
module inside therefore it is necessary to place an
external ADC in the application. The ADC unit uses
several operational amplifiers to apply a gain and
an offset to the input signal so the resulted signal
could be handled by the ADC itself and by a set of
Page |3
comparators.
The comparators are used to check the signal
for upper and lower limitations set according to
the setup configuration. These limitations are the
over-voltage, under-voltage, over-current, undercurrent. The input voltage range of the ADC is
between -10 to 10 V, or the current range
between -20 mA to 20 mA.
The gate output unit handles all the
communications between the FPGA and the
actual gate drivers. The serial communication
with the FPGA is handled by a CPLD (Xilinx
XC9572) that operates at 5 VDC.
An IGBT unit has, beside the high power
components, the gate drivers and the interface
boards which form the control section of the unit.
The user interface of the system is represented
by a VGA based screen, a standard computer
keyboard and mouse on a PS/2 connector. The
user input, mainly the keyboard and the mouse,
are handled by an external microcontroller which
implements the PS2 protocol in order to
communicate with the user input devices. The
communication with the FPGA is done using serial
communication. The user output device, the VGA
screen, is handled directly by the FPGA.
The DSP which only connects with the FPGA
has Parallel connection with a data bus of 32 lines
and an address space of 1 MByte. The rest of the
application modules, due to their relative large
number are designed with serial communication
on 4 wires (Clock, Chip Select, Data In, Data Out)
plus several other lines for application specific
signals (Fault, Reset, Output enable, etc).
In order to avoid having every unit running at a
different clock frequency, the system general
frequency was established at 31.75 Mhz, (1/4 of
the FPGA clock frequency of 125 MHz). The
frequency of 31.75 MHz has a clock period of 32
ns which represents the amount of time required
for a bit to be sent.
Most of the communication between
components is done on single ended lines, except
for the gate drivers. The communication between
Page |4
Gates CPLD line drivers and the gate drivers
interface is done over differential lines. The lines
are twisted pairs shielded cables. This reduces the
influence of external noise over the signal.
III.
∑ ! ∑
CONTROL STRATEGY
The sixteen units inverter can produce 9
voltage levels. Only 8 units are on at any time
therefore only 8 capacitors are active at all time.
A mandatory constraint that assures constant
voltage across all units is given by the following
expression:
8
(1)
Where:
m – are the units switched on (open) in the upper
part of the leg;
n – are the units switched off (close) in the upper
part of the leg;
The close and open states are represented in
figure 5, where the orange line represents the
current flow. These figures are just two examples
for the positive current flow.
a)
b)
Figure 5. a) Example for close state of a unit; b) Example for
open state of a unit
Assuming that the voltage on one unit does
not vary and the DC bus is fed with 600 VDC ( )
the amount of voltage on one unit is:
V
DC
V VDC
75 VDC
The main power supply (VDC ) feeding the
inverter can be up 4,8KV/150Amps.
The total number of switching combinations of
the inverter leg units results from the expression:
(2)
!
! !#$%!
255
(3)
All these states are valid and thus used to
attain optimized capacitor voltage balancing. This
is required because the capacitors tend to
discharge due to their internal resistance and the
resistance placed in parallel with each of them
(fig. 6).
In order to achieve the multilevel voltage at
the output, the control structure is based on a
combination of staircase algorithm and square
wave. The staircase algorithm is necessary in
order to obtain the multi-level waveform at the
output. The method is used to determine which
voltage level should be outputted next. The
various unit combinations are made also with
respect to the optimization of voltage balancing.
The square wave modulation is present due to the
voltage balancing. After the voltage level was
determined by the staircase algorithm, the
voltage balancing uses the square wave
modulation to control each of the IGBT units.
The voltage balancing is based on a sorter
algorithm that is applied at each step. A step does
not have a constant time period, it varies with
respect to computation time and with the desired
output frequency. The step time only has a
minimum duration of 750 ns, time required by the
gate drivers to switch from a state to another.
The staircase modulation algorithm calculates
the length of each stair as a function of the DC bus
voltage, number of levels and a threshold value,
which adjusts the algorithm with respect to the
reference sinusoidal waveform. This threshold
value is given by the expression:
'()*++ +,*- . /
Where:
'()*++ - voltage threshold;
(4)
Page |5
- voltage step;
- coefficient that belongs to the interval
/ 0 #0; 1%;
This coefficient increases the effectiveness of
the control structure of the staircase modulation.
The 0 and 1 values of the coefficient were
excluded from the interval because in this case
the 0 respectively 4300 are no longer
obtainable.
The threshold value can have one value for the
rising slope and another one for the falling slope,
or even one for each quadrant. For this
application, two values were used, one for
quadrant I and III and another one for quadrant II
and IV. The two values differ because of the
capacitor effect over the voltage: the voltage on
the capacitor cannot vary rapidly.
The threshold for the rising edge is called
Vlo_threshold and for the falling edge if Vhi_threshold. The
formulas for these two thresholds are:
+,*/
9·;<=
67'()*+(768 >
?@A@?B
. /67
Figure 6. Staircase wave form of the control structure
IV.
EXPERIMENTAL RESULTS
Each unit contains four IGBTs placed into an Hbridge configuration and 6 capacitors. Unit
components are presented in figure 7.
(5)
Figure 7. IGBT high power unit
('()*+(768 9·;<=
>?@A@?B
. /(
(6)
Where:
/67 , /( - Coefficient between [0; 1) for the lo and
hi thresholds;
VDC
- DC Bus voltage;
nlevels
- The number of levels of the inverter
('()*+(768 , 67'()*+(768 – The voltage threshold
limit for a step;
A graphical representation of the staircase
waveform of the control structure is depicted in
figure 6.
One unit consists of two types if IGBTs: one
Semikron with a rated voltage of 1200 V at 900
Amps (named IGBT subunit A in the report) and
one Toshiba IGBTs which is rated 1200 V at 150
Amps (named IGBT subunit B in the report). The
Semikron IGBT has 6 IGBTs inside, two by two
connected in parallel in order to obtain the 900
Amps capability. The two IGBT units are different
in size of the large unit not because of special
purposes but due to the lack of components. Due
to the extent of the setup it is presented in three
distinct figures (fig 8, 9, 10). An FPGA board was
used for its capability of achieving parallel
computation and a relative high processing speed.
In figure 8 the main diodes interface the main
setup with the main power supply. The charging
resistors are placed close to the diodes in order to
reduce the losses.
Page |6
The cables that link the FPGA extension board
(placed on top the FPGA board) are all shielded
and have a small ferrite core on each pair in order
to reduces the noise. The cable shield for the
command section is connected to ground only at
the end towards the FPGA.
Figure 8. Setup power units
The resistors are 20 Ohms at 8 Amps. The two
contactors near the diodes and charging resistor
represents the main contactor, used to connect
the power supply to the system, and the charging
contactor. On the right side of the charging
resistors are the load contactors one for the load
transformer and the other for the load resistor.
The voltage sensor is placed on the DC bus
between module A and B of each unit.
Figure 9. Setup control boards
Figure 9 presents the control boards. The
control boards are placed on an aluminum plate
in order to ground it and every cable shield. The
gates line driver boards are also grounded as they
are connected to the high power section. Even if
there is isolation to the gate driver, voltage can be
induced in the wires and a certain degree of
protection must be assured. The FPGA board is
placed in the middle of the motherboard in order
to maintain a reduced distance for the cables.
Figure 10. Main setup desk
Figure 10 shows the whole setup desk where
all the control boards are located. The power
supplies used to power the system and the
control boards are found on top of the desk. The
large number of power supplies is mandatory
because these cannot handle large currents. The
main power supplies that are used to power the
main system are 300 V at 2 Amps each while the
power supplies for the control boards are rated 3
Amps. The amount of power required for the
control boards at 15 VDC measured 7 Amps when
no switching occurred and 11 Amps when the
system is fully operational. The other voltages 24
V, 5 V and -15 V does not require more than 1,2
Amps each. The boost capacitors (90 mF) are used
at power-up, when the power supply from the
mainboard is switched on. The current absorbed
when the control part at powered up is around 16
Amps (average measured).
The function generator from the picture has
been used to test the CPLD board at high speeds.
Page |7
The data was then monitored with the
oscilloscope and the logic analyzer.
The VGA screen and the Keyboard are
connected to the FPGA and are part of the user
interface.
The final results are shown in figure 11. The
voltage output at a frequency of 320 Hz and with
the VDC at 80 V. The output voltage of the
bidirectional modularized multilevel inverter arm
is a staircase waveform with small amount of
noise due to inductances, switching and voltage
level on the capacitors. The obtained waveforms
are similar with those obtain through simulations
therefore the experimental work confirms the
simulations.
connected. This can be notices again from the
difference between channel 1 and 3. The
difference between the two outputs represents
the voltage on the capacitor. The pulses of the
module B have a variable density with respect to
the output waveform. When the output voltage is
increasing from – VDC to + VDC, the density of the
pulses also increases until a constant 0 level is
maintain when the output is at the maximum
value.
Figure 12. Unit pulses (zoom out)
V.
Figure 11. Output voltage waveform at 320 Hz
In figure 12 a unit from the lower section of
the inverter leg is monitored. The channel 1 and 3
are connected between the output of module B
respectively A, and the negative line of the unit
internal DC Bus. The channel 2 represents the
main output of the inverter leg. The pulses from
channel 1 coincide with the output wave form. In
the period when the output voltage is low, the
channel 1 shoes that the module is behaving like a
wire. The voltage difference between the channel
1 and 3 is 0 therefore the capacitor is not
connected in the system. In the case when the
output voltage is high, the unit capacitor is
CONCLUSIONS
The aim of this paper was to present the
design and the control of a bidirectional
modularized multilevel inverter arm. This type of
inverter can produce nine voltage levels. The
control strategy that involved staircase
modulation with voltage balancing gave the
expected results and confirming the simulations.
The experimental work was conducted at
various input voltages ranging from 80 V up to
600 V (like in simulations) and output frequencies
between 2 to 600 Hz. In each case the system had
showed good responses
Page |8
REFERENCES
[1] Tomasz Biskup, "Multilevel converter for
power systems with SMES".
[2] A. A. Sneineh, "A new topology of capacitorclamp cascade multilevel converters" 2006.
[3] Lena Max, "System efficiency of a DC/DC
converter based," in Nordic Wind Power
Conference, ESPOO, Finland, May 22-23,
2006.
[4] L. Max, "An investigation of different DC/DCconverters for wind farm applications," in
Technical Report, 2005.
[5] http://www.elkraft.ntnu.no/norpie/10956873
/Final%20Papers/009%20%20Eval_WF_lay_00
9.pdf
Autonomous mobile robots avoiding obstacles using
ultrasonic sensor or video camera
Dan Novischi
Department of Electrical Engineering, Polytechnic University of Bucharest
Splaiul Independentei Nr. 313, Bucharest, Romania
dan.novischi@gmail.com
Abstract— This paper presents two mobile robots that have the
ability to avoid obstacles when going from a start point to an end
point. The robots were built using the NXT Lego Mindstorms kit:
one uses an ultrasonic standard sensor (US) and the other a
custom vision subsystem MindSensors NXT camera. The
software was written in two languages: NXC (NXC/NBC1.7
firmware) and Java (Lejos0.7 firmware). The main algorithm,
implemented both with ultrasonic sensor and camera, computes
the minimum trajectory for avoiding the obstacle. Experiments
were performed to evaluate the robots performance.
Key words— mobile robots, ultrasonic sensor, video camera,
Bluetooth communication, obstacle avoidance, detection, travel,
processing, experiments
I. INTRODUCTION
Mobile autonomous robots are increasingly becoming more
important in people lives. Today, these robots are used in
various applications like household cleaning, self driving
vehicles, intelligent security systems and space exploration,
for the expected benefits of their improved efficiency. In
these applications robots have to deal with unstructured
environments that contain lots of uncertainties, that is,
environments for which there is no prior knowledge of the
landscape and the locations or shapes of the obstacles. The
ability to avoid obstacles gives robots the necessary
intelligence to make optimal decisions while travelling in such
environments.
This paper focuses on obstacle avoidance in an indoor
environment. Simply stated, for a given starting point, the goal
is to make the robot reach a final destination point, where the
environment has a flat landscape and still objects are
randomly lying around. The main goal of this work is not only
to develop obstacle avoidance robots, but also to do this with a
low cost hardware. We propose two hardware and software
approaches to address this problem. The first approach is
based on ultrasonic sensors and the second is based on a
vision subsystem (camera). For physically implementing the
robotic systems we chose the Lego Mindstorms NXT kit and a
custom NXT camera from MindSensors. The software is
written in two languages: NXC and Java. The aplication for
the ultrasonic based system is implemented in NXC and the
aplication for the video camera based system is implemented
in Java. Experiments were conducted to determine the robots
performances. Results show that the camera based robot has
significantly higher performances.
Following this short introduction section 2 reviews the
previous work, section 3 describes the hardware and the
implementation of robotic systems. Section 4 describes the
two languages used for implementation and section 5
summarizes the Bluetooth communication between the NXT
microcontroller and a PC. Section 6 describes the obstacle
avoidance alorithm for both robots, section 7 presents the
experiments and section 8 draws the conclusions.
II. PREVIOUS WORK
The available literature on obstacle avoidance for
autonomous mobile robots is immense. One reason for this is
that many approaches are necessarily specific to the robot in
question. Early work on obstacle avoiding algorithms both for
ultrasonic sensor and camera based robotic systems was done
in the late 80’s and was focused on different strategies to
determine the size and location of the obstacles. This work
includes the research of Borenstein, Koren [12] and Khatib
[13].
Till today most important work was done as a part of
EUREKA Prometheus Project [10] by Ernst Dickmanns. This
project was the largest R&D European project ever in the field
of self driving cars at very high speeds. The project
achievement was the basis for most subsequent work on self
driving cars.
Today, a large variety of competitions and platforms
address the problem of obstacle avoidance for different
proposes. Most significant competitions and platforms are:
 DARPA Grand Challenge [9] – is a competition
between self driving cars sponsored by the Defence
Advanced Research Projects Agency (DARPA) and
provides some good, and bad, examples of the
current state of the art in obstacle avoidance.
 iRobot [11] – is a robotic platform developed by the
company with the same name that is used to
implement a series of household cleaning, indoor and
outdoor, autonomous robots.
III. HARDWARE AND ROBOTIC SYSTEMS
The robotic systems are physically implemented using the
Lego NXT Mindstorms kit [1] and a custom, NXT Camera [2],
vision subsystem. The NXT Mindstorms kit is a robotic
platform composed of the following parts: a microcontroller,
three DC motors and four sensors, including the ultrasonic
sensor.
The NXT microcontroller has a 32-bit ARM7
programmable CPU running at 48 MHz, Bluetooth and USB
hardware modules, an 100 x 64 pixel LCD graphical display,
four analogue/digital input ports and three digital output motor
ports. Practically, the NXT is an intelligent, computercontrolled Lego brick that lets a Mindstorms robot “come
alive” and perform different tasks [1].
The three DC motors are 9V electric motors which can
achieve a maximum rotation speed of 200 rpm. Each motor
has a built-in rotation encoder. The rotation encoder measures
motor rotations in degrees or full rotations with an accuracy of
+/- 1 degree [1].
The ultrasonic sensor (US) contained in the Mindstorms
kit is an 8-bit digital sensor which measures distance in
centimetres and in inches. It is able to measure distances from
0 to 255 centimetres with a precision of +/- 3 cm [1].
Experiments done with this sensor revealed that:
 distances smaller than 3 cm can not be measured.
 the sensor should always be placed in a horizontal
position because other positions decrease maximum
measured distance.
 the measured distance is strongly influenced by
mechanical vibrations.
 in some distance intervals the sensor has a high
probability to return 255 cm instead of the actual distance.
The NXT Camera from MindSensors is a custom vision
subsystem for real time image processing. The camera can
track up 8 distinct objects, at 30 frame/second [2], based on
the object R.G.B. map ( R – red, G – green, B – blue ). This
vision subsystem offers post processed images which are used
in programming. The post processed image, at one time, is an
interpretation of the real image that the camera sees. The
object being track is represented in form of blobs. A blob is,
here, represented by rectangles shapes with corresponding
RGB map to the tracked object.
A
B
C
Fig. 1 Camera images: real image (A), processed image for obstacle
identification (B) and post processed image - simplified model for the
obstacle, as used in programming (C)
Experiments done with this camera revealed that:
 a tracked object is interpreted at one time by two or more
blobs.

the change rate of the post processed image is very high
 the camera is more sensitive to the red colour then any
other colour.
 tracked object recognition is strongly influenced by
ambient light.
With this hardware two robots were built: one with two
ultrasonic sensors and the other with NXT Camera. Both
robots use a differential locomotion system for propulsion,
implemented using two DC 9V motors. The difference
between the two is that ultrasonic sensors are placed on a
mobile arm. This arm is built using a third DC 9V motor.
Fig. 2 Robot equipped with NXT camera (left) and US sensor (right)
IV. SOFTWARE IMPLEMENTATION
The software algorithms are implemented in two languages:
NXC using NBC/NXC 1.7 firmware and Java using leJOS 0.7
firmware. Both languages offer a wide variety of functions to
control the motors, sensors, Bluetooth module, LCD display
and NXT user interface buttons. The main difference between
the two is that NXC (similar to C) is not object oriented while
Java is an object oriented language. Java has an enhanced
programming interface for
controlling the NXT
microcontroller. The most important advantages of Java over
NXC are listed below:
 Synchronization mechanism is based on monitors and is
implemented without busy waiting
 Support for PC to NXT Bluetooth communication and
also support for third party devices to NXT Bluetooth
communication
 Support for float primitive data type
V. BLUETOOTH COMMUNICATION
In some situations, it is desirable to remotely control the
robot and to be able to monitor and analyse different
characteristics (data) of a certain task that the robot is
performing without interfering with that task. In this project
we want to remotely send the current robot location and the
destination coordinates and we, also, want to be able to
analyze characteristics like robot path from a start point to an
end point to determine the robot’s performances. Bluetooth
module is the most suited for this propose. Communication
and interaction with the mobile robot is implemented via
Bluetooth protocol. Bluetooth communication takes place
between a PC, which plays the master role and NXT
microcontroller, which plays the slave role. Both master and
slave performe similar tasks: initiate bluetooth ports and
transmit or receive data. The difference between the two is
that master initiates the Bluetooth connection and the slave
waits for master request and then responds to it accordingly.
VI. OBSTACLE AVOIDANCE ALGORITHM
Obstacle avoiding algorithm is structured in three
individual tasks: travel, detection and processing. The main
program logic, that reunites these tasks, starts from the idea of
robot movement on three segments: (AB) from the start point
to an obstacle, (BC) from the obstacle to a determined
intermediate point and (CD) from the intermediate point to the
final point. These segments are illustrated below:
Fig. 3 Robot travel path segmentation
In the algorithm for the main program the travel task is a logic
function (method) that returns true if the robot has stopped
after a detection of an obstacle or returns false if the robot
arrived successfully to the final point. “foundObstacle” and
“foundInterObstacle” are two flags signalling that there is an
obstacle on the segment from the initial to final point,
respectively there is an obstacle on the segment form the
current obstacle to the intermediate point. The pseudo code for
this algorithm is presented below:
Main program algorithm
1.
2.
3.
4.
5.
6.
foundInterObstacle = false;
foundObstacle = false;
Start Detection;
foundObstacle = Travel( );
Processing( );
While (foundObstacle)
6.1. foundInterObstacle = true;
6.2. While (foundIntermediateObstacle)
6.2.1. foundInterObstacle = Travel( );
6.2.2. Processing( );
6.3. foundObstacle = Travel( );
6.4. Processing( );
Since the NXT is communicating with the PC via
Bluetooth, the slave part for the Bluetooth communication
task is an additional thread that runs in the background,
without interfering with the thread synchronization of the
obstacle avoidance algorithm.
The detection task’s goal is to identify an obstacle that is
on the robots travel path and then signal the travel task. For
this purpose we use the global flag “foundObstacle”. After
obstacle identification the detection task is temporary
suspended and it resumes activity after the processing task is
finished, according to the synchronization diagram. The
algorithm for this task is almost the same for both sensors.
The difference between the two is that distinct “obstacle”
characteristics are analysed with each of the sensors, namely:
distance to the obstacle with ultrasonic sensors and obstacle
width recognition with the NXT camera. The general
algorithm for this task is presented below:
Detection task algorithm
1.
2.
3.
foundObstacle = false;
Enable sensor;
While (true)
3.1. Read sensor registers;
3.2. if( obstacle indentified )
3.2.1. foundObstacle = true;
3.2.2. Notify travel task;
3.2.3. While ( !finishedProcessing )
3.2.3.1. Wait;
The travel task is task that drives the robot between two
points. When travelling from the starting point to the final
point, the robot could meet obstacles. In this case the robot
movement must be stopped suddenly by the travel task upon
notification of the obstacle detection. Notification for obstacle
An important aspect for the obstacle avoiding algorithm is detection is accomplished by the detection task (described
thread synchronization. Therefore main program logic, the previously) through “foundObstacle” global flag. Because the
travel task and the processing task were united in a single robot stops before the destination point is reached, travel task
thread and the detection task is performed by a separate thread. must update the current robots (x, y) coordinates and current
Synchronization between the two threads is achieved with two angle () that the robot is facing relative to x axis. The triplet
flags, named: “foundObstacle” and “finishedProcessing”. The (x, y,) is also named robot pose. The travel task algorithm is
presented below:
diagram for thread synchronization is presented below:
Travel task algorithm
1.
2.
3.
4.
5.
6.
7.
8.
9.
Fig. 4 Synchronization diagram
Get robot initial pose (xi, yi, i);
Get final coordinates;
Compute (xr, yr) relative to initial (xi, yi);
Convert (xr, yr) into angular coordinates (R, );
Compute (r) relative angle to initial (i) angle;
Rotate motors anti-synchronous with (r) angle
While( !foundObstacle && !( distance R crossing))
7.1. Synchronous move motors forward;
Synchronous brake motors;
Update current pose;
The processing task is the most important part of the
obstacle avoidance algorithm. This task determines the
relative position of the obstacle to the robot and then it
computes the shortest path to avoid it. The algorithm has 3
steps: obstacle analysis, data processing and computation of
minimum trajectory to avoid the obstacle. These steps are
accomplished differently for each of the two robots. For the
ultrasonic sensor based robot, the processing task has the
following steps: the scanning of the detected obstacle,
correction of retrieved data and the computation of the
intermediate (x, y) coordinates. The scanning process is done
by measuring distances at increasing angles starting at 0
degrees (where the ultrasonic sensors are parallel with the
robot direction of travel), up to 90 degrees. The scanning is
done similarly for both the left and right sides. Because of the
ultrasonic sensor limitations incorrect measured distances are
uniformized by the correction process. Before computing the
intermediate (x, y) coordinates, 3 additional steps are taken for
correct determination of obstacle relative position to the robot:
the aligning of sensors with the direction perpendicular to the
obstacle, the scanning of the obstacle and the data correction.
The algorithm for the processing task is presented below:
US processing task algorithm
1.
2.
3.
Scan obstacle;
Correct data retrieved ;
If( ! aligned )
3.1. Compute US alignment angle;
3.2. Align US;
3.3. Scan with US aligned;
3.4. Correct data retrieved;
4. Compute minimum trajectory
5. Determine intermediate coordinates
6. finishedProcessing = true;
7. Notify detection task;
For the NXT camera based robot the steps of the
processing task are different than those for the ultrasonic
sensor based robot. The steps are the following: the
determinations of the obstacle location in the camera window
frame (obstacle analysis), the computation of the distance for
avoiding the obstacle based on its width (data processing) and
the computation of (x, y) intermediate target coordinates
based on the distance to the obstacle and the orientation angle
(the computation of the minimum trajectory).
7.1. Compute width from right obstacle margin to left
frame margin;
7.2. Compute relative angle to avoid the obstacle;
7.3. Compute absolute angle to avoid the obstacle;
8. Else
8.1. Compute width from left obstacle margin to right frame
margin;
8.2. Compute relative angle to avoid the obstacle;
8.3. Compute absolute angle to avoid the obstacle;
9. Compute intermediate coordinates;
10. finishedProcessing = true;
11. Notify detection task;
VII.
EXPERIMENTS
Experiments were performed with the two mobile robots
in order to determine the following characteristics:
 The precision with which the two robots reach a
(final) destination point, in two cases: with and
without an obstacle on the robots travel path.
 The overall time it took each of the two robots to
reach the destination point.
 The total distance travelled, between the initial
and the final point, for each robot.
The experiments were conducted in two ways: by direct
physical measuring of the robot state and by using Bluetooth
communication with the PC to send subsequent robot states.
The robot state, at a given moment, is defined as the current
robot (x, y) coordinates and the time passed to reach the
coordinates in question. The states were saved in a text file.
This text file was processed with a Matlab application that
was developed afterwards.
The experimental results are summarized in the table and
graphics below:
Sensor
Final
Point
x, y
[mm]
Precision
Overall
Time
[%]
2.1
[s]
63
Total
distance
travelled
[mm]
1030
400, 300
0.7
5
910
US
Camera
NXT camera processing task algorithm
1.
2.
3.
4.
5.
6.
7.
Read x left coordinate of the blob;
Read x right coordinate of the blob;
Compute the obstacle width;
Compute the distance to the obstacle;
Compute width from the obstacle to left frame margin;
Compute width from the obstacle to right frame margin;
If ( widthToLeftMargin > widthToRightMargin )
Fig. 5 Travel path graphic for the robot equipped
with NXT camera
which the robots reach a final point, the overall time and the
distance travelled between the two points. This characteristics
show that the camera based robot has a significantly higher
performance.
Obstacle avoidance by autonomous mobile robots is a
problem that can be solved even with a limited and low cost
hardware like the Lego NXT Mindstorms kit. Therefore, in
the future, autonomous robots could have a greater impact in
people lives.
ACKNOWLEDGMENT
First and foremost I would most like to thank professor
Constantin Ilas for his support and encouragement during the
development of this project. Also, I would like to thank my
family for supporting me every step of the way.
Fig. 6 Travel path graphic for the robot equipped
with ultrasonic sensors
VIII.
CONCLUSIONS
This paper presents two mobile robots that have the ability
to avoid obstacles when going from an initial (starting) point
to a final (destination) point. The two robots were built using
the NXT Mindstorms kit. The first of the two robots is based
on two ultrasonic sensors and the second is based on a NXT
camera from MindSensors.
Experiments conducted both with ultrasonic sensor and
video camera revealed some of their limitations.
The software was created in two programming languages:
NXC using NXC/NBC 1.7 firmware and Java using leJOS 0.7
firmware.
Communication and interaction with the robots was
implemented via Bluetooth protocol. This communication
takes place between the PC (Master) and the NXT
microcontroller (Slave).
The starting point idea for the main program logic is to
divide robot movement on three segments between the initial
point and the final point. These segments are:
 First segment – from the initial (starting) point to the
obstacle
 Second segment – from the obstacle to an
intermediate point
 Third segment – from the intermediate point to the
final (destination) point
The algorithm for obstacle avoidance was structured in
three individual tasks, namely: the detection task, the travel
task and the processing task.
The most important part of the obstacle avoidance
algorithm is the processing task. The job of this task is to
determine the obstacle relative position to the robot and then
compute the minimum trajectory to avoid it. This task was
implemented differently for each robot.
Both robots were able to successfully avoid the obstacles
and reach the final point.
Experiments were performed with the two robots to
determine the following characteristics: the precision with
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
Lego NXT Mindstorms: http://mindstorms.lego.com/, Jun 2009
MindSensors: http://www.mindsensors.com/, Jun 2009
J. Hansen, ” NXT Power Programming: Robotics in C”, Feb 2008,
ISBN-13 978-0973864922
J. Hansen, E-book: “Not eXactly C Programmer's Guide”,
Oct 10 2007
RobotC: http://www.robotc.net/, Jun 2009
B. Bagnall, “Maximum LEGO NXT: Building Robots with Java
Brains”, July 15 2009, ISBN 0973864958
Juan Antonio Brenda Moral, E-book: “Develop leJOS programs”,
Apr 2009
NXT Cam View User Guide: http://nxtcamview.sourceforge.net/,
Jun 2009
NXT motor internals: http://www.philohome.com/, Jun 2009
DARPA Grand Challenge: http://www.darpa.mil, Jun 2009
EUREKA Prometheus Project: http://en.wikipedia.org/, Jun 2009
IRobot: http://store.irobot.com, Jun 2009
J. Borenstein and Y. Koren, “Obstacle avoidance with ultrasonic
sensors”, IEEE Journal of Robotics and Automation, vol. 4, Apr. 1988
O. Khatib, “Real-time obstacle avoidance for robot manipulator and
mobile robots”, Internat. J. Robotics Res. 5 (1) (1986) 90–98
J. Michels and A. Sexna, “High speed Obstacle Avoidance using
Monocular Vision and Reinforcement Learning”, Autonomous Systems
Laboratory, CSIRO ICT Centre
Joseph Jones, “Robot Programming: A practical guide to BehaviorBased Robotics”, McGraw-Hill, 2004
Ronald C. Arkin, “Behavior-Based Robotics”, MIT Press, 1998
Ulrich Nehmzow, “Mobile Robotics: A practical introduction, second
edition”, Springer-Verlag London, 2003
Gregory Dudek, Michael Jenkin, “Computational Principles of Mobile
Robotics”,Cambridge University Press, 2003
Radu B. Rusu, (Romanian): “Arhitecturi moderene pentru roboti
mobili” : Javaclient and ZeeRO, Master dissertation thesis, Technical
University of Cluj-Napoca, Apr 2005
Design of Inverter-Fed SPMSM-Motor Drive Line in a
FC Truck System
Reference topic: Design of power electronic converters and systems
Jorge Varela Barreras
Institute of Energy Technology
Aalborg University
Denmark
Abstract - The purpose of this paper is to design and implement an
efficient control for SPMSM, between their upper and lower speed
limits, under the requirements of a FC truck system. The simulation
of the SPMSM, truck load, VSI, modulation strategy and control
structure is performed in MATLAB/Simulink. For the simulations,
real parameters of the provided SPMSM obtained in the laboratory
are used. A Field Oriented Control strategy is implemented and
tested on several situations.
PMSM drives
PMSM are increasing applied in several areas such as traction,
automobiles, robotics and aerospace technology.
In general, PMSM machines have the efficiency advantage and can offer
high dynamic performance when are controlled by field oriented
control, so they are an attractive alternative to the IM and DC machine
for automotive applications
A requirement in PMSM control is the synchronization of the AC stator
Index Terms - SPMSM, VSI, SVM PWM, FOC, Ackerman steering voltage frequency and the rotational speed of the shaft - to achieve this a
shaft-mounted position/speed sensor can be used (Fig. 1.2).
geometry
1
INTRODUCTION
Nowadays, due to the massive use of energy in automotive applications
coupled to the limited petrol resources, one promising application deals
with the employment of electric motors instead of combustion engines
in vehicles.
Electric vehicles can be more efficient, do not produce noise and air
pollution emissions, and are less complex in a mechanical sense
compared with combustion systems. But their future in industry present
many challenges: to be desirable power electronic systems must offer
low cost, endurance to harsh environments, reliability, light weight, ease
to install, ease to maintain and low cost energy storage devices with
high energy densities, power density and long life.
In 2004, Semikron received General Motor's "Supplier of the Year"
Award [3] because of their work on the SKAI modules. These are a
series of three-phase VSI specially designed for hybrid vehicles.
In this paper, a truck model Stama Eltruck ST-L-100, produced by GMR
Maskiner A/S., is driven by two SPMSM fed from two SKAI VSI.
These SKAI modules employ DSP controllers manufactured by Texas
Instruments.
Thus, a robust control strategy should be designed and implemented in
these DSP boards fulfilling the main demands of the truck system:
reduced energy consumption and high dynamic performance.
Fig. 1.2 Self synchronization concept for PMSM [15]
Another way is to use sensorless control methods, which consist on an
on-line estimation of the rotor angular position and speed employing
measurements from the PMSM stator windings or the DC bus.
The position sensor is expensive, so it is not desirable in low cost
applications (heating, air conditioning…), but for automobile
applications the cost of the sensor is not as representative [4].
Three-phase PWM-VSI is widely employed in automotive applications.
It can achieve control on motor performance (speed, torque) controlling
the AC voltage and frequency of the stator three-phase balanced system.
Desired machine performance is reached programming accurate duty
cycles for the switches through PWM methods. Also, thanks to its
bidirectional power characteristic, DC-AC and AC-DC, it is especially
useful to develop regenerative braking systems.
2
TRUCK ANALYSIS
The graveyard truck model Stama Eltruck ST-L-100 is analysed based
on data from the producer, GMR Maskiner A/S.
PARAMETER
VALUE
Empty weight
760kg
Battery weight
200kg
Max. Cargo
1000kg
Max. Speed
15Km/h
Gearing ratio
15:1
Wheel radius
22.4 cm
Max. Traction
430 kp
Table 2.1 Main parameters of the truck.
Fig 1.1 Block diagram of the part of the system tackled in this paper.
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
Load Model
Employing the main parameters of the truck a load model is designed in
MATLAB/Simulink based on the next equations:
=
+
=
+
=
2·
·
=
=
+
·
bends. The relationship between both speeds can be found through a
geometric study (Fig. 2.5).
·
·
· |cos |
2·
>0
·
2·
·
·
·
2·
=0
To check the variations in the load torque a simple track has been
designed (Fig 2.1). It is assumed that the truck is driven straight and it
runs without skidding. The truck is tested under the two load situations
proposed by the manufacturer.
=
tan
Payload
Full (2200kg)
Normal (1000kg)
Fig 2.1 Diagram of the track.
Surface
Rolling friction
Gravel
Paved road
0,04
0,015
Acceleration
0,6
1
Table 2.2 Load specifications.
Fig. 2.18 Geometry of the movement
+
=
−
2
=
=
+
2
Then the relation between the mechanical speed of the inboard tire and
exterior is:
,
=
·
=
·
The values of W (width between middle points of wheels of the same
axle) and L (length between axles) can be obtained in the laboratory.
This equation can be used to implement an electric differential.
3
MATHEMATICAL MODEL OF A SPMSM [4, 8]
In order to design the control system a mathematical model of the
SPMSM is needed.
Voltage equations in stationary a, b, c reference frame
The voltage equation for each of the three stator phases is easy to derive
from the equivalent circuit of a star-connected SPMSM [9].
(
(
(
· )
· )
· )
=
· +
+
+
+
Fig. 2.2 Load torque obtained for case 1 (left) and case 2 (right)
These results are used to dimension the requirements of the system.
(
(
(
· )
· )
· )
Ackerman steering geometry [5, 6, 7]
=
· +
+
+
+
The Ackerman steering geometry indicates that, when a vehicle takes a
(
)
(
)
(
)
·
·
·
bend, the inboard front tire (assuming it is front wheel steering) turns
=
· +
+
+
+
around an smaller circle than the outside front tire. Because of that, to
avoid skidding in any wheel, the inboard front tire has to turn an angle Employing previous assumptions these voltage equations can be
rewritten in matrix form as:
closer than the outside front tire.
sin ( )
This geometry is widely used in automotive industry, since it makes
⎡
⎤
1
driving more comfortable and reduces as much as possible the wear of
sin
− · ⎥·
⎢
= ·
+
+ ·
·
+ ·
2
⎢
the tires.
· ⎥
−
⎣sin
⎦
The basic purpose of the Ackerman Steering geometry is to have all the
Voltage equations in rotating dq reference frame
wheels (4) rolling around a common point during a turn (Fig. 2.3).
The three phase stationary abc reference frame can be transformed into
a orthogonal two phase stationary αβ reference frame with a single
mathematical transformation known as Clarke transform (x can express
either the voltage, the current or the flux linkage):
=
1
( +2 )
=
√3
+ + =0
Fig 2.3 Four Wheels rolling around a common point.
Through
this transformation the previous voltage equations in matrix
The geometric solution comes from Langensperger in 1816, but
form
can
be represented in one simple vector. This notation is called
Rudolph Ackermann was who patented the invention by arrangement in
Space
Vector
Representation.
London in 1817 (Fig. 2.4).
But
the
equations
of the model are really simplified when the coordinate
This principle is still being applied to automobiles. However, modern
four wheels vehicles do not use exactly true Ackermann steering, since system is just positioned on the rotor. It allows defining a new equation
it does not take into account important dynamic effects, but it is absolute non dependent on the rotor rotating angle.
valid in applications for low speed cars, like the Stama Eltruck ST-L100. The propulsion system of the truck is made up two SPMSM in each
wheel of the rear axle. It means, according to the geometry of the
movement, that the speed of both motors should be different during
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
Using these reference voltages the modulator generates the proper duty
cycles to run the VSI which drives the SPMSM.
Control property
Several control properties have been developed for the SPMSM. In this
paper the Constant torque angle property is chosen, since this ensures
reduced energy consumption and high dynamic performance.
The torque angle is defined as the angle between the rotor d-axis and
the stator current space vector (Fig 4.2).
Fig. 3.2 Reference frame transformations.
To reach this simplification another transformation called Park is needed
to relate these components of the stationary αβ reference frame to the
new rotating dq reference frame:
= ∙ cos( ) + ∙ sin( )
= − ∙ sin( ) + ∙ cos ( )
Fig 4.2 Current and voltage vector, power factor angle ,
Applying both transformations (Parke and Clark) the voltage equations
stator flux linkage angle and torque angle .
based on the dq rotating reference frame are derived:
In this method the torque angle is keeped constant = 2 . This is
easily reached making the d-axis current zero:
= ·
+ · −
·
=0 ⇒ | |=
Then the relationship between torque and current is expressed as:
Then the magnetic flux linkages in the direct and quadratic directions
3
= ∙
∙
[
]
are defined:
2 2
 = + ·
5 CONTROLLERS DESIGN
 = ·
The design of the PI controllers is based on pole placement method, so
So the voltage equations can be written as:
this is just based on the knowledge of the system transfer functions.
Plant analysis - decoupling
=
+
−
Previously, in the presented voltage equations of the SPMSM, the crosscoupling between the d-axis and the q-axis can be observed. This
=
+
+
+

interaction between axis should be avoided for purposes of the FOC
These equations are used to develop the SPMSM MATLAB/Simulink strategy: the d-axis stator current and the q-axis stator current should
be controlled indepently (decoupled control). The only way of doing this
model. Also the next dynamic equation is employed:
is to decouple the stator voltage equations.
1
=
− −
−
The previous stator voltage equations in the rotating d-q reference frame
2
/2
can be written again separated into two main parts - linear and coupling
4
PRINCIPLE OF FIELD ORIENTED CONTROL
components:
FOC perform real-time control of torque variations demand, to control
the rotor mechanical speed and to regulate phase currents in order to
avoid current spikes during transient phases [10]. This control strategy
uses mathematical reference frame transformations (Clarke and Park).
Applying a decoupling algorithm (Fig. 5.1) “the nonlinear SPMSM
model is transformed to linear equations which can be controlled by
general PI controllers instead of complicated controllers”[4].
=
·
+
·
+
·
Fig. 4.1 Block diagram for the FOC of a PMSM.
The control structure is a cascade connection of two feedback loops.
The inner current loop uses a control property to transform an
electromagnetic torque reference into d-q currents reference while the
outer speed loop regulates the rotating speed. A speed reference is given
and compared with the feedback from the position sensor in the speed
Fig. 5.1 Decoupling (current controllers design).
loop, while the feedbacks for the current loop are the three-phase
Current controller design
currents from the SPMSM stator. These currents are transformed into dq The transfer functions of the PI current controllers are:
reference frame and feedback to two separate current controllers, one for
( )=
+
each axis. The dq voltages references are the outputs from the current
controllers and the inputs for the SVM.
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
( )=
+
,
To design the current controllers only one of the system transfer
functions is considered, the electrical transfer function, as it has the
dominant pole for the current loop.
The electrical transfer function of the system after decoupling is written
as:
( )
1
( )=
=
( )
+
And the closed loop transfer function for the current controllers is
calculated as:
+
·
( )· ( )
( )
( )=
=
=
( ) 1+ ( )· ( ) 1+
+
·
· +
=
· +
·
· +
Since the plant of the SPMSM is the same for the q-axis and d-axis
current loop, the controller parameters should be the same for both axis.
It can be seen that the closed loop systems are 2nd order systems. An
easy method to reduce these systems to 1st order and estimate the
controller parameters is explained in followings.
Firstly, the open loop poles of the electrical system transfer functions
and the open loop zeros of the PI controllers transfer functions are
calculated. Then these poles and the zeros are placed in the same points
in the left side of the complex plane and the next relationships are
derived:
( )=
( )·
1+ ( )·
=
( )
=
( ) 1+
· +
+
·
+
·
· +
·
· +
It can be seen that this speed closed loop transfer function is analog to
the current closed loop transfer function derived before. Then the same
approach can be employed to reduce the closed loop system to a first
order system.
Then if the pole and the zero are placed in the same point on the left side
of the complex plane, the next relationships is stated:
0.00024
=
· =
=
· 1.283
0.000187
Further, during simulations, it was observed that a design based on this
relationship does not give a good performance, since this does not take
into account the behaviour of the SPMSM when is mounted in the truck.
A new value of the parameter is found based on the next equations of
the load torque model:
·
=
=
2·
·
⇒
·
·
=
= 0.11819 [
·
·
2·
·
]
A constant mass of the vehicle of 1060 kg is assumed to calculate .
Then the next new relationships is stated:
0.00024
=
·
=
=
· 0.002
( + ′)
0.1184
− =−
⇒
=
6
MODULATION METHOD
Each SPMSM is fed from a single Three-Phase Inverter, which operates
− =−
⇒
=
from a fixed voltage DC-link of 48V. Each inverter is controlled with
Pulse Width Modulation (PWM), which allows amplitude and phase
These equations can be applied in the open loop transfer functions:
control of the output voltage at the same time.
· · +
· +
1
1
( )= ( )· ( )=
·
=
·
Basically the PWM of an inverter reach the control of the power sent to
+ ·
+ ·
a load modulating its duty cycle. In fact, the PWM is just a method to
·( · + )
1
1
=
·
=
·
generate switching sequences.
·
+ ·
Finally from this open loop transfer function a closed loop transfer The main topology of the three-phase Voltage Source Inverter (VSI)
employed in this project is shown in the next figure.
function reduced to 1st order is derived:
1
·
( )
( )=
=
=
( ) 1+ ·1
1+
· +
Therefore, the PI controller integer gain
than the proportional gain
order system.
=
=
should be
times higher
to reduce the closed loop system to a 1st
0.01382
46.115 · 10
= 299.69 ·
PI Speed controller design
The transfer function of the PI speed controller is:
( )=
+
As the current closed loop has higher bandwidth than the speed loop, the
reference current can be assumed equal to the SPMSM current for a
proper designed current controller. Then the inner current loop is not
considered for the speed controller design.
Thus, only the mechanical transfer function of the motor is considered.
( )
1
( )=
=
( )− ( )−
( )
+
If the load torque is treated as a perturbation to the system, and is set at
zero, and the dry friction is neglected, the motor transfer function is
simplified as:
( )
1
( )=
=
( )
+
Using previous equations, the closed loop transfer function of the speed
loop can be determined:
Fig. 6.1 Topology of the three-phase Voltage Source Inverter.
Digital devices offer several advantages over analog systems to
implement the modulation method:
•
Stability (no drift, offsets, temperature or aging effects)
•
Precision (noise immunity)
•
Flexibility (can be customized by changing software)
•
More compact (less number of devices)
Space vector modulation is chosen as PWM method instead of SPWM
or THIPWM because:
•
Less computation time is required
•
Modulation index 15% higher than SPWM (equal to
THIPWM)
•
Better harmonic performance in comparison with SPWM and
one-sixth magnitude THIPWM
Although a deeper knowledge of the theory is needed (conceptually
more complex and harder to implement).
7
SIMULATION
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
The proper performance of the whole modeled system for one SPMSM
is checked through the next simulation. The overall block model
designed in MATLAB/Simulink is presented in Fig. 7.5. The values of
the parameters employed in the simulation are presented in the
nomenclature table. The real parameters of the provided SPMSM have
been obtained through several laboratory tests.
The truck is tested considering a constant mass of the vehicle of 1060
kg. The vehicle is driven straight (no bends) and through a paved road
with no inclination.
Fig. 7.4 Speed reference – Speed measured.
CONCLUSSION
In this paper the design of an Inverter-Fed SPMSM-Motor drive line
under the requirements of a FC Truck system is presented.
The truck and the motor provided have been analyzed and a useful
model of the load of the truck has been designed. The needed of an
electrical differential has been clearly explained and the basis geometric
Fig. 7.1 Speed reference – Speed measured.
theory behind its design is presented.
The speed response is fast enough - 0 to 250 rad/s in less than 5 s. The The SPMSM has been carefully studied, its electrical and dynamic
performance is stable and there is a steady state error of about 5% at equations have been derived and a mathematical model have been
maximum speed due to the several friction torques (they are designed and implemented in MATLAB/Simulink. The VSI has been
disturbances of the system).
modelled, as well as the SVM PWM. The proper performance of the
models has been checked in simulations. The real parameters of the
provided SPMSM have been obtained through several laboratory tests.
The structure of the FOC method is presented and the theoretical design
and tuning of the controllers explained step by a step. The performance
of the whole system MATLAB/Simulink simulation is shown.
Fig. 7.2 Measured abc currents.
A regenerative braking system could be designed based on this
simulations – it should be noted that there is a current peak of 400 A
when the speed reference of the truck drops suddenly This maximum
current could damage the SPMSM stator windings.
The maximum possible save of energy is shown in the next figure (Fig.
7.3).
ABBREVIATIONS
CTAC
DSP
FC
FOC
LV
MTPAC
PI
PWM
SKAI
SPMSM
SPWM
SVM
THIPWM
VSI
Constant torque angle control
Digital signal processor
Fuel cell
Field-oriented control
Low voltage
Maximum torque per ampere control
Proportional-Integral control
Pulse width modulation
Semikron advanced integration
Surface magnets type permanent-magnet
synchronous machine
Sinusoidal pulse width modulation
Space vector modulation
Third harmonic injection pulse width modulation
Voltage source inverter
NOMENCLATURE
Fig. 7.3 Total energy flowing into the machine.
Finally, from next simulation it can be stated that the system is stable for
any changes of the speed setpoint reference (‘setpoint disturbances’).
Parameter
Motor parameters:
Number of poles
Stator phase resistance
Stator phase inductance
Permanent magnet flux
Viscous friction
Dry friction torque
Mass moment of inertia
Electromagnetic torque
Mechanical torque
Mechanical frequency
Mechanical angular speed
Symbol

Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
Value
Units
8
0.01382
46.115
0.0182
0.00024
0.00085
0.000187
-
[·]
[Ohm]
[µH]
[Wb]
[Nms]
[Nm]
2
[kg·m ]
[Nm]
[Nm]
[Hz]
[rad/s]
Mechanical position
Electrical parameters:
Electrical frequency
Electrical angular speed
Electrical position
D axis current
Q axis current
D axis voltage
Q axis voltage
Modulation parameters:
DC link voltage
Sampling frequency
Switching frequency
Truck parameters:
Mass of the vehicle
Mass moment of inertia
(for
= 1060 )
Gearing ratio
Wheel radius
Acceleration of the vehicle
Total load torque
Rolling friction torque
Inertia torque
Inclination torque
Drag torque
Rolling friction coefficient
Control parameters:
Proportional constant in
current controllers
Integration constant in
current controllers
Proportional constant in
speed controller
Integration constant in
speed controller
-
′

[rad]
REFERENCES
-
[Hz]
[rad/s]
[rad]
[A]
[A]
[V]
[V]
48
5000
5000
[V]
[Hz]
[Hz]
-
[kg]
0.11819
[kg·m2]
15
0.224
-
[·]
[m]
[m/s2]
[Nm]
[Nm]
[Nm]
[Nm]
[Nm]
[·]
[1] D. Grahame Holmes, Thomas A. Lipo, “Pulse Width Modulation for power
converters, Principles and practice”, IEEE Press series on Power Engineering,
Mohamed E. El-Hawary, Series Editor,USA, 2003.
[2] Mohan, Undeland, Robins, “Power Electronics – Converters, applications and
design”, J. Wiley and Sons, New York, 1989.
[3] Erik Mayer, “Development of motor controls using the Semikron advanced
integration power module”, Electrical Insulation Conference and Electrical
Manufacturing Expo, 2005. Proceedings Volume, Issue , 26-26 Oct. 2005, pp.
236 – 239.
[4] Perera, P. D. Chandana, “Sensorless Control of Permanent-Magnet
Synchronous Motor Drives”, Ph. D. Thesis in Electrical Engineering, Institute of
Energy Technology, Aalborg University, 2002.
[5] D. Foito, J. Esteves, J. Maia, “Electric differential and regenerative braking
EV”, 11th IEEE Mediterranean Conference on Control and Automation, Rhodes,
Greece, June 2003.
[6] D. Foito, J. Esteves, J. Maia, “Experimental and simulation results of an
electric vehicle”, Proceeding of Mechatronics 2002, Netherlands, June 2002.
[7] D. Foito, A. Cordeiro, M. Guerreiro, “A sensorless speed control system for
an electric vehicle without mechanical differential gear”, IEEE MELECON,
Benalmádena (Málaga), Spain, May 2006.
[8] Sen, P. C. Principles of Electric Machines and Power Electronics. John
Wiley & Sons, Inc., 2 edition, 1997.
[9] Young-Jin Lee, Young-Jin Yoon, Young-Ho Kim and Man-Hyung Lee, “A
Study on the Sensorless PMSM Control using the Superposition Theory”,
International Journal of the Korean Society of Precision Engineering Vol. 4, No.
2, March 2003.
[10] Erwan Simon, “Implementation of a Speed Field Oriented Control of 3phase PMSM Motor using TMS320F240”, TEXAS INSTRUMENTS
Application report SPRA588, 1999.
0.16
[·]
479.5
[·]
0.08
[·]
0.000162
[·]
Fig. 7.5 Whole system MATLAB/Simulink block model
FOC OF SPMSM
Filter _Iabc
[Vabc ]
[Iabc ]
Iabc
From
[Wm]
Vd
Vd
Vq
Vq
Goto 1
Wm
Valfa
From 1
Vq
Speed_Ref
Vabc
Vabc
Val,be_chop
Vq
Iabc
Demux
In 1
Out1
Vbeta
Speed _Ref
[Iabc ]
Speed ref
Low _Pass_Filter
FOC
Vd
[Wm]
SVM +INVERTER
Wm
[Wm]
From 3
From 2
Vd
Wm
Wm
alpha , beta -- d,q
trans d,q => abc
Tload_2
dw/dt
Tload_1
[Vabc]
From 5
[Wm]
Goto 3
Tload _1/Tload _2
Demux
SPMSM MODEL
[Iabc ]
TRUCK LOAD MODEL
Pin
From 4
Product
Tload_1
dw/dt
1
s
Tload_2
Integrator
Ein
Create PDF files without this message by purchasing novaPDF printer (http://www.novapdf.com)
Goto 2
Doubly Fed Induction Generator Fault Simulation
*Krisztina Leban, *Ewen Ritchie, **Alin Argeseanu, ***Ileana Torac
*Aalborg University, Pontoppidanstræde 101, Aalborg, Denmark
**Politehnica University Timisoara, Bl. V. Parvan 2, 300223, Romania
***Romanian Academy – Timisoara Branch. Bl. M. Viteazu 24, 300223, Romania
e-mail: krisztina_leban@yahoo.com.au
aer@iet.aau.dk; alin_argeseanu@yahoo.com; ileana_torac@yahoo.com
Abstract
This paper focuses on validating a simplified wind turbine system having a fault
rid through protection [1]. The machine used was a doubly fed induction generator (DFIG).
As protection against short circuit transients, the crowbar protection was employed in the
simulation.
An equivalent model was constructed. Simplifications were made so as to have
a system composed of grid, transformer, line and generator represented by elementary circuit
elements (R, L, C and voltage sources). Equivalent circuit models were simplified so that the
fault models may be used for synchronous machine parameters. The assumption that the
mechanical system cannot respond during the short time of a three phased short circuit was
made.
The simplified simulation model was compared to simulations constructed with
complex library subsystems. The SimPower Systems Simulink library was used for this
purpose.
Keywords: wind energy plant, variable speed generator, induction generator, double fed
induction generator, DFIG simplified model, Simulink model
1. Introduction
The Doubly Fed Induction Generator (DFIG) is widely used in wind energy power
generation. DFIG are variable speed generators, used more and more in wind turbine
applications due to easy controllability, high energy efficiency and improved power
quality,[2], [3], [4]. As power converters in a DFIG system only deal with the rotor power,
electronic costs are kept low, about 20-25% of the total generator power. This implies that the
converter is dimensioned to the rotor parameters [5], [6], [7], [8]. This makes the system more
economical than using a fully rated converter in a series configuration. Turbines are
commonly installed in rural areas with unbalanced power transmission grids. For an induction
machine an unbalanced grid imposes negative effects like overheating and mechanical stress
due to torque pulsations. For an unbalance of 6% for example the induction generator is
stopped from generating in the grid. By control of the rotor currents of a DFIG the effects of
unbalanced stator voltage may be compensated. The drive system operates in four quadrants.
This implies that a bidirectional flow of power is possible. The possibility of supplying and
consuming reactive power enables the generator system to act as a power factor compensator.
By the control of the back to back inverters the slip may be controlled. In the case of the
squirrel cage induction machine, for example, as the rotor cannot be driven, the slip only
depends on the stator and load inputs [8], [9]. As for synchronous machines a relatively large
torque may cause the machine to oscillate. The DFIG does not encounter any synchronization
problems. To observe the system and the flow of active and reactive energy a dynamic model
is needed [10]. The machine may be simulated as an induction machine having three phases
supply in the stator and three phases supply in the rotor. The rotor circuit is connected through
slip rings to the back to back inverter, arrangement controlled by PWM strategies. The
voltage magnitude and the power direction between the rotor and the supply may be varied by
controlling the switch impulses. Back to back converters consists of two voltage source
converters (ac-dc-ac) having a dc link capacitor connecting them. The generator side
converter takes the variable frequency voltage and converts into dc voltage. The grid side
converter has the ac voltage from the dc link as input and voltage as grid parameters as
output. The gearbox has the role of adjusting the speed between the blades and the rotor. The
transformer couples the generator to the grid adjusts the parameters of the machine voltage to
the grid voltage. The stator is connected directly to the grid. For a normal generation regime
the energy obtained by processing the wind speed as an input is fed into the network by both,
the stator and the rotor. In this paper the simulation of the DFIG included in a wind turbine
system [11], [12], [13], [14] is presented. With this an idea of the validity of the simplified
model is tested. Models elaborated in this paper are detailed. Normal duty and the fault ride
through models are described. Consequences of the simplified assumptions implemented in
the model are shown and the acceptability of results is discussed.
2.
DFIG System Description
In the following, the modelled drive system having a DFIG is described. Function of
the DFIG is analysed and the basic elements of the drive are presented. The aim is to
represent the drive using an equivalent circuit in two cases: normal operation and crowbar
active.
Figure 1: DFIG wind turbine system
In Figure 1 the basic normal duty diagram for the wind turbine is presented. From the
blade and shaft, the rotor of the generator receives the torque produce by the wind. The
energy is fed to the grid through the back to back inverter system and the three phase
transformer. The initial model of the transmission line consisted of a pi diagram containing
the parasitic elements of the conduction cable. According to [14] lines shorter than 100 km
are considered short lines and may be modelled as an inductance. The grid was modelled as a
three-phase voltage source.
The grid side converter is connected to the transmission line through a three phased
step down transformer. The grid is modelled like a 3 phase voltage source.
In Figure 2 the equivalent circuit of the system is presented. The grid was represented
as an alternative voltage source. For the transmission line, a pi equivalent circuit was used and
for both the transformer and the machine, a T equivalent circuit.
Figure 2: Normal Duty Equivalent Circuit
In order to simplify the equivalent circuit and consequently the mathematical
model derived from it, a set of simplifying assumption were considered. Arguments are
brought for each assumption to show their validity.
As the line is considered to be short (< 100km [15]), the model may be
simplified to a single parasitic inductance. The magnetisation branch of the transformer was
neglected ([16]). This is done under the assumption that the current is too small in that branch
and the reactance is small compared to the horizontal branch reactance.
3. Crowbar Activated
Another protection for a short circuit is called the crowbar. This fault handler is a
set of resistors used to short circuit the rotor windings in case of a severe fault. In literature it
is also called ‘beak resistor’ because it has an electrical breaking effect on the accelerating
rotor. The role of the circuit is to contribute to system stability during transients. The extra
resistance introduced in the circuit dissipates the surplus energy generated during the fault in
extremely high current conditions. The rotor of the generator is disconnected from the back to
back inverter system and short-circuited with resistors (see Figure 3). Automatically
disconnect after the fault had passed. K is a symbolical switch representing the connection
apparatus and control.
Figure 3: Crowbar Activated Diagram
The equivalent circuit of the crowbar activated diagram is shown in Figure 4. As
can be seen, the magnetisation inductance of the transformer was neglected in the short circuit
simulation.
Figure 4: Crowbar Activated Equivalent Circuit
Equations derived from Figure 4 are presented in the following. In the first circuit loop, the
equation is written as:
dL ech
dL
ig + m im
dt
dt
The equivalent elements are written as sums of series components:
R ech = R sec + R prim
ug = Rech ig +
(1)
(2)
(3)
L ech = L sec + L prim + L line
For the second circuit loop, as the rotor is disconnected from the voltage source,
only the energy stored in the motor magnetising branch intervenes. This is the main reason
why the magnetising element must not be neglected in this case.
dψr
Rr
(4)
0 = Ra ir +
dt
+ (Rcrowbar +
S
)ir
If the slip is considered to be constant, the resistor value of the rotor would
remain constant.
0 = Ra ir +
dψr
+ (Rcrowbar + Rr ' )ir
dt
ψr = Lr ir + Lm is
(5)
(6)
The circuits presented above were used as a base for fault simulation models. Each circuit
was simulated and analysed separately as shown if the next chapter. the separation has been
made in order to facilitate the basic understanding of the model and to shorten computational
time.
4. DFIG System Simulation
Figure 5 shows the main simulation level of the crowbar fault response. The protection is
inserted at 0.25s via ideal switches driven by a step signal. The fault is introduced from 0.2 to
0.3 seconds. Parameters used: Rs = 0.0021; Ls = 0.11; Lr = 0.07; Rr = 0.0021; Lm = 2.5;
Rline=0.05; Lline = 0.007;Rcrowbar = 0.1; Lprim=0.007;Rprim=0.05; Lsec = 0.007; Rsec =
0.05;
Figure 5: Crowbar Protection – Main Simulation Level
(a) Library Model
(b) Detailed Model
Figure 6: Crowbar Resistor Simulation Results - Stator Current vs. Simulation Time
(a) Library Model
(b) Detailed Model
Figure 7: Crowbar Resistor Simulation Results - Stator Voltage vs. Simulation Time
5.
Conclusion
As it may be observed from the simulation results, an over simplified model should not
be used to analyse transients. The errors introduced in the results make the simplified model
inacceptable.
Besides the structure of the circuit its self, parameters and initial conditions must be
accurately introduced in the simulation. The parameters used were not of a real system and
the initial conditions were generated by the 0.2 s simulation prior to the fault.
By comparing a complex (Sympower Systems) model to the simplified model, the last
one has been proven to be unreliable due to the reduced number of elements.
Acknowledgements
The authors would like to thank the proposer of the project Jorje Martinez of Vestas for
this support of the current work.
References
[1] Doubly Fed Induction Generator Fault Simulation Krisztina Leban Final masters Project at Institute
on Energy Technology Aalborg Denmark 2009
[2] A novel doubly-fed induction wind generator control scheme for reactive power control and torque
pulsation compensation under unbalanced grid voltage conditions Brekken, T. Mohan, N. ;Power
Electronics Specialist Conference, 2003. PESC '03. 2003 IEEE;ISBN: 0-7803-7754-0
[3] Adjustable speed generators for wind turbines based on doubly-fed induction machines and 4quadrant IGBT converters linked to the rotor; Muller, S. Deicke, M. De Doncker, R.W.; Industry
Applications Conference, ISBN: 0-7803-6401-5 DOI: 10.1109/IAS.2000.883138
[4] P. W. Carlin, A. X. Laxson, and E. B. Muljadi, “The History and State of the Art of Variable-Speed
Wind Turbine Technology,” National Renewable Energy Lab., Tech. Rep. NREL/TP-500-28 607,
Feb. 2001.
[5] Modelling of the Wind Turbine With a Doubly Fed Induction Generator for Grid Integration studies;
Yazhou Lei, Alan Mullane, Gordon Lightbody, and Robert Yacamini IEEE TRANSACTIONS ON
ENERGY CONVERSION, VOL. 21, NO. 1, MARCH 2006
[6] Vladislav Akhmatov, "Variable-Speed Wind Turbines with Doubly-Fed Induction Generators,
Modelling in Dynamic Simulation Tools," Wind Engineering Volume 26, No. 2, 2002
[7] M. Machmoum, R. L. Doeuff, and F. M. Sargos, “Steay state analysis of a doubly fed
asynchronous machine supplied by a current controlled cycloconverter in the rotor,” Proc. Inst.
Elect. Eng. B, vol. 139, no. 2, pp. 114–122, 1992.
[8] P. G. Holmes and N. A. Elsonbaty, “Cycloconverter excited divided winding doubly fed machine as
a wind power converter,” Proc. Inst. Elect. Eng. B, vol. 131, no. 2, pp. 61–69,1984.D.P. Sen Gupta
and J.W. Lynn ’ Electrical machine dynamics’, Macmillan1980, ISBN:0333138848
[9] Transformers and Electrical Machines; Ion Boldea; Editura Politehnica Timisoara 2002 ISBN 9739389-97-X
[10] Dynamic modelling of a wind turbine with doubly fed induction generator Slootweg, J.G. Polinder,
H. Kling, W.L. ; Power Engineering Society, 2001. IEEE 2001 ISBN: 0-7803-7173-9
[11] Doubly fed induction generator using back-to-back PWM convertersand its application to variablespeed wind-energy generation Pena, R. Clare, J.C. Asher, G.M.; Electric Power Applications, IEE
Proceedings -May 1996; ISSN: 1350-2352;
[12] S. Doradla, S. Chakrovorty, and K. Hole, “A new slip power recovery scheme with improved
supply power factor,” IEEE Trans. Power Electron., vol. PE-3, no. 2, pp. 200–207, Apr. 1988.
[13] Y. Tang and L. Xu, “A flexible active and reactive power control strategy for a variable speed
constant frequency generating system,” IEEE Trans. Power Electron., Jul. 1995.
[14] Ion Boldea, ” Variable Speed Generators “CRC Press ISBN : 0849357152
[15] Power System Analysis John Grainger, William Stevenson 1994 McGraw Hill ISBN 0-07-061293-5
[16] The Self and Mutual Inductances of Linear Conductors, By Edward B. Rosa, Bulletin of the
Bureau of Standards, Vol.4, No.2, 1908,
Magnetic Field Enhancement using Ferrofluid and Iron
Powder
*Krisztina Leban, *Ewen Ritchie, **Alin Argeseanu
*Aalborg University, Pontoppidanstræde 101, Aalborg, Denmark
**Politehnica University Timisoara, Bl. V. Parvan 2, 300223, Romania
e-mail: krisztina_leban@yahoo.com.au
aer@iet.auu.dk
alin_argeseanu@yahoo.com
Abstract This paper focuses on studying the variation the magnetic field of a PM
when covered with ferrofluid and iron powder. For testing purposes, four neodymium
magnets were used. One was covered with ferrofluid; one was left ‘clean’; one
covered with magnetic powder and one with ferrofluid and iron powder combined.
The magnetic field growth was quantified through measurements and the results
presented herewith.
Keywords: Ferrofluid, iron powder, magnetic field
1. Introduction
Ferrofluids are colloidal mixtures composed of ferromagnetic, or ferrimagnetic
nanoparticles suspended in a carrier fluid[1], [2]. Typically a ferrofluid is about 5%
magnetic solids, 10% surfactant, and 85% carrier, by volume. [3], [4], [5], [6].The
carrier fluid may be an organic solvent or water. The nano-particles are usually
coated with a surfactant to prevent agglomeration. This is important because the
properties of the fluid otherwise decrease.
The ferrofluid is a non Newtonian fluid as it changes its properties (most importantly
the viscosity) when exposed to a magnetic field.The permeability of the ferrofluid is
higher that that of air because of the particles suspended in the carrier fluid.
Ferrofluids do not display remnant magnetization after removing the external field.
In fact, ferrofluids display paramagnetism, and are often referred to as being
"super-paramagnetic" due to their large magnetic susceptibility [7], [8]0
To evaluate a fluid, one must take account of the magnetic field strength, viscosity,
density, thermal, electrical, acoustic and optical properties [9].
All properties of ferrofluids are derived from the properties of the components, the
base fluid, the surfactant, and the particles. The important parameter relationships
are different for each fluid depending on the composition [10].
2. Experimental Arrangement
1.1 Materials Used
The measured magnets are shown in Figure 1: a- PM-Sintex NdFeB covered
with ferrofluid(X-002-015 from www.neotexx.com); b- NdFeB ‘clean’ magnet; c-
NdFeB covered with IP CM 40 to saturation (would not attract any more
powder); d- magnet covered with ferrofluid on the top and one side and IP on
the other 3 sides.
a
b
c
d
Figure 1: Magnets used showing the various coatings a: ferrofluid covered magnet; b: ‘clean’
magnet; c: iron particle covered magnet; d: iron particles and ferrofluid covered magnet
The setup used to measure the field around the permanent magnet at different
distances from the surface of the magnet is shown in Figure 2. The magnetic
induction of each magnet was measured ad various distances between the probe
and the magnet. For each case a curve was drawn(see Figure 3). The value of the
field depends strongly on the spatial coordinated. The measurements were made
by positioning the probe in the middle of the large surface of the magnet.The gauss
meter was calibrated several times during the measurements.
Figure 2: Measurement Setup: 1- magnet to be measured; 2- gauss meter probe ;
3-gauss meter; 4- probe support; 5- mobile arm; 6- plastic ruler; 7- table;
1.2 Results
In the following, the results from the field measurements are presented in Figure
3.It may be observed that the ‘clean’ magnet exhibits the lowest field values. The
best results are obtained for the permanent magnet covered with iron particles. At
small distances measurement errors were introduced as the probe is emerged in
the attracted iron particles and the position of the probe could not be precisely
observed. The read value of the field depends strongly on the relative position
between the probe and magnet.
Figure 3:Distance as function of magnetic induction – all magnets
The ferrofluid covered magnet was left undisturbed for 6 days. As can be seen in
Figure 1, no spikes are present. This sample was not disturbed for 6 days. This is
believed to be caused by the fact that the iron particles were drawn to the surface
of the magnet where the magnetic field is strongest. Measurements were carried
out on both spiky and nonspyky samples. Results are shown in Error! Reference
source not found.Figure 4. The ‘fresh’ ferrofluid gives better results than the
undisturbed sample. Practically, the iron particles are drawn to the surface of the
magnet and the magnetic permeability becomes that of the carrier fluid which is
closer to that of the air.
3. Conclusions
Due to the permeability of the ferrofluid, the magnetic field is ‘preserved’ for a
longer distance from the source.
It is of great importance to have a ferrofluid with iron particles in suspension. If the
particles settle (for whatever reason), the properties of the ferrofluid decrees. By
disturbing the settled ferrofluid, the quality of response may be regained.
Figure 4: Spiky and nonspiky PM+FF
References
[1] Timko, M. Zentko, A. Zentkova, M. Koneracka, M. Kellnerova, V. Zentkova, A. Stepan,
M. Barbora, J. Inst. of Exp. Phys., Slovak Acad. of Sci., Kosice ; Magnetorheological
properties of some ferrofluids Magnetics, IEEE Transactions on; Mar 1994 Volume: 30,
Issue: 2, Part 1-2; page(s): 1117-1119;ISSN: 0018-9464
[2] Ferrofluids and Magnetorheological Fluids Ladislau Vékás Advances in Science and
Technology Vol. 54 (2008) pp 127-136 online at http://www.scientific.net © (2008) Trans Tech
Publications, Switzerland Online available since 2008/Sep/02
T. Albrecht, C. Bührer et al. (1997), "First observation of ferromagnetism and ferromagnetic
domains in a liquid metal (abstract)", Applied Physics A Materials Science & Processing
(Applied Physics A: Materials Science & Processing) 65: 215, doi:10.1007/s003390050569
[3] K. Raj, B. Moskowitz, and R. Casciari, “Advances in ferrofluid technology”J. Magn. Magn.
Mater., vol. 149, pp. 174–180, 1995.
[4] V. Bashtovoy, B. Berkovsky, and A. Bislovich, Introduction to Thermomechanics of Manetic
Fluids. Bristol, PA: Hemisphere, 1988.P. Cambell, Permanent Magnet Materials and Their
Applications. Cambridge, MA: Cambridge Univ. Press, 1994, pp. 57–61.
[5] Ralph Jansen Ohio Aerospace Institute Brook Park, Ohio; EIiseo DiRusso Lewis Research
Center Cleveland, Ohio Passive Magnetic Beating With Ferrofluid Stabilization NASA
Technical Memorandum 107154; February 1996 National Aeronautics and Space
Administration
[6] Fernando D. Goncalves, Jeong-Hoi Koo and Mehdi Ahmadian ”A Review of the State of the
Art in Magnetorheological Fluid Technologies - Part I: MR fluid and MR fluid models” The
Shock and Vibration Digest. Vol. 38. No.3. May 2006 203-219, rD2006 SAGE Publications,
DOl: 10.1177/0583102406065099
[7] Rheology Fundamentals Alexander Ya. Malkin, ChemTec publishing 1994 ISBN 1-895198-097
[8] Rheology of Field polymer systems Aroon V. Shenoy Kluwer Academic Publishers 1999 ISBN
0-412-83100-7
[9] J.-C. Bacri and R. Perzynski, V. Cabuil Colloidal Stability and Transport Properties of
Ferrofluids Brazilian Journal of Physics, vol. 25, no. 2, June, 1995
[10] Ralph Jansen Ohio Aerospace Institute Brook Park, Ohio; EIiseo DiRusso Lewis Research
Center Cleveland, Ohio Passive Magnetic Beating With Ferrofluid Stabilization NASA
Technical Memorandum 107154; February 1996 National Aeronautics and Space
Administration
Novel Actuator I
Magneto rheological (MR) Fluid
*Krisztina Leban, *Ewen Ritchie, **Alin Argeseanu
*Aalborg University, Pontoppidanstræde 101, Aalborg, Denmark
**Politehnica University Timisoara, Bl. V. Parvan 2, 300223, Romania
e-mail: krisztina_leban@yahoo.com.au
aer@iet.auu.dk
alin_argeseanu@yahoo.com
Abstract: This paper presents a novel soft actuator using magneto rheological fluid
enclosed in flexible membranes [1]. The goal was to obtain the largest amount of
force with a given excitation and layered membranes. The amount of force resulted
from the interaction between the multilayered membranes filled with the magnetic
field produced by the excitation was measured. The phenomena were described and
resulting forces are presented and explained herewith.
Keywords: Actuator, magneto rheological (MR) fluid
1. Introduction
MR fluids are magnetically controllable fluids that consist of iron powder (commonly
20-40%) suspended in a carrier fluid [2], [3], [4], [5]. In the presence of a magnetic
field, the magnetic particles in the fluid align to the field lines. This implies an
increase stiffness and yield strength of the exposed fluid [5].
The fluid was invented in the 1940’s [4] by Jacob Rabinow. He carried out
experiments to determine the ingredients and the concentration to be used. Also
performance tests were carried out to determine the torque obtained when the fluid
is exposed to a magnetic field. Only recently these kinds of fluids started to be used
in industrial applications.[6], [7].
2. Experimental Arrangement
In the following, the experiment performed is detailed. Materials used are
presented. Results of the experiments are discussed and a mathematical model
describing the phenomena is argued for.
1.1
Experiment Description
The excitation is supplied from a laboratory supply source was set to output 3A,
10V. In Figure 1, components of the actuator are presented. The soft component of
the actuator consists of multilayered membranes filled with MR fluid. The magnetic
field is produced by the two windings placed on the iron core. In this context, the
field path is the one shown in Figure 1. The air gap of the circuit is where the soft
component of the actuator is. Because the membrane contains iron, the
permeability of the air gap increases. This makes the air gap seem smaller as far as
the field is concerned.
The membrane filled with ferrofluid is places in the excitation. The bolt was left in,
so a partial flux guide will touch the bottom of the membrane. As the reluctance of
the MR fluid is smaller than that of air, the flux would pass through it and close in
the upper iron core.
With this, the active length of the magnetic circuit becomes approximately 20mm.
As a result, forces acting on the membrane walls pull the filled membrane the
direction of the force. The effect of the phenomena is latching of the mobile part of
the actuator.
General information on the electromagnetic circuit:
Force required to move the entire magnetic circuit setup along the table is 3, 15 N
(unlaquered wood-support frame of the iron core- on lacquered wood-table. Weight
of the magnetic circuit was 980g.
4
2
1
1
3
F
F
B
B
6
5
Figure 1: Working principle of MR actuator
In Figure 1:
1 – windings
4 – MR filled membrane
2 – soft component of the actuator
5 – lamination core
3 – vinyl protection membranes
6 – bolt
The setup used for the actuator is the one presented in Figure 2. The actuator
windings are supplied with DC voltage from the laboratory supply. Voltage and
current are measured at the output of the source. The goal is to lift the excitation
and additional weights by pulling on the membrane when the actuator is latched.
Figure 2: Test Setup for the M2 and M3
1.2
Materials Used
The membranes used were vinyl glove material: 3 filled with MR fluid and 1 safety
membrane. The weight of the powder filled membrane was 36 g with a diameter of
the membrane of 18.5mm. The length of the active membrane = 56.5mm. Fluid
used to fill he membranes was MR 2434.
Excitation windings were supplied from a laboratory DC supply
2. Results
The opposed force by the hardened powder was 30N (measurement done using a
cup). The weight measurements on the actuator revealed a two Kg lifting capability
of the layered membrane arrangement [1].
3. Mathematical Model of the Soft Actuator Load
When testing the soft actuator at a certain load, the actuator releases.
In the following, the forces acting on the membrane are analysed. Types of forces
that intervene in the functioning of the actuator are
Magnetic forces, Fm with the help of which the membrane is drawn toward the iron
core of the excitation. There are two forces perpendicular on the surface of the
membrane and implicitly on the symmetry axe (see Figure 3). Both forces are
oriented towards the core.
As the two forces are perpendicular on the surface of the membrane and also on
the red line of the core and having the specific direction shown in the figure, the
forces will pull the membrane on both sides.
Gravitational force acting on the actuator (Gtot) represents the influence of the
gravitational constant of the earth and the masses of both the actuator and the
additional weights added for testing purposes.
Friction forces on the boundary surface between the soft membrane and the iron
core (marked with red in Figure 3). Friction forces have two main origins, namely:
friction coefficients of the meeting surfaces and the action of the magnetic forces.
When the actuator breaks (force test) friction forces become equal to the total
weight of the system.
Notations:
Ffl - left side friction force; Ffr - right side friction force
F fl = F fr = F f
F f = c f Sδ Fm
(1)
Where
c f - friction coefficient between the membrane and the iron core
Sδ - air gap cross section (where B is uniform - at the red meeting surface)
Fm - attraction force determined by the electromagnetic field and the MR fluid
Figure 3: Force Analysis of the Prototype
Fm may be estimated by using the formula of the transfer force for one air gap in
electromagnets
1 Φ2
Fm =
(2)
2 µ 0 Sδ
where
Φ =magnetic flux through Sδ
Φ = Sδ B
(3)
Total friction force
F fT = 2c f Sδ FT = 2 c f Sδ
1 Φ2
1 Φ2
1 2
= c f Sδ
=cf
Φ
2 µ 0 Sδ
µ0 Sδ
µ0
(4)
Determination of the friction coefficient the following steps were followed:
the MR filed membrane is placed in the excitation frame like in Figure 3.
contact surface is estimated. The contour of the surface is drawn on squared
paper in addition to the dimension of the rest of the actuator, letting the
surface be visible. With a piece of string the actuator is tied to a force
measuring equipment
the actuator is pulled by pulling on the dynamometer (T vector in Figure 3).
The movement must be uniform.
measured force represents the friction force and it is equal to:
F f = c f S f GSA
(5)
were
Sf - contact surface between the membrane and the excitation; GSA -weight of the
membrane
GSA = mSA g
(6)
The friction coefficient is:
cf =
Ff
S f mSAG
(7)
mSA - mass of the membrane;
g - acceleration due to gravity
Figure 4: Friction Coefficient Determination
When friction forces (influenced by Sδ and c f ) are greater or equal to the equivalent
weight of the system, the excitation and additional weights are lifted.
When the friction forces are smaller than the weight, the membrane releases the
excitation and the additional weights. This is what is called breakdown test of the
actuator.
At the limit, equality between the two forces is assumed:
GT = F fT
GT = Gmm + Geem
where
(8)
GT - total weight
Gmm - weight of additional weights
Geem – weight of the excitation
By measuring the mechanical force, an expression of the electromagnetic force
may be obtained. This is done by assuming the moving part of the actuator is acting
like an electromagnet.
From the final form of the friction force
FfT = c f
1 2
Φ
µ0
(9)
It could be observed that the magnetic flux at the active surface may be estimated.
Knowing Sδ (marked with red in Figure 3), a value of the magnetic inductance B
may be calculated.
Calculated value may be verifying it by measurements. These equations could be
used for simulating the actuator.
4. Conclusion
It is possible to build a soft actuator having smart fluid. In essence the actuator is
a prime mover. By study of the smart materials a vast number of applications
could be envisaged. Combining movements of an array of simple actuators, a
complex moving trajectory may be obtained. By using multiple layers, the
mechanical force of the actuator will increase.
A safe level of current can produce a substantial amount of force. The flexibility of
the membrane makes the actuator more flexible than a solid one.
By using layers of MR fluid, each layer interacts with the adjacent membrane. The
content of the membrane produces forces that push the container walls in the
direction set by the magnetic field (see Figure 1). In this way each layer facilitates
latching.
As the membrane interacts with the oil from the MR fluid, additional membranes
will shield the MR container thus making it more resistant. A secondary function of
extra skin is to increase mechanical resistance.
References
[1]
[2]
[3]
[4]
[5]
Krisztina Leban, Ewen Ritchie, Alin Argeseanu, Novel Soft Actuator. Power Electronics and
rd
Drives MSc Programme 3 semester project January 2009
Rheology of Field polymer systems Aroon V. Shenoy Kluwer Academic Publishers 1999 ISBN 0412-83100-7
Rheology Fundamentals Alexander Ya. Malkin, ChemTec publishing 1994 ISBN 1-895198-09-7
Fernando D. Goncalves, Jeong-Hoi Koo and Mehdi Ahmadian ”A Review of the State of the Art
in Magnetorheological Fluid Technologies - Part I: MR fluid and MR fluid models” The Shock
and Vibration Digest. Vol. 38. No.3. May 2006 203-219, rD2006 SAGE Publications, DOl:
10.1177/0583102406065099
Development and Experiments of Actuator Using MR Fluid Naoyuki Takesue, Hiroy
asu Asaoka, Jing Lin, Masamichi Sakaguchi, Guoguang Zhang, Junji Furusho Departament
of Computer-Controled Mechanical Systems Graduate School of Engineering, Osaka
University, Japan IEEE TRANSACTIONS ON MAGNETICS, VOL. 40, NO. 4, JULY 2004
Ralph Jansen Ohio Aerospace Institute Brook Park, Ohio; EIiseo DiRusso Lewis Research
Center Cleveland, Ohio Passive Magnetic Beating With Ferrofluid Stabilization NASA Technical
Memorandum 107154; February 1996 National Aeronautics and Space Administration
[7] Fernando D. Goncalves, Jeong-Hoi Koo and Mehdi Ahmadian ”A Review of the State of the Art
in Magnetorheological Fluid Technologies - Part I: MR fluid and MR fluid models” The Shock
and Vibration Digest. Vol. 38. No.3. May 2006 203-219, rD2006 SAGE Publications, DOl:
10.1177/0583102406065099
[6]
Partial Discharge Data Denoising and Feature Extraction
S.Vivekananthan, S.D.R.Suresh
Department of Electrical &Electronics Engineering, College of Engineering, Guindy, Anna University, Chennai-25

Abstract—One of the major challenges of modern-day online partial discharge (PD) measurement in High voltage
electrical equipments is the recovery of PD signals from a noisy
environment. The different sources of noise include thermal or
resistor noise added by the measuring circuit and high-frequency
sinusoidal signals that electromagnetically couple from radio
broad casts and/or carrier wave communications. Sophisticated
methods are required to detect PD signals correctly. Fortunately,
advances in Analog-to-Digital conversion (ADC) technology, and
recent developments in Digital Signal Processing (DSP) enable
easy extraction of PD signals. This paper deals with the analysis
of noise and denoising of PD signals using wavelet method. This
denoising method is employed on both simulated as well as real
PD data using MATLAB. Also statistical analysis is done on
different kind of partial discharges for feature extraction.
Index Terms—Partial discharge, wavelet, Threshold,
Corona discharge, De-noise, Coupling capacitor, High voltage
I. INTRODUCTION
P
artial discharge (PD) tests and its analysis plays a vital
role in insulation quality assessment .Partial discharge is a
localized electrical discharge that only partially bridges the
insulation between conductors and which may or may not
occur adjacent to a conductor. PD can result from
breakdown of gas in a cavity, breakdown of gas in an
electrical tree channel, breakdown along an interface, or
breakdown between an energized electrode and a floating
conductor, etc. Based upon this partial discharge can be
classified as corona discharges, surface discharges , and
internal discharges such as treeing and cavity discharges.
Determination of PD activities in measurements can be
extremely difficult due to sources of interference such as
radio and telecommunication signals and switching
transients especially in online monitoring.
II. PARTIAL DISCHARGE
AND MEASUREMENT
Figure 1 PD pulse characteristics
In practical measurements, discharge voltage signals are
captured by feeding the discharge current through a
detection circuit. On this basis, detected voltage signals are
likely to have different pulse shapes, depending on the
configuration of the detection circuit. In practice, the current
pulse i ( f ) caused by a PD is not an ideal Dirac current,
because it has duration in time. When a practical PD pulse
passes i(t) through a detection circuit, the output pulse
produced always has a finite rise time. Considering the shape
of a PD current pulse and the characteristics of detection
circuits, the pd pulse is classified as damped exponential
pulse (DEP) and the damped oscillatory pulse (DOP) and
further displayed in Figure below.

DEP(t )  A e t / t1  e t / t 2


DOP(t) = Asin(2fc t ) e t / t1  e t / t 2

Where A gives the pulse peak value, t1 , t2 , the damping
coefficients, whilst f is the oscillatory frequency of the DOPtype pulse
CHARACTERISTICS
A PD pulse in a dielectric gives rise to an electromagnetic
pulse (after the cavity collapses) with a rise time in the ns
range and a pulse width in the range of 1.5 ns-300µs. The
resulting voltage pulse propagates in both directions away
from the PD source. The optimum bandwidth for detection
of such a “fast” pulse is in the range of 2kHz-300 MHz.
Discharge from electrical trees in aged solid dielectrics often
takes the form of a cascade of such fast pulses. In aged
cavities, the PD pulse can broaden to have a rise time of up
to several tens of ns and a pulse width up to some hundreds
of ns. Figure 1 demonstrates a typical PD current pulse,
which could be characterized by a series of discharge
parameters including A (pulse peak value), f, (pulse rise time
from 10% to 90% levels), t, (pulse width between 50%
levels) and fd (pulse decay time from 90% to 10% levels.
Figure 2
(a) DEP Signal (b) DOP Signal
The detection and measurement of discharges is based on
the exchange of energy taking place during the discharge.
Over the past forty years, several methods have been
developed to detect PDs within electrical equipments These
can be grouped into four categories, based on the PD
manifestation that they measure chemical, electrical,
acoustic and optical detection. Among these electrical
(coupling capacitor) method is selected for detection of
partial discharge. Because online monitoring needs
continuous measurement of the PD pulses. In order to
measure the PD signal during operation of the HV
equipment this is the effective measuring technique
Figure 3 Coupling capacitor PD measuring system
In figure 3 U is a step voltage on the detection impedance
Zd, resulting from the apparent charge in the presence of PD
in the test specimen Ca, with Cc is the coupling capacitor.
III. NOISE ANALYSIS
A major bottleneck encountered with on-line/on-site PD
measurements is the ingress of external interferences
(usually of very high amplitude comparable to PD signal)
that directly affects the sensitivity and reliability of acquired
PD data. Digital signal processing techniques must be
applied to on-line PD measurements to recognize PD pulses
within the EMI background, which often swamps the PD
signal on-site. The ability to discriminate the PD signal from
the noise requires knowledge of both the PDs and the noise.
In general, noise sources may be divided into the following :
1. Discrete spectral interference (DSI), narrowband
interference caused by e.g., radio broadcasts due to
amplitude
modulation/frequency
modulation
(AM/FM) radio emissions and communication
networks
2. Periodical pulse shaped disturbances, repetitive
pulses caused by e.g., corona discharges, other
discharges due to transformers or power electronics
3. Stochastic pulse shaped disturbances, random
pulses caused by e.g., lightning or switching
operations, PD and corona from the power system
which can get coupled to the apparatus under test
4. White noise, broadband interference caused by e.g.,
the measuring instrument itself.
In terms of its nature DSI is a narrow band signal, the other
three types of noise are wide band signals. DSI can be
removed by applying digital notch filters with the help of
spectrum analyzer. Repetitive pulses can be rejected through
gating circuits in time domain . Random pulse interference is
the most common interference source and presents the
biggest problem to online PD measurement. An advanced
technique has to be implemented to separate PDs from the
random pulses. Several denoising methods are analyzed by
sathish et al and Nasser and they concluded based on the
Mean Square Errors (MSEs) and the time taken to perform
denoising wavelet based denoising method is suitable for
denoising the PD signals.
IV. WAVELET TRANSFORM
Wavelet transform theory has well explained by many
authors. Briefly the wavelet transform decomposes a signal
from the time domain into a time scale domain with
expression of a set of shifted and scaled versions of a single
prototype function
Ψa,b(t)=a-1/2 *Ψ((t-a)/b), a,b Є R, a≠0
Where a and b are the wavelet scaling and shifting
parameters respectively.
Choosing scales and positions based on powers of two
gives a more efficient analysis with equal accuracy . the
discrete wavelet transform (DWT) can be obtained through
use of multi-resolution signal decomposition . The time
domain original signal is passed through a series of
complimentary high pass filters (H) and down sampled by
two to generate higher frequency coefficients , passed
through low pass filters (L) and down sampled by two to
produce lower frequency coefficients (approximation) at
different scales . These filters are called quadrature mirror
filters (QMF). QMF enable signals to be decomposed
without loss of original signal lower frequency coefficients
can be further decomposed into next level approximation
coefficients and detail until the desired resolution is
achieved .
Figure 4 Wavelet Decomposition
Reconstruction of the signal is an inverse process of
decomposition. Coefficients obtained through DWT or
modified coefficients can be used to reconstruct the original
signal.
Figure 5 Reconstruction
V. WAVELET DENOISING TECHNIQUE
(a) ALGORITHM FOR PD SIGNAL DENOISING
1.Simulate a noise free pulse, which is similar to the
detected real PD signal. This may require the knowledge of
the likely PD source, propagation media and path, method of
detection, the frequency response of the sensor or the
detection circuit and so on.
2.Convolute between a series of mother wavelets and the
simulated pulse to determine the best candidate mother
wavelet.
3.Calculate the maximum of the approximation coefficients
of the simulated pulse on each level to determine the
possible numbers of the decomposition levels.
4 .Determine the threshold method according to the purpose
of the application
5. Determine the coefficients which are mainly associated
with noise and set them to zero
6.Decompose the noisy PD signal via the DWT
7.Apply the threshold to the remaining coefficients
8.Reconstruct the signal by using the modified coefficients
pure signal and this signal having rise time in micro seconds
and bandwidth of 150Khz to 800Khz with 50mv. This signal
is taken in a shielded room. And is shown in figure 6.
(b) MOTHER WAVELET SELECTION
The selection of the mother wavelet is crucial in the
DWT-based denoising algorithm. As each mother wavelet
associates with a filter, the result of DWT or the value of the
convolution between the pulse under analysis and the mother
wavelet implies how much energy of the pulse passes
through the filter. More energy of the pulse passing through
the filter means more information of the pulse is kept.
Hence, the better denoising effect could be achieved.
Different kind of mother Wavelets are Hear wavelet,
Mexican Hat, Biorthogonal, Coiflets, Morley, Daubechies
wavelet , Symlets .A comparison of Signal-to-Noise Ratio
(SNR) has been used for the estimation of the performance
of the mother wavelet for de-noising application. Noisy
signal has been considered as I(i) = O(i)+σe(i).Where I is
original signal and O is the de-noised signal.
SNR=10(log10 ∑I(i)2-log10 ∑(I(i)-O(i))2).
According to this Daubechies wavelet (DB2) has higher
SNR ratio of 12 when compared to mother wavelets so this
is chosen for the denoising of PD signal.
Figure 6
Pure PD signal
Using MATLAB this PD pulse is taken and it is mixed with
simulated noise such as DSI, gauss ion white noise and
Random noises .These noises are simulated with
mathematical models .DSI is with 12 different frequencies
from 0.1 to 12 MHz and maximum magnitude in 16 my.
White noise is standard zero mean Gaussian white noise
with maximum magnitude of 18mv. Stochastic noise is
simulated by random entries chosen from a uniform
distribution on the interval (0.1,1.0) multiplying with a gain
of 16mv. The DSI, Gaussian white noise and random noises
are shown in figure respectively.
(c) NUMBER OF DECOMPOSITION LEVELS
PD pulse and noise all show different frequency
spectrum. Their associated coefficients appear on different
levels following the DWT. On the level where the main
energy of signal is located, the approximation of the signal
achieves its maximal value. PD pulse reaches its maximal
value of approximation coefficients on level 3 in the DWT
Which is called as the critical level for the signal. The
optimal decomposition level should be around the critical
level. It may be a level more or less due to the unknown
influence caused by noise. That is to say, for denoising PD
pulse, the optimal number of the decomposition levels might
be 2, or 3, or 4. Here the level taken as 3.
Figure 7 Noise characteristics
After adding these noises with the pure signal becomes
noisy. the noisy PD signal is shown in below figure
THRESHOLDING METHOD
Thresholding is a key step in wavelet denoising. It
applies the threshold to the decomposition coefficients,
keeping those coefficients above the threshold and
discarding those below. Various investigations by Donoho
have shows that zero can replace small wavelet coefficients,
because they are dominated by noise and carry only a small
amount of signal information. Thus, by ignoring these
coefficients when reconstructing the original signal from the
rest of the wavelet coefficients, will result in a de-noised
version of the original. This idea is the basis for method of
thresholding. Intuitively hard thresholding preserves the
peak height of signal better but yields less smooth fits.
Whilst soft thresholding smoothes the signal but reduces the
signal magnitude. Here the fixed form of thresholding is
chosen to denoise the noise PD signal.
Figure 8 Noisy PD signal
Using MATLAB wave menu the extraction of PD signal
is done. The extraction of PD signal is obtained by the
proposed method in this paper. The PD pulse and their
approximation and detailed coefficients are shown in figure
below
VI. SIMULATION AND ITS RESULTS
The standard PD pulse according to the IEC 60270 was
produced by the ADG PD generator this signal is taken as
Figure 9 Denoised PD signal
parameters using Principle component analysis for data
reduction .
VIII.
CONCLUSION
1. In this paper PD signal characteristics and various noises
interrupting the online Measurement was analyzed. Also the
selection of suitable wavelet and its selection was analyzed.
2. The proposed wavelet method is implemented to extract
PD signal from noisy PD signal
Figure 10 Approximation and Detailed coefficients
The same method can also be used for the extraction of train
of PD pulses from the noisy environment as shown in below
figure 11.
3. The correlation between the pure and denoised PD signal
is calculated by the correlation formula
4. The correlation between the pure PD signal and noised
signal = 85.6,The correlation between the pure and denoised
PD signal = 94.3 But the reduction of the amplitude of
denoised PD pulse is 9.66% when compared to the pure PD
pulse. And statistical analysis can be studied further.
IX. REFERENCES
1.
Figure 11 Train of Noisy PD pulses and Denoised
Pulses
VII. STATISTICAL ANALYSIS
The feature Extraction of PD pulses is carried out by
conducting various kind of experiment on different
specimens and capturing the PD signals. A typical set up is
shown in figure 12.
2.
3.
4.
5.
Hao Zhang, T.R. Blackburn, B.T. Phung and D. Sen “A
Novel Wavelet Transform Technique for On-line Partial
Discharge
Measurements Part 1: On-site Noise
Rejection Application” IEEE Transactions on
Dielectrics and Electrical Insulation ,February 2007
Sathish and B.Nazneen“Wavelet Based denoising of
Partial discharge signals buried in excessive noise and
interference” IEEE Transactions on Dielectrics and
Electrical Insulation April 2003
S. Sriram,S.Nitin, K.M.M. Prabhu, and M.J. Bastiaans,
“Signal Denoising Techniques for Partial Discharge
Measurements” IEEE Trans. Signal Processing .p1-p9
Wang Hang, Tan Kexiong, Zhu Deheng “Extraction of
Partial Discharge Signals Using Wavelet Transform”
Proceedings of IEEE 5th International Conference on
Properties and Applications of Dielectric Materials
,May 25-30,1997, Seoul, Korea
Feng Zhang, jian Li, Rujinin Lioa “Aged Oil –paper
classification using statistical parameters and clustering
analysis” 2007 IEEE conference on Electrical insulation
and Dielectric phenomena.
X. ABOUT THE AUTHORS
Figure 12 Air corona experimental setup
A set of 20 data has been taken for each Air corona, cavity
discharge and pressboard corona and surface discharge PD
signals through coupling capacitor method. Later this set of
data is fed to the Matlab to find out the statistical parameters
such as mean , kurtosis, skewness, and number of pulses.
And these parameters is used for determination of mean
pulse height distribution, maximum pulse height distribution
and pulse count distribution for classification of PD signals.
For online monitoring these signals are first denoised by the
above methodology described in this paper and then these
27 statistical parameters are reduced to 9 statistical
Vivekananthan.s obtained his B.E (EEE) degree from
Kamban engineering college, Anna University in the year
2007. Presently he is pursuing M.E (High voltage engg)
degree in College of engineering, Guindy, Anna University,
chennai.
Suresh.S.D.R obtained his B.E.(EEE) from Government
college of engineering, Manonmanium Sundaranar
university,Tirunelveli and completed his M.E (High voltage
engineering)degree in college of engineering,Guindy, Anna
university, Chennai in the years 1996,2001 respectively.
Presently he is pursuing Ph.d degree (High voltage engg.) in
College of engineering, Guindy, Anna University Chennai.
1
Brain-computer interfaces: to optimise or to
commercialise?
Suzan de Goede
Center for Sensory-Motor Interaction, Department of Health Science and Technology, Aalborg University,
Aalborg, Denmark
Abstract—Brain computer interfaces (BCI) provide
a mean of communication for severely paralysed patients. These systems bypass the brain’s usual output
channels, such as the muscles or speech and use the
brain signals to control an external device. Currently,
there are no BCI systems commercially available for
paralysed patients. A review of the literature has been
performed to assess what the current state of art in BCI
technology is. An analysis is made of the capabilities
and limitations of clinically applied BCI. The performance and sophistication of the currently available BCI
technology is at a high level. However, for a commercial
system to be marketed, some gaps need to be bridged.
Some essential requirements for a BCI system to be
used on day-to-day basis will be discussed.
I. Introduction
N the 1980’s, the first brain computer interface (BCI)
that has been used by severely paralysed patients was
created by Farwell and Donchin [1], [2]. The electroencephalogram (EEG) recorded from the scalp of a subject
is used to control a spelling machine. A matrix with 9
letters is presented to the subject. When the desired letter
lights up on the screen, a positive peak can be noticed
approximately 300 ms after the presentation [1].
This is the first time that signals of the brain are used
to succesfully control an external device. Several different
types of BCIs have followed since and the main focus of
these systems is to control an external device without
requiring muscle activity. A generally accepted defenition
has been established by Wolpaw [3]:
’A BCI is a communication system that provides the brain
with a new, non-muscular communication and control
channel.’
This is a very usefull technology for people who are
completely paralysed, due to spinal cord injury, cerebral
palsy or degenerative diseases such as Parkinson’s or amyotrophic lateral sclerosis (ALS). Such patients are severly
restricted in their means of communication. Currently,
any means possible are used to communicate [4]. The
question that will be addressed in this review is whether
BCI technology can become a method, which can be used
on a day to day basis for such patients.
The search methods to identify articles relevent in BCI
literature is first mentioned. There after, the components
of a general BCI system are mentioned, followed by the
brain signals that have been clinically applied to control a
BCI. In section V, the state of art of BCI systems used by
paralysed subjects is reviewed. Finally, the steps that need
to be taken to commercialise a BCI system are discussed.
I
II. Methods
Exhaustive searches were performed in PubMed and
Ovid Medline, using the keywords ’Brain Computer Interfaces’ and ’BCI’ to identify all recent literature in this field.
The identified references were then sorted into different
categories:
• BCI methods
• Noise reduction
• Detection and classification algorithms
• Feedback paradigms
• Applications
• Reviews
Despite the broad literature search, this article focusses
on the state of art of clinically applied BCIs that are used
online.
III. BCI components
There are several translational steps before the input to
the BCI, the user’s EEG signals, results in the control of
an external device, these are shown in figure 1.
Firstly, the brain signal is recorded from the scalp, from
the cortex or from single neurons. Currently, several BCI
systems using EEG signals are now clinically applied. BCI
systems using more invasive recording methods have not
yet reached this stage [5]–[8], therefore the focus of this
article will be on EEG based BCIs. These signals are
usually filtered using both analogue and digital filters to
enhance the signal quality. The EEG frequencies of interest
are usually below 30 Herz, where distinct frequency bands
are related to different functionalities [9]–[11]. Despite the
fact that the higher frequencies have not been described,
they are also investigated for potential BCI use [12].
Besides general filtering techniques, much attention has
been paid to seperation of sources in EEG signals.
The surface EEG is usually contaminated with electrooculographic (EOG) and electromyographic (EMG) signals, besides the usual background noise. Different source
seperation methods [13] and component analyses [14], [15]
have been applied to improve the acquired signals.
From the filtered EEG, specific features must be chosen
to differentiate between two or more system outputs.
These features are related to the task performed by the
user. The user can be asked to select a letter from a
matrix and and event-related potential (ERP) will occur
involuntarily when the selected letter is presented [1], [16].
But the user may also be asked to voluntarily produce
specific brain signals, for example to achieve movement of
2
Figure. 1: The input of a BCI system are the user’s brain signals
recorded intra- or extracranially. This signal often requires filtering
to improve the signal quality, where after specific features are extracted from the signal. Such features can be the amplitude, specific
frequency content, event-related potentials or single-neuron activity.
These features can be translated into a specific command, to control
a prosthetic limb or to select a letter from a sequence [adapted from
[3] and [6]].
a cursor [17]–[19].
In either situation, a feature that distinguishes between
different outputs must be selected. Such a feature can
be chosen a-priori, based on physiological knowledge,
for example. Another option is to use a feature extraction method, for example wavelets or principle components analysis, to select features. Hereafter, the system
is trained using labeled datasets, containing EEG signals
corresponding to the different outputs the system should
provide. When an optimal classifier has been obtained, the
system is able to distinguish between the different classes
of signals that were present in the training data. Each class
of signals can be used as a command to control a device
[5], [6], [17], [20].
Figure. 2: The different components of an event-related potential.
The P3 or P300 positive peak and the slow negative wave, consisting
of the initial and the terminal contingent negative variation (iCNV
and tCNV, respectively) are used for BCI control [30].
The first BCI was based on the detection of this peak
in the EEG. Many different algorithms have been used
in an attempt to correctly detect the signal. From simple
peak picking algorithms [1] to discrete wavelet transforms
[2]. Three methods have succesful: a stepwise discriminant
algorithm (SWDA) [31], a SWDA in combination with
discrete wavelet transform (DWT) and independent component analysis (IDA) [16].
Though the output of this BCI is binary, it can be very fast
due to the short latency of the signal. In addition, the user
does not need to learn to control the BCI, since the brain
signal of interest is produced involuntarily. However, the
system can only be used if the user is able to interpret the
stimulus and receive feedback with respect to its selection.
A. P300 BCIs
B. Slow cortical potentials
The SCP is also shown in figure 2, it is the slow wave of
negative following the P300. However, the SCP is not always related to a specific event. Negative SCPs are usually
associated with a preparatory depolarisation of the related
cortical areas, also referred to as Bereitschaftspotential.
Positive SCPs are on the other hand related to inhibition
or distraction. It has been shown possible to control SCP
in an area specific manner. SCP control is learned using
biofeedback regarding the SCP shift. When control over
the SCP has been achieved, it is possible to change it
without feedback [6], [30], [32].
The amplitude of the SCP is used to control the BCI
in different ways. The amplitude over the center of the
scalp, recorded from Cz according to the 10-20 system, is
used to control a cursos vertically. Horizontal movement of
the cursor is controlled by the SCP gradient over the left
and right hemispere (C3 and C4) [17].This BCI provides
a continuous output, but the system requires that the
user can control eye movement in order to see and follow
the cursor. Furthermore, the user must learn to control
the slow cortical potentials that occur over the scalp.
Though BCI control is considered a skill [3] and is thus
comparable to learning, for example, how to write; the
learning trajectory might be a barrier for potential users.
ERP consists of several positive and negative variations
in the EEG, as shown in figure 2. The P300 is a positive
peak which appears in the EEG approximately 300 to
1000 ms after a stimulus is recognised, depending on
the complexity of the stimulus. This involuntary response
cannot be learned or controlled. The amplitude of the
P300 peak depends on the relevance and the probability
of occurence of the stimulus [5].
C. Sensorimotor rhythm
The sensorimotor rhythm occurs over the left and right
hemispheric sensorimotor areas of the cortex and can
be voluntarily controlled [5], [33]. The manner in which
this EEG feature is used to control a BCI shows much
resemblence to the SCP BCI. The sensorimotor rhythm
is typically applied to control a cursor. The power in a
IV. Brain signals for BCI control
Different features of the brain signals can be and have
been used in different ways to control an external device.
There are vast amounts of different pattern recognition
classification methods that have been applied to the signals mentioned above. Examples of methods to detect an
event range from a simple peak picking algorithm [1] to
wavelets [21]–[23]. To classify the detected events, methods have an equal extent of complexity. Both linear and
non-linear methods have been applied in supervised and
unsupervised manners. Examples of classification methods
are neural networks [24] and support vector machines [21],
[25], [26]. Reviews of the methods that have been applied
in the past can be found in [27]–[29].
Three types of signals have been applied and verified online: slow cortical potentials (SCP), sensorimotor rhythm
and P300 event related potentials (ERP).
3
specific frequency band of the EEG spectrum determines
the movement of the cursor. The frequencies used range
from 8 to 12 Herz, the µ rhythm, and from 18 to 26 Herz,
the β rhythm [33]. These two frequencies are suppressed
during imagination of movement and actual movement [5],
[6].
The weighted sum of the power of the beta and mu
frequencies is used to control a cursor. The movement
of the cursor in the second dimension is defined by the
difference between the two hemisperes [33].
Though generation and control of the sensorimotor rhytm
does not require any muscle activity, the user must be able
to control gaze in order to control the cursor. In addition,
controlling the sensorimotor rhythm must be learned.
V. State of the art in BCI performance
A. Information transfer rates
The performance of BCI systems that have been applied online, in terms of speed, accuracy and information
transfer rate (ITR), is shown in table I. In this table,
only experimental data with paralysed subjects reported
in literature has been used and converted to obtain the
same units for all BCIs.
Higher ITR have been reported using the Berlin BCI, 2030 bits/min [6], [36]. However, this system has only been
used by able bodied subjects and performance levels may
be different in patients. ITR, also called bit rate, is a
measure that combines accuracy and speed of communication [3]. A higher ITR can be achieved through an
increase in accuracy, number of options each trial, and a
decrease in the duration of each trial. However, the speed
and accuracy of the algorithm and the latency of the signal
are also important to be able to interpret the ITR.
Accuracy describes the agreement between what the user
desires to choose and what the BCI interprets. This implies
that accuracy is both dependent on the user’s ability to
use the BCI and on the ability of the BCI to interpret the
user’s brain signal. Accuracy implies that the user’s choice
is correct, which is not always the case using, for example,
a keyboard or even speech [20].
Increasing the options in each trial to increase the ITR
can negatively influence the accuracy. The probability of
misclassification by the BCI will increase, as there is only
1 right option among the number of options provided.
Furthermore, the user may be distracted or confused,
which affects the brain signals used for classification.
Decreasing the duration of a trial may increase the bit
rate, but can also negatively affect the accuracy. The signal
may not capture all information necessary for correct
classification and a short decision time may be confusing
for the user as well. [3], [18].
Several studies [18], [37]–[39] have been performed to
optimise these parameters to increase the ITR. However,
results have not provided one consistent method to increase ITR. Rather, the number of targets and the trial
duration may be optimised for each user individually to
achieve an optimal ITR.
What has been found useful is the presence of an error
potential, first described by Schalk [40] in the light of BCI.
This has later been applied to detect errors [38], [39], [41],
but has not yet been implemented in online BCI systems.
Implementation must also be carefully considdered, as it
provides additional information for the user to process.
This may be confusing and thus decrease the ITR. The
most straight-forward implementation method would be
to simply interrupt execution of the command, when an
error potential is detected.
B. User interfaces
The second important aspect that could improve the
usability of BCI is the actual user interface. The user
interface of the BCI consists both of the recording electrodes and apparatus and the media providing the task
and feedback. Focus will be on the latter.
Presentation of a task to the user can be done in many
different ways. The most common one has been the P300
speller matrix from Farwell and Donchin [1], shown in
figure 3. From a functional point of view, it is a good design, though other designs for spelling machines have been
proposed. One of them is the Hex-o-spell, as presented in
figure 4. This interface was implemented in the Berlin BCI
[42].
Since patients may not be able to control gaze and may
thus not see the spelling matrix, an auditory BCI has been
proposed. This has been tested using both healthy and
paralysed subjects [43], [44]. However, results were not
satisfactory. This may be due to the fact that an auditory
matrix with a visual support matrix is rather complex.
Another interface may improve this method significantly,
since the users were able to use the visual P300 spelling
BCI.
The interfaces used for SMR and SCP based BCIs, shown
in figure 5, were initially very basic as well. The task
presented to the user is to move the cursor to a target.
For experimental research, such tasks are useful and the
results are easy to interpret, but for the user the task may
not be interesting.
A first step to improve this has been to use virtual reality.
The subject is now in a street or an appartment in which
tasks should be fullfilled. This was applied first in 2004
Figure. 3: The P300 spelling matrix [2].
4
Figure. 4: Hex-o-spell [42].
[45] and has been investigated extensively [46]–[49], even
with paralysed subjects [50]. The results from these studies
have not been quantified in terms of ITR, but it has been
shown that feedback from a virtual environment decreases
the minimal error rate [51]. Furthermore, the experiment
in virtual reality has been applied succesfully with patients
controlling the course of their wheel-chairs [50]. However,
as mentioned in section IV, even the most basic task
cannot be performed when a subject cannot control gaze.
Therefore, Bianchi et al [52] propose another type of
feedback. Instead of moving the cursor to the target, as
shown in figure 5, they propose that the background is
moved and the cursor remains at the center of the screen.
Figure. 5: 1. Presentation of the cursor and target. 2. Moving
the cursor towards the target. 3. The target flashes when it is hit.
[Adapted from [18]]
The last aspect of the user interface is feedback. In most
of the discussed BCI protocols, the feedback is provided
visually or vision is required to even perform the task.
Attempts have been described to use auditory systems for
the P300 spelling BCI [43], [44], without much success.
A simpler approach has been applied to the sensorimotor
rhythm BCI. This has been more succesful, though users
required a longer training period to achieve the same
accuracy as with a visual system [53].
VI. Optimisation versus commercialisation
A lot has been achieved in the area of BCI, though there
still is much that can be optimised. What should be the
area of attention?
Detection and classification algorithms have had a lot of
attention in recent years. But without properly filtered and
denoised signals, the accuracy of such algorithms would
not improve much. Another option is to look for new signal
types and signal features to use for BCI control or to look
for methods that require less training than SMR and SCP
BCI do.
But perhaps the more relevant question is, will the potential user benefit from such efforts? Currenly, no BCI
are commercially available and only a selected group of
patients benefits from the efforts made in BCI research.
Judging the state-of-art in BCI technology, the technology
may be ready to be commercialised.
A. Hardware
However, several steps need to be taken in order to bring
a BCI system on the market. The systems that are used
Table I: Performance of currently available BCIs in terms of accuracy, speed and information transfer.
Developers
Wadsworth center [34]
Collaboration of Italian research groups [16]
Graz BCI [35]
University of Tuebingen [17]
Wadsworth center [18] 1
Graz BCI [19]
1
Based on
P300
P300
P300
SCP
SMR
SMR
Trial duration
175 ms
2.5 sec
1.68 s
4s
3s
3-5.5 s
The results from this study include both healthy and disabled subjects.
Accuracy
78.8%
8.6 %
75 %
85 %
77 %
82.2 %
ITR
9.7 bits/min
7.77 bits/min
22.04 bits/min
4.9 bits/min
7.49 bits/min
8-17 bits/min
5
for research require a shielded lab to record EEG signals,
a set of equipment to amplify, filter and digitally acquire
the signal and a fast PC to run the algorithms and user
interfaces. A potential user would be interested in a system
that can be used the whole day, that does not take a lot
of space and appears as a common device or perhaps even
as a gadget.
In order to live up to such wishes, the BCI should use
a minimal number of channels for classification. The
electrodes should also be small in size or perhaps made
invisible by embedding them in a fashionable accessory.
From a engineering point of view, the electrodes should
still be capable of recording proper signals and maintain
a good impedance between the electrode and the skin.
Secondly, the amplifier should be small and preferably
portable or even wearable. Such a device has been presented recently by Thorbergson et al [54]. This is a very
small device that is designed for interfaces with the nervous system. The system has a sufficient sampling rate and
can transmit data wirelessly and continuously, allowing
online detection and classification of signals. However, the
number of channels of this device is limited and its battery
life is only 23 hours. This means that it must be recharged
every other day. These drawbacks can be overcome and
finally practical tests of this system in combination with
a BCI should validate its use.
Lastly, the user interface should appear as a very common
everyday use technology, like a PC. During its design, it
should be taken into account that the potential users may
lose more and more control over their muscles. This means
that in the final stages of the disease, they might not be
able to focus gaze or to see at all. Interfaces should be
available for the different stages of the disease, so the user
does not need to purchase new devices or device extensions
as the disease progresses.
B. Software
The software used in BCI technology can be divided
into three parts: filtering and noise reduction, signal
detection and pattern classification. All of these fields
have been investigated extensively and many algorithms
are available that are able to perform these tasks.
What is important, however, is not how well these
algorithms perform on experimental data, which is
usually acquired in a laboratory, but how noisy data
is handled. It is very important to investigate which
features of the signal are used for classification, in order
to design an appropriate filter. Besides filtering, the signal
needs to be improved further, for example by separating
contributions from different sources, for example EMG
and EOG. Numerous techniques are available for this
purpose; [13]–[15] are the most recently investigated
methods.
Secondly, the detection and classification methods seem
always open to improvement. However, making such
improvements is very time consuming and the level
of improvement is usually marginal. It may therefore
be more usefull to invest some time in designing a
user-friendly method to correct mistakes to improve the
usability of the system, instead of its performance.
At last, it has been shown that not every user can learn
to work with a BCI [4]. However, in the experimental
setting, only one type of BCI is usually available. Since
learning how to use a BCI is considered a skill, it may be
usefull to implement different types of BCI (e.g. SMR,
SCP, P300, motor imagery) in one system. Such efforts
have been made in the past [55], [56], but could be
extended.
Currently, different application modes have been
integrated into one BCI system. However, more signal
types could be implemented in the same BCI. This would
increase the number of patients that could benefit from
one system. Depending on the capabilities of the user,
switches between the different application modes could be
implemented or a combination of different signals could
be used in one application mode. For example, moving
the cursor towards a target, using SMR or SCP, and
using a P300 potential to select the target.
C. Potential users
Finally, it all depends on the user. The system should
be flexible, providing sufficient applications for skilled
users, but having the option of reducing the number of
applications not to confuse unexperienced or less skillful
users. It is very important to include both patients and
care-givers into the design phase of a commercial BCI, as
they are the ones who can benefit most from this device.
Inevitably, not every potential user will benefit from this
device. Some patients may not want to use this device
and some may not be capable of learning to control a
BCI. It is very important to define a selection method to
identify those patients that will benefit from the device. It
has been shown that initial performance at BCI control
is a good predictor of final performance [57]. However,
this has only been examined for one type of BCI control,
using SCP. This should be investigated for all types of BCI
control, in order to avoid dissatisfaction among patients,
as the training protocol to acquire BCI control can be very
intensive [57], [58].
Because of the intensity of training, patients should be
well-informed regarding the training procedure and duration. The patient should enroll in training as early as
possible (in the case of degenerative diseases) to enlighten
the burden on both the trainer and the patient. When the
patient is in or near the locked-in state, communication
between the trainer and the patient is very difficult, yet
extremely important [58]. Lastly, it is important that
the patient is supported by care-givers. They should be
able to set-up the system for the patient and use it in
collaboration with the patients.
VII. Conclusions
The state of art in BCI technology suggests that it is
time to commercialise this technology for the benefit of
many paralysed patients. Though there are still many aspects of BCI that can be optimised, the final improvements
6
in ITR are expected to be marginal. In addition, they may
not be decisive for the acceptance of this technology by
potential users.
In order to commercialise BCI, its usability should be
optimised instead of its technology. Most can be achieved
in the area of hardware design and user interfaces. Though
patient selection and training are areas that require attention as well.
References
[1] L. Farwell and E. Donchin, “Talking off the top of your head:
toward a mental prosthesis utilizing event-related brain potentials,” Electroencephalography and clinical neurophysiology,
vol. 70, pp. 510 – 523, 1988.
[2] E. Donchin, K. M. Spencer, and R. Wijesinghe, “The mental
prosthesis: assessing the speed of a p300-based brain-computer
interface,” IEEE Transactions on rehabilitation engineering,
vol. 8, pp. 174 – 179, 2000.
[3] J. R. Wolpaw, N. Birbaumer, D. J. McFarland, G. Pfurtscheller,
and T. M. Vaughan, “Brain-computer interfaces for communication and control,” Journal of clinical neurophysiology, vol. 113,
pp. 767 – 791, 2002.
[4] N. Neumann and N. Birbaumer, “Predictors of successful self
control during brain-computer communication,” Journal of neurology, neurosurgery and psychiagtry, vol. 74, pp. 1117 – 1121,
2003.
[5] N. Birbaumer and L. G. Cohen, “Brain-computer interfaces:
communication and restoration of movement in paralysis,”
Journal of physiology, vol. 597, pp. 621 – 636, 2007.
[6] J. J. Daly and J. R. Wolpaw, “Brain computer interfaces in
neurological rehabilitation,” The lancet: neurology, vol. 7, pp.
1032 – 1043, 2008.
[7] T. N. Lal, T. Hinterberger, G. Widman, M. Schroeder, J. Hill,
W. Rosenstiel, C. E. Elger, B. Schoelkopf, and N. Birbaumer,
Advances in neural information processing systems. MIT Press,
2005, ch. 17.
[8] P. R. Kennedy, R. A. E. Bakay, M. M. Moore, K. Adams, and
J. Goldwaithe, “Direct control of a computer from the human
central nervous system,” IEEE Transaction on rehabilitation
engineering, vol. 8, pp. 198 – 202, 2000.
[9] T. H. Swanson, Comprehensive clinical neurophysiology. W.
B. Saunders Company, 2000, ch. Basic cellular and synaptic
mechanisms underlying the electroencephalogram, pp. 354 –
356.
[10] E. Niedermeyer and F. L. da Silva, Electroencephalography,
5th ed. Lippincott Williams and Wilkins, 2004, ch. The normal
EEG of the waking adult, pp. 167 – 170.
[11] C. S. Herrmann, M. H. Munk, and A. K. Engel, “Cognitive functions of gamma-band activity: memory match and utilization,”
Trends in cognitive sciences, vol. 8, pp. 347 – 355, 2004.
[12] R. G. de PeraltaMenendez, Q. Noirhomme, F. Cincotti, D. Mattia, F. Aloise, and S. G. Andino, “Modern electrophysiological
methods for brain-computer interfaces,” Computational intelligence and neuroscience, vol. 2007, pp. 1 – 8, 2007.
[13] S. Halder, M. Bensch, JurgenMellinger, M. Bogdan, A. Kuebler,
N. Birbaumer, and W. Rosenstiel, “Online artifact removal
for brain-computer interfaces using support vector machines
and blind source separation,” Computational intelligence and
neuroscience, vol. 2007, pp. 1 – 10, 2007.
[14] S. Acharya, M. Mollazadeh, K. Murari, and N. Thakor, “Spatiotemporal source tuning filter band for multiclass eeg based
brain computer interfaces,” Proceedings of the IEEE, vol. 28,
pp. 327 – 330, 2006.
[15] N. Nicolaou and S. J. Nasuto, “Robustness of mutual information to inter-subject variability for automatic artefact removal
from eeg,” Proceedings of the IEEE, vol. 27, pp. 5991 – 5994,
2005.
[16] F. Piccione, F. Giorgi, P. Tonin, K. Priftis, S. Giove, S. Silvoni,
G. Palmas, and F. Beverina, “P300-based brain computer interface: Reliability and performance in healthy and paralysed
participants,” Clinical neurophysiology, vol. 117, pp. 531 – 537,
2006.
[17] A. Kübler, B. Kotchoubey, T. Hinterberger, N. Ghanayim,
J. Perelmouter, M. Schauer, C. Fritsch, E. Taub, and N. Birbaumer, “The thought translation device: a neurophysiological
approach to communication in total motor paralysis,” Experimental brain research, vol. 124, pp. 223 – 232, 1999.
[18] D. J. McFarland, W. A. Sarnacki, and J. R. Wolpaw, “Braincomputer interface (bci) operation: optimizing information
transfer rates,” Biological Psychology, vol. 63, pp. 237–251,
2003.
[19] G. Krausz, R. Scherer, G. Korisek, and G. Pfurtscheller, “Critical decision-speed and information transfer in the "graz braincomputer interface",” Applied Psychophysiology and Biofeedback, vol. 28, no. 3, 2003.
[20] A. Kuebler, B. Kotchoubey, J. Kaiser, J. R. Wolpaw, and
N. Birbaumer, “Brain-computer communication: unlocking the
locked in,” Phsychological bulletin, vol. 127, pp. 358 – 375, 2001.
[21] D. Farina, O. F. do Nascimento, M.-F. Lucas, and C. Doncarli,
“Optimization of wavelets for classification of movement-related
cortical potentials generated by variation of force-related parameters,” Journal of neuroscience methods, vol. 162, pp. 357 – 363,
2007.
[22] D. Vautrin, X. Artusi, M.-F. Lucas, and D. Farina, “A novel
criterion of wavelet packet best basis selection for signal classification with application to brain-computer interfaces,” In press,
2009.
[23] B. hua Yang, G. zheng Yan, R. guo Yan, and T. Wu, “Feature
extraction for eeg-based brainŰcomputer interfaces by wavelet
packet best basis decomposition,” Journal of neural engineering,
vol. 3, pp. 251 – 256, 2006.
[24] B. Liu, M. Wang, L. Yu, Z. Liu, and H. Yu, “Study of feature
classification methods in bci based on neural networks,” Proceedings of the IEEE, vol. 27, pp. 2932 – 2935, 2005.
[25] Y. Li and C. Guan, “A semi-supervised svm learning algorithm
for joint feature extraction and classification in brain computer
interfaces,” Proceedings of the IEEE, vol. 28, pp. 2570 – 2573,
2006.
[26] J. Qin, Y. Li, and W. Sun, “A semisupervised support vector
machines algorithm for bci systems,” Computation intelligence
and neuroscience, vol. 2007, no. 1 – 9, 2007.
[27] N. S. Dias, M. Kamrunnahar, P. M. Mendes, S. J. Schiff, and
J. H. Correia, “Comparison of eeg pattern classification methods
for brain computer interfaces,” Proceedings of the IEEE, vol. 29,
pp. 2540 –2543, 2007.
[28] A. Bashashati, M. Fatourechi, R. K. Ward, and G. E. Birch,
“A survey of signal processing algorithms in brainŰcomputer
interfaces based on electrical brain signals,” Journal of neural
engineering, vol. 4, pp. 32 – 57, 2007.
[29] F. Lotte, M. Congedo, A. Lťecuyer, F. Lamarche, and B. Arnaldi, “A review of classification algorithms for eeg-based
brainŰcomputer interfaces,” Journal of neural engineering,
vol. 4, pp. 1 – 13, 207.
[30] N. Birbaumer, T. Elbert, A. G. M. Canavan, and B. Rockstroh,
“Slow potentials of the cerebral cortex and behavior,” The
American physiological society, vol. 70, pp. 1 – 41, 1990.
[31] E. W. Sellers, D. J. Krusienski, D. J. McFarland, T. M.
Vaughan, and J. R. Wolpaw, “A p300 event-related potential
brain-computer interface (bci): The effects of matrix size and
inter stimulus interval on performance,” Biological psychology,
vol. 73, pp. 242 – 252, 2006.
[32] T. Hinterberger, R. Veit, B. Wilhelm, N. Weiskopf, J.-J. Vatine,
and N. Birbaumer, “Neuronal mechanisms underlying control of
a brain-computer interface,” European journal of neuroscience,
vol. 21, pp. 3169 – 3181, 2005.
[33] D. J. McFarland and J. R. Wolpaw, “Sensorimotor rhythmbased brain-computer interface (bci): feature selection by regression improves performance,” IEEE Transactions on neural
systems and rehabilitation engineering, vol. 13, pp. 372 – 379,
2005.
[34] F. Nijboer, E. Sellers, J. Mellinger, M. Jordan, T. Matuz,
A. Furdea, S. Halder, U. Mochty, D. Krusienski, T. Vaughan,
J. Wolpaw, N. Birbaumer, and A. Kübler, “A p300-based
brainŰcomputer interface for people with amyotrophic lateral
sclerosis,” Clinical neurophysiology, vol. 119, pp. 1909 – 1916,
2008.
[35] A. Lenhardt, M. Kaper, and H. J. Ritter, “An adaptive p300based online brain-computer interface,” IEEE transactions on
neural systems and rehabilitation engineering, vol. 16, pp. 121
– 130, 2008.
7
[36] R. Krepki, B. Blankertz, G. Curio, and K.-R. Müller, “The
berlin brain-computer interface (bbci) Ű towards a new communication channel for online control in gaming applications,”
Multimedia tools and applications, vol. 33, pp. 73 – 90, 2007.
[37] J. Kronegg, G. Chanel, S. Voloshynovskiy, and T. Pun, “Eegbased synchronized brain-computer interfaces: a model for optimizing the number of mental tasks,” IEEE transactions on
neural systems and rehabilitation engineering, vol. 15, pp. 50 –
58, 2007.
[38] B. Blankertz, G. Dornhege, C. Schäfer, R. Krepki, J. Kohlmorgen, K.-R. Müller, V. Kunzmann, F. Losch, and G. Curio,
“Boosting bit rates and error detection for the classifcation of
fast-paced motor commands based on single-trial eeg analysis,”
IEEE transactions on neural systems and rehabilitation engineering, vol. 11, pp. 127 – 131, 2003.
[39] P. W. Ferrez and J. del R. Millán, “Error-related eeg potentials
generated during simulated brain-computer interaction,” IEEE
transactions on biomedical engineering, vol. 55, pp. 923 – 929,
2008.
[40] G. Schalk, J. R. Wolpaw, D. J. McFarland, and G. Pfurtscheller,
“Eeg-based communication: presence of an error potential,”
Clinical neurophysiology, vol. 111, pp. 2138 – 2144, 2000.
[41] A. Buttfield, P. W. Ferrez, and J. del R. Millán, “Towards a
robust bci: error potentails and online learning,” IEEE transactions on neural systems and rehabilitation engineering, vol. 14,
pp. 164 – 168, 2006.
[42] B. Blankertz, M. Krauledat, G. Dornhege, J. Williamson,
R. Murray-Smith, and K. R. Müller, “A note on brain actuated
spelling with the berlin brain-computer interface,” Lecture notes
in computer science, vol. 4555, pp. 759–768, 2007.
[43] A. Furdea, S. Halder, D. J. Krusienski, D. Bross, F. Nijboer,
N. Birbaumer, and A. Kübler, “An auditory oddball (p300)
spelling system for brain-computer interfaces,” Psychophysiology, vol. 46, pp. 617 – 625, 2009.
[44] A. Kübler, A. Furdea, S. Halder, E. M. Hammer, F. Nijboer,
and B. Kotchoubey, “A brain-computer interface controlled auditory event-related potential (p300) spelling system for lockedin patients,” Disorders of consciousness, vol. 1157, pp. 90 – 100,
2009.
[45] R. Leeb and G. Pfurtscheller, “Walking through a virtual city
by thought,” Proceedings of the annual international conference
of the IEEE EMBS, vol. 26, pp. 4503 – 4506, 2004.
[46] R. Leeb, R. Scherer, C. Keinrath, C. Guger, and
G. Pfurtscheller, “Exploring virtual environments with an
eeg-based bci through motor imagery,” Biomedical technik,
vol. 52, pp. 86 – 91, 2005.
[47] R. Leeb, F. Lee, C. Keinrath, R. Scherer, H. Bischof, and
G. Pfurtscheller, “Brain-computer communication: motivation,
aim and impact of exploring a virtual appartment,” IEEE
transactions on neural systems and rehabilitation engineering,
vol. 15, pp. 473 – 482, 2007.
[48] R. Leeb, D. Friedman, G. R. Müller-Putz, R. Scherer, M. Slater,
and G. Pfurtscheller, “Self-paced (asynchronous) bci control
of a wheelchair in virtual environments: a case study with a
tetraplegic,” Computational intelligence and neuroscience, vol.
2007, pp. 1 – 8, 2007.
[49] R. Scherer, F. Lee, A. Schloegl, R. Leeb, H. Bischof, and
G. Pfurtscheller, “Toward self-paced brain-computer communication: navigation through virtual worlds,” IEEE transactions
on biomedical engineering, vol. 55, pp. 675 – 682, 2008.
[50] M. N. F. Galan AND, E. Lew, P. Ferrez, G. Vanacker, J. Philips,
and J. del R. Millan, “A brain-actuated wheelchair: asynchronous and non-invasive brainŰcomputer interfaces for continuous control of robots,” Clinical neurophysiology, vol. 119,
pp. 2159 – 2169, 2008.
[51] R. Ron-Angevin and A. Díaz-Estrella, “BrainŰcomputer interface: Changes in performance using virtual reality techniques,”
Neruoscience letter, vol. 449, pp. 123 – 127, 2009.
[52] L. Bianchi, D. Pronestí, M. Abbafati, L. R. Quitadamo, M. G.
Marciani, and G. Saggio, “A new visual feed-back modality for the reduction of artifacts in mu-rhythm based braincomputer interfaces,” Annual international IEEE EMBS conference, vol. 30, pp. 1323 – 1327, 2008.
[53] F. Nijboer, A. Furdea, I. Gunst, J. Mellinger, D. J. McFarland,
N. Birbaumer, and A. Kübler, “An auditory brainŰcomputer
interface (bci),” Journal of neuroscience methods, vol. 167, pp.
43 – 50, 2008.
[54] P. T. Thorbergsson, M. Garwicz, J. Schouenborg, and A. J. Johansson, “Implementation of a telemetry system for neurophysiological signals,” Annual international IEEE EMBS conference,
vol. 30, pp. 1254 –1257, 2008.
[55] G. Schalk, D. J. McFarland, T. Hinterberger, N. Birbaumer,
and J. R.Wolpaw, “Bci2000: a general-purpose brain-computer
interface (bci) system,” IEEE transactions on biomedical engineering, vol. 51, pp. 1034 – 1043, 2004.
[56] G. Pfurtscheller, G. R. Müller-Putz, A. Schlögl, B. Graimann,
R. Scherer, R. Leeb, C. Brunner, C. Keinrath, F. Lee,
G. Townsend, C. Vidaurre, and C. Neuper, “15 years of bci research at graz university of technology: current projects,” IEEE
transactions on neural systems and rehabilitation engineering,
vol. 14, pp. 205 – 210, 2006.
[57] C. Neuper, G. Müller, A. Kübler, N. Birbaumer, and
G. Pfurtscheller, “Clinical application of an eeg-based brainŰcomputer interface: a case study in a patient with severe motor
impairment,” Clinical neurophysiology, vol. 114, pp. 399 – 409,
2003.
[58] N. Neumann and A. Kübler, “Training locked-in patients: a challenge for the use of brain-computer interfaces,” IEEE transactions on neural systems and rehabilitation engineering, vol. 11,
pp. 169 – 173, 2003.
1 Switch Mode Power Supply for Inverter Application
Msc IET Valentin Costea in collaboration with Associate Professor Stig Munk-Nielsen
Institute of Energy Technology, Aalborg University, Denmark
Abstract –The aim of this paper is to design a
switch mode power supply for auxiliary
systems belonging to the inverter such as :
control and monitoring systems. The switch
mode power supply provides multiple
voltage outputs necessary for the control or
other auxiliary systems. This kind of supply is
a step down or a buck converter.
I.
Introduction
The idea with a switch mode power comes
from the necessity to provide output
regulated energy for the auxiliary components
of the inverter. The inverter was considered to
by a medium voltage multilevel one. Each cell
of the inverter has its own power supply
embedded in the control board placed on top
of the power IGBT’s and capacitor banks. The
board is considered to be very important due
to the fact that has to provide insulation
between the medium voltage and the low
voltage area. Also on the board positioned on
top of the power units are place the gate
drivers for the IGBT units. The supply its going
to provide the following outputs[1]:
The energy necessary for the supply comes
from the common DC link of the cells.
One of the goals is to reduce the size of the
supply taking in consideration that the supply
implies a transformer which gives the size of
the supply. In now days, the transformer can
be a small component because of new
materials and new manufacturing procedures.
One of the reasons for using new materials is
also to reduce the losses caused by old
materials. The material used in this situation is
ferrites, taking in consideration that the
transformer is going to be subjected to high
frequency waves.
Basic schematic of the supply
Table 1-1 Outputs of the SMPS
No
1
2
3
4
5
6
7
8
9
Name
Symbol
Value
Minimum
Input
voltage
Output 1Voltage
Output 1Current
Output 2Voltage
Output 2Current
Output 3Voltage
Output 3Current
Output 4Voltage
Output 4Current
ܸ௣
60 VDC
ܸௌଵ
800
VDC
15 VDC
‫ܫ‬଴ଵ
2A
0.25 A
ܸௌଶ
12 VDC
n/a
‫ܫ‬଴ଶ
1A
0.25 A
ܸௌଷ
5 VDC
n/a
‫ܫ‬଴ଷ
5A
0.25 A
Figure 1.1.1 Principle diagram[2]
ܸௌସ
-12
VDC
1A
n/a
Basic steps for design calculation are showed
in [2].
‫ܫ‬଴ସ
n/a
0.25 A
The materials where chosen based steps from
[1]
The chosen core has ETD-34 form and is made
out of ferrite material. The core has the
following characteristics[1]:
In the Figure1.1.2 the primary winding is
represented by W1 coloured in red, the
secondary windings are: W2 coloured in green
which is related to the 15 V, W3 - purple to 12
V, W4 – blue to 5V, W6 – light green to -12V
and W7 which is common with W1. Between
the winding special insulation yellow polyester
tape (UL) was put to ensure a proper
insulation [2].
To ensure that the transformer design was
correct some experiments were conducted for
each one of the out windings.
Figure below show the output results:
Figure 1.1 Dimensions for ETD Ferrites Cores
Table 1-2 Dimensions data for ETD-34
Part
No.
ETD-34
A
[cm]
3.5
B
[cm]
2.56
C
[cm]
3.46
D
[cm]
1.11
E
[cm]
1.11
G
[cm]
2.36
The parameters represent the physical
dimensions of the transformer core. The
magnetic core plays an important role due to
the fact that it stores energy for each
conduction period.
Figure 1.1.3 Measurements for 5V secondary winding
After the design parameters where
determined, the transformer was realised
according to the calculation. The winding
where placed on bobbin on core ETD34 given
by design calculations. The Figure1.1.2 shows
the construction of the transformer including
even the placement of the windings.
Figure 1.1.4 Measurements for 15V winding
Figure 1.1.2 Transformer Construction
II.
Figure 1.1.5 Measurements for 12V winding
The carried tests showed the expected results.
The ratios were achieved considering the
calculations made in [2].
Another important part in the design of the
supply is to choose a switch , usually MOSFET
are used in this type of application. Taking in
consideration the voltage level in the DC Link
the switch has to withstand voltage between
60- 800 V as nominal voltage and a duty cycle
bigger then 50 %. Due to the fact that current
controller was used more exactly the UC3842.
The controller is limited to 50% duty cycle so a
small adjustment was made to ensure that
100 % was achieved. The adjustment consists
in adding a resistor to the current sense route
to provide slope compensation [3].
So the primary power for such a converter is
determined by:
ܲ௢௨௧ ≤ ݂௦௪ ∙
మ
௅೛ೝ೔ ∙ூ೛ೖ
ଶ
(1)
Where:
-
ܲ௢௨௧ output power
‫ܮ‬௣௥௜ primary inductance
݂௦௪ the switching frequency
‫ܫ‬௣௞ peak current
The
necessary
calculations
for
the
components were made in [2] together with
the lists of necessary components.
Experimental Work
In order to provide experimental results a PCB
was made. The PCB was designed using OrCAD
Layout Design. All the important things were
considered like insulation between the layers.
The trace of the PCB where made considering
all the facts like RFI radiation. Like any other
traces these ones also have resistance and
inductance. These influences can lead to high
voltage transitions as consequence of big
variations of the current that passes trough
the PCB traces. In the design it is good to
consider having traces thick and short in order
to minimize the inductive effect and the
resistive. Additional attention was paid to the
layout that is around the capacitive filter. For
example if the capacitors where placed in
parallel within a straight line and placed
nearby the source will get hot due to the
ripple current. There are many aspects to
consider when designing a PCB, most of them
very important in well functionality of the
supply[3].
Figure 1.1.6 Bottom PCB of SMPS[1]
III.
Conclusions
Overall conclusions are deducted after
conducting experimental test on the supply.
Problems have been encountered due to the
PCB
imperfections
occurred
during
manufacturing process. The process has been
adjusted until wanted results where
accomplish [2].
In order to perform the laboratory tests some
minimum conditions have been fulfilled:
Figure 1.1.7 Top PCB of SMPS[1]
RTest were carried out in order to see how the
SMPS will behave so also a load was placed.
The results are displayed in the images below:
-
The minimum voltage necessary for the
supply to work is 60 V(deducted from
calculations). Due to insulation problems
the tests for maximum voltage was not
done [2].
The resistive loads where chosen so the
nominal currents have been reached [2].
2 References
[1]. Wm.T.McLyman, Colonel. Designing
Magnetic Components for High Frequency DCDC Converters. San Marino : Library of
CongressCataloging in publication data, 1993.
1-883107-00-8.
Figure 1.6 Test results for fist secondary output
[2]. Cristian Sandu, Nicoleta Cârnu, Valentin
Costea. Medium Voltage Modular Multi-Level
Inverter. Aalborg : Aalborg University, 2009. p.
320, Master Thesis.
[3]. SWITCHMODE Power Supplies Reference
Manual and Design Guide. Brown, Marty.
1999, p. 138.
Figure 1.7 Test results for the second output
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
1
Experimental Analysis of the Magnetic
Behavior of NdFeB Permanent Magnets
Veronica PANAITE, and Ewen RITCHIE

Abstract—This paper presents a possible method to analyse the
NdFeB permanent magnets. They are powerful magnets, but they
have a low Curie temperature. This parameter influences their
magnetic properties, mainly modifying the magnets’ hysteresis.
This means both the magnetization and the demagnetization
curves can be modified. Two experiments have been performed
on some samples of NdFeB magnets in order to analyze the effect
of applied magnetic field to them.
Index Terms—NdFeB, Neodymium-Iron-Boron, Rare-Earth
Permanent Magnet, Magnetic Properties
I. INTRODUCTION
T
HE neodymium-iron-boron permanent magnet is an
artificial magnet. This material was first discovered in
1983 by General Motors Inc. in the United States and by
Sumitomo Special Metals in Japan. [1, page 1/207]
Neodymium: This is a rare earth lanthanide metal and it is
found in nature as a compound of minerals, like most
lanthanides. [2] It represents an estimated 38 ppm of the
Earth’s crust. [3] It has a silver white color with a slight yellow
tinge. Since it is a reactive rare earth metal, it must be handled
with care: when exposed to air, its surface quickly oxidates,
turns green and peels off, further exposing the remaining metal
to the air; the process repeats itself until all the metal has
oxidated. [2]
Iron: Iron is a metal commonly found in a considerable
quantity on Earth as well as in the rest of the Universe: in the
Sun and various other stars. It corrodes at contact with the
moist in the air, especially at higher temperatures. Iron metal
has a silvery lustrous color with a grey tinge. [2]
Boron: This is a semi-metallic material with properties from
both metallic and non-metallic materials. It is rather a
semiconductor than a metallic conductor. [2] In nature it is
never found as a stand-alone material, only as a compound of
borax [3], tourmaline and kernite. Pure Boron is obtained by
isolating it from the compound material in which it is found.
[2]
II. NDFEB’S MAGNETIC PROPERTIES AND THEIR BEHAVIOR
UNDER HIGH TEMPERATURE
1) Hysteresis magnetic properties:
Magnetization curve:
The initial magnetization curve (also called the virgin curve
[5, page 827]) marks the beginning of the hysteresis process.
Here, the permanent magnet is magnetized to saturation. The
minimum field required for a full saturation value varies
according to the magnet type, alloy composition and
processing parameters, therefore a higher coercivity material
requires a higher saturating field than a lower coercivity one.
Figure 1 illustrates this principle by comparing the behavior of
high and low coercivity of NdFeB alloys. [5, page 589]
Figure 1: Magnetization behavior of high and low coercivity
NdFeB alloys
In figure 2 below there is a comparative presentation of the
form and approximate value of the virgin curves for four of the
most popular types of permanent magnets. As observed, for
NdFeB magnets, we need magnetic fields around 1200kA/m in
order to magnetize them to saturation. [5, page 827]
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
Figure 2: Initial magnetization curves (virgin curves) for
SmCo and NdFeB permanent magnets
Demagnetization curves:
In the image 3 below we have the demagnetization curve for
the neodymium magnet in comparison with some types of
permanent magnets [5, page 827] and in 4 we have second
quadrant demagnetization curves for three types of neodymium
magnet:
 Curve N1 is for sintered NdFeB
 Curve N2 is for high coercivity NdFeB
 And curve N3 is for bonded ‘magnequench’ NdFeB [4,
page 318]
2
Maximum energy product:
When a permanent magnet is implemented in a magnetic
circuit, the magnetic field generated in a gap of that circuit
depends on the permanent magnet’s volume V, magnetic
induction B and coercivity H. So, if we want to have the
minimum volume, we need to maximize the BH product. By
designing the magnet in a certain way, we can obtain the
maximized BH product. [6, page 1172] The permanent
magnet’s maximum energy product (BHmax) can be
approximated using the second quadrant demagnetization
curve by calculating the area described by the curve (figure 5
below):
B
-H
Maximum
Energy
Product
Figure 5: Energy product of a magnet expressed through the
BH curve
Basically a permanent magnet’s performance is defined by
its energy product. Below (figure 6) we have a graph of the
historical evolution of BHmax throughout the years starting with
the 1880’s (the ordinate containing the BHmax values is a
logarithmic scale) [5, page 11]:
Figure 3: Demagnetization curves for permanent magnets
Figure 6: Historical evolution of permanent magnets’ BHmax
values
2) Senzitivity to environmental influence:
Figure 4: Demagnetization curves for three types of NdFeB
magnets
The high temperature effect on the demagnetization curve:
Mostly, the elevated temperature will affect the coercivity
by decreasing it and this decrease further affects the
demagnetization curve. [5, page 641] As a matter of fact, the
temperature dependence of the coercivity is mostly determined
by the temperature dependence of the anisotropy energy. [5,
page 153]
Figure 7 describes the shape of the demagnetization curve
particularly for the NdFeB 35 magnet under three different
temperatures: 20 C
̊ (room temperature), 100 C
̊ and 140 C
̊ :
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
3
been substituted by Dy (didymium), fact which increases the
magnet’s coercivity. [7, page 72]
Figure 7: NdFeB 35 – demagnetization curves at various
temperatures
When returning the magnet to room temperature, the
domain walls do not return to their original positions because
there is no driving force available while the magnet cools.
Therefore a change in bulk magnetization appears. In
conclusion, if a magnet fully magnetized at room temperature
is heated, it will suffer both reversible and irreversible
changes. [5, page 641]
The Curie temperature and material substitution in NdFeBs:
As far as temperature dependence goes for a magnet, there
is a certain temperature above which the magnet is so much
affected, that spontaneous magnetization doesn’t even occur.
This is called the Curie temperature Tc. [7, page 59]
NdFeB magnets have a low Curie temperature, around
300 C
̊ , but it can be raised by partially substituting Fe with Co,
which has a higher Curie temperature. [7, page 59] However,
there are also two disadvantages to consider: introducing Co
will increase the alloy’s cost and reduce the remanence Br. [8,
page 4 (368)]
As temperature increases , there is not only a rapid drop in
the magnetization , but also in the magnet’s coercivity at only
250 C
̊ . Therefore, without partial substitutions of material
(either Nd or Fe), the NdFeB magnets cannot be used much
above 100 C
̊ . For the NdFeB permanent magnets
, the
reversible temperature coefficient is α=-0.15%/ ̊C. [7, page 72]
Another possibility is to add Al to the alloy. This will
increase the coercivity, but unfortunately it will decrease the
saturation magnetization and even the Curie temperature. [9,
page 1 (916)]
Reference [10] describes a successful way to substantially
reduce the reversible temperature coefficient. This was
achieved by introducing a strip along the magnets of 30%-NiFe alloy of 1.25mm thickness.
Figure 8 shows the demagnetization curves for NdFeB
permanent magnets with and without partial substitution of the
initial material. In this case it is the percentage of Nd that has
Figure 8: Demagnetization curves for NdFeB with and without
partial substitution
III. EXPERIMENTS – MAGNETIC RESPONSES FROM NDFEBS
A. The length experiment:
Experimental Setup and Components:
This experiment uses the magnetometer, the Gaussmeter,
two magnets and two pieces of soft magnetic material. In
figure 9 below there is a photo of the entire setup and used
components:
1
2
4
5
3
1. FW Bell 5170 Gaussmeter
2. Measuring probe
3. Magnetometer
4. Nd magnet
5. Soft magnetic material
Figure 9: The length experiment – setup and components
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
Description of the Experiment:
This experiment represents a study of the influence of a
neodymium magnet upon a soft magnetic material. It consists
of two series of magnetic field measurements of the soft
magnetic material. The magnetic field is measured at various
distances from the magnets, ranging from 35mm to 2mm.
The first set of measurements is taken with two neodymium
magnets and one piece of magnetic material, then one magnet
is removed and a piece of soft magnetic material is added; the
measurements are taken again at the same lengths.
Results:
The results of this experiment are the magnetic field values
of the soft magnetic material taken at different distances from
the neodymium magnets. The first set of results was taken by
using two magnets and one piece of soft magnetic material and
the second by using one magnet and two pieces of soft
magnetic material. The measurements are written below:
The measurements were introduced in a XY chart using
Matlab. Each chart represents the magnetic field values
function of the distance between the magnets and the soft
magnetic material. The charts are represented below in figure
10:
Figure 10: The increasing magnetic field values in soft
magnetic material
The plotted magnetic field values show that there is a
considerable increase in the soft magnetic material’s magnetic
field intensity. From the initial values of 21.7G and 31.7G,
respectively, it increases approximately 30 to 40 times,
reaching 868G and 921G, respectively.
B. The paper experiment:
Experimental Setup and Components:
During this experiment I have used a Gaussmeter, two
neodymium magnets and a scaled piece of math paper. I
marked the place where the magnets stood and then measured
and marked various distances from the magnets’ edge. The
distances are between 5 and 55mm and represent the distances
where I have measured the magnetic field. In figure 11 below
there are the setup and components used:
2
4
1
3
1. Measuring probe
2. Nd magnet
3. Marked lengths on a math paper
Figure 11: The paper experiment – setup and components
Description of the Experiment:
This experiment studies the values of the magnetic field
intensity at various distances from the magnet. There are two
measuring sessions: one with one neodymium magnet and the
second with two neodymium magnets placed one on top of the
other.
Results:
The measurements were compiled in a XY plot in Matlab as
the magnetic field intensity function of the distance from the
magnets. The plots can be viewed in figure 12 below:
Figure 12: The decreasing magnetic field values at various
distances from the magnets
APPENDIX: SAFETY HANDLING OF NDFEB PERMANENT
MAGNETS
A. Possible damaging done by NdFeB permanent magnets
The information presented here in this documentation is
provided for information only. [15]
NdFeB permanent magnets are very powerful and when
mishandled or misplaced around people and certain objects
can cause damage. [14]
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
5
People:
When handling NdFeB magnets, fingers and other body
parts can get severely pinched between two attracting magnets.
[12, 14, 15] The larger magnets can easily bruise fingers and
even break finger bones as they attempt to connect together.
[13] One should always wear gloves when handling large
magnets. Very large magnets can pose a crushing hazard and
should not be man-handled. [15]
Eye protection should be worn when handling these magnets
because shattered magnets can launch pieces at great speeds.
[12, 14]
Magnets should never be used to lift objects over people.
[15]
Children should not be allowed to handle NdFeB magnets.
Small magnets pose a choking hazard and should never be
swallowed or inserted into any part of the body, such as the
ear, nose or mouth. If magnets are ingested or aspirated to the
lungs, immediate medical attention is required. NdFeB
permanent magnets are not suitable for children to play with,
and should only be handled under strict adult supervision. [13,
15]
magnets are made to very demanding standards, however uses
should be restricted to operating temperatures below 180°
Fahrenheit (80° Celsius) or they will lose their magnetic
properties. [13]
NdFeB magnets should never be burned [12, 14, 15], as
burning them will create toxic fumes. [12, 14] These magnets
can ignite and burn at high intensity. [15]
Like any tool or toy, NdFeB permanent magnets can be fun
and useful, but must always be treated with care. [12, 14]
Household appliances or personal objects:
The strong magnetic fields of NdFeB magnets can also
damage magnetic media such as floppy disks, credit cards,
magnetic I.D. cards, cassette tapes, video tapes [12, 13, 14], or
any magnetic based storage devices such as desktop or laptop
computers [15], and other CRT displays. Never place these
magnets near electronic appliances. [12, 15] Keep them away
from VCR'S and TV's, non-electronic wrist watches etc. [13]
Never allow them near a person with a pacemaker or similar
medical aid. The strong magnetic fields of the NdFeB magnet
can affect the operation of such devices. [12, 14] Consult your
physician and the manufacturer of your medical device to
determine its susceptibility to static magnetic fields prior to
handling magnets. All of our magnetic products should be kept
at a safe distance from individuals with these devices. [15]
[2]
Working with NdFeB magnets:
Rare-Earth magnets are fragile and can break easily. [12, 13,
15] NdFeB magnets are brittle, and can peel, crack or shatter if
allowed to slam together. [12, 14] Even though they are coated
with tough protective nickel plating, do not allow them to snap
together with their full force or they may break, and possibly
send small pieces of metal flying on impact. [13]
The magnetic fields from the more powerful magnets can
affect each other from more than 12 inches [30.5 cm] away. A
distance of at least 12" should be kept between magnets and
these items at all times. [13] They do not take kindly to
machining. [12, 14]
NdFeB permanent magnets will lose their magnetic
properties if heated above 175° F (80° C). These NdFeB
B. Disposal
Rare-earth magnets should be disposed of in compliance
with local, state, and Federal law. All strong permanent
magnets should be thermally demagnetized prior to disposal.
Alternatively, all strong permanent magnets should be placed
in a steel container prior to disposal so the magnets do not
attract waste disposal equipment or refuse containers. [15]
REFERENCES
[1]
Petrie, Roger – ‘Permanent Magnets in Review’, Electrical Electronics
Insulation Conference and Electrical Manufacturing & Coil Winding
Conference, 1993, Proceedings, Chicago '93 EEIC/ICWA Exposition 47 Oct. 1993 Page(s): 207 – 210
www.webelements.com, Copyright 1993-2007 Mark Winter [The
University of Sheffield and WebElements Ltd, UK].
[3]
www.wikipedia.org, Wikipedia® is a registered trademark of the
Wikimedia Foundation, Inc.
[4]
Jiles, David – ‘Introduction to Magnetism and Magnetic Materials’,
Chapman and Hall, 1994
Commission of the European Communities – ‘Concerted European
Action on Magnets (CEAM)’, Elsevier Applied Science, 1989.
‘Smithells Metals Reference Book 7th Edition’, edited by E. A. Brandes,
G. B. Brook, 1,792pp, 250 figures - 1997, ISBN: 0750636246.
Campbell, Peter – ‘Permanent Magnet Materials and their Application’,
Cambridge University Press, 1994.
Trout, S. R. –‘Material Selection of Permanent Magnets, Considering
Thermal Properties Correctly’, Electrical Insulation Conference and
Electrical Manufacturing & Coil Winding Conference, 2001.
Proceedings, 16-18 Oct. 2001 Page(s):365 – 370, Digital Object
Identifier 10.1109/EEIC.2001.965683
Baa-Min Ma; Narasimhan, K. –‘NdFeB Magnets with Higher Curie
Temperature’, Magnetics, IEEE Transactions on Volume 22, Issue 5,
Sep 1986 Page(s):916 – 918
Kim, S. H. and Dooise, C. –‘Temperature Compensation of Ndfeb
Permanent Magnets’, Particle Accelerator Conference, 1997.
Proceedings of the 1997, Volume 3, 12-16 May 1997 Page(s):3227 3229 vol.3 Digital Object Identifier 10.1109/PAC.1997.753163
Panaite, Veronica – ‘General Study upon the NdFeB Permanent
Magnet’, Aalborg University, Aalborg, 2007
http://www.powermagnetstore.com/acatalog/Safety_Info.html
http://www.unitednuclear.com/magnets.htm, All rights reserved. ©
1998-2008 United Nuclear Scientific Supplies, LLC.
http://www.kjmagnetics.com/safety.asp, K&J Magnetics, Inc.
http://www.rare-earth-magnets.com/Policies/magnets_safety.htm,
©
2002 – 2008 – National Imports, LLC. All Rights Reserved.
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
1
Types of NdFeB Permanent Magnets and
Application Possibilities
Veronica PANAITE, and Ewen RITCHIE
Institute of Energy Technology, Aalborg University, Pontoppidanstræde 101
Abstract—This paper presents the main types of neodymiumiron-boron (NdFeB) permanent magnets and their application.
They are strong magnets, but susceptible to corrosion. Therefore
protective coatings are important for NdFeB magnets. Since they
are powerful magnets, they are sought to be implemented in
magnetic related applications. The NdFeB magnets can be used in
all magnet-related fields, as table 4 shows.
Boron: This is a semi-metallic material with properties from
both metallic and non-metallic materials. It is rather a
semiconductor than a metallic conductor. [2] In nature it is
never found as a stand-alone material, only as a compound of
borax [3], tourmaline and kernite. Pure Boron is obtained by
isolating it from the compound material in which it is found.
[2]
1) Magnets grades relative to NdFeB magnets
Index Terms—NdFeB, Neodymium-Iron-Boron,
Magnet, Bonded Magnet, Applied Permanent Magnet
Sinterred
I. INTRODUCTION
HE neodymium-iron-boron permanent magnet (Figure 1)
is an artificial magnet. This material was first discovered
in 1983 by General Motors Inc. in the United States and
by Sumitomo Special Metals in Japan. [1]
T
Figure 1: Neodymium, Iron, and Boron
Neodymium: This is a rare earth lanthanide metal and it is
found in nature as a compound of minerals, like most
lanthanides. [2] It represents an estimated 38 ppm of the
Earth’s crust. [3] It has a silver white color with a slight yellow
tinge. Since it is a reactive rare earth metal, it must be handled
with care: when exposed to air, its surface quickly oxidates,
turns green and peels off, further exposing the remaining metal
to the air; the process repeats itself until all the metal has
oxidated. [2]
Iron: Iron is a metal commonly found in a considerable
quantity on Earth as well as in the rest of the Universe: in the
Sun and various other stars. It corrodes at contact with the
moist in the air, especially at higher temperatures. Iron metal
has a silvery lustrous color with a grey tinge. [2]
A magnet grade is defined by the maximum energy product
of the magnetic material in it. The energy product is measured
in Gauss Oersted (GOe) units, but usually it is expressed in
mega Gauss Oersted (MGOe); 1MGOe=1,000,000GOe. In a
simpler manner, the magnet’s grade defines how ‘strong’ that
permanent magnet is. For example: a Neodymium permanent
magnet with the grade forty, will be expressed with the symbol
N40, and thus has a maximum energy product of 40 MGOe.
[4]
As most manufacturing issues, a magnet’s grade depends
upon the manufacturers and distributors. Throughout the world
there are many NdFeB magnet suppliers that have different
magnet grade symbols as well as maximum energy product
values. Neodymium magnets from Shin/Etsu and Neomax
(former Sumitomo Special Metals) have magnet grades up to
52MGOe. [5]
A permanent magnet’s performance is defined by its energy
product. Below (figure 2) there is a graph of the historical
evolution of BHmax throughout the years starting with the
1880’s (the ordinate containing the BHmax values is a
logarithmic scale) [6]:
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
Type of bonded magnets
Compression molding
process
Hot compression molding
process
Injection molding process
Figure 2: Historical evolution of permanent magnets’ BHmax
values
Extrusion process
II. TYPES OF NDFEB PERMANENT MAGNETS
Bonded anisotropic
magnets
There are mainly four types of NdFeB permanent magnets:
powdered, sinterred, bonded and nanocomposite. Each is
briefly described in this section through the susceptibility to
corrosion (the main reason magnets are worn out) and
protective coatings.
2
Coatings
Epoxies
Thermoplastic materials
Epoxies
Thermoplastic materials:
polyprophenylene sulfide
(PPS); nylon
Polymer additives:
antioxidants, plasticizers,
lubricants, etc.
Nylon 12
Lubricants
Antioxidants
Polymers:
polyprophenylene sulfide
(PPS); nylon
Table 1: Typical coatings and additives for powdered bonded
magnets
2) Sinterred NdFeB Magnets:
1) Powdered NdFeB Magnets:
Corrosion:
Rare earth (especially neodymium based) permanent
magnets can be manufactured through a powder processing
route. The oxidation experiments performed on NdFeB
powders show that the oxidation rate is dependent of the
powder size. Basically fine powders were shown to oxidize
more rapidly than coarse powders. [6]
Coatings:
As far as bonded magnets go, their particles are usually
coated with thermosetting epoxies. The epoxies have a series
of properties such as adhesion, impact strength, hardness,
thermal shock, abrasion and chemical resistance, and
penetration. These help protect the magnets’ particles and have
the following advantages:
 After coating, the magnetic particles are ready for
pressing;
 Their powder flow characteristics are improved;
 The uniformity of the molded parts is also improved;
 By using coated particles, the production rate is
increased;
 With relatively low pressure, higher density is achieved;
 A higher solid loading level can be achieved;
 Coated particles will have an improved protection against
corrosion. [10]
The following table 1 represents a short list with typical
coatings and additives for various types of bonded NdFeB
magnets [10]:
Corrosion:
All rare earth types of permanent magnets are prone
especially to oxidation, process which permanently alters their
metallurgical structure. The changes depend entirely of the
exact composition and process method, but it is more severe in
sintered NdFeB magnets. These have to be entirely
encapsulated in order to protect them from corrosion. [8]
Opposed to these, the ceramic, alnico and SmCo types of
permanent magnets are not sensitive to corrosion and are
usually coated for mechanical reasons or just for a different
aspect. [11]
The sintered NdFeB magnets have many types of
composition and hence different electrochemical potential,
[12] but generally speaking, the sintered or heat treated
magnets are very susceptible to corrosion if not thoroughly
dried after machining or slicing operations. Under controlled
test conditions surface corrosion is visible within hours. [6]
They have a corrosion rate of 9. [12]
In order to improve corrosion resistance, the sintered
NdFeB magnets may be alloyed with Co, V, Cr or Ni, but this
can affect their magnetic properties. [13]
Tests performed on the Sintered Hitachi HX94EA revealed
that the demagnetization curves were considerably lower in a
corroded magnet. Figure 3 below (taken from [6]) shows the
effect of corrosion particularly on the magnet’s
demagnetization. This effect could be critical when discussing
thin segments of magnet. [6]
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
3
Corrosion:
Polymer bonded magnets are mostly used at elevated
temperatures and severe environmental conditions. [10] In
reference [6] tests were performed on polymer bonded (MQ1)
and sintered Hitachi magnets and they revealed the fact that
the bonded magnets are more resistant to corrosion than the
sintered ones. Figure 4 below shows the comparison between
BHmax curves of bonded and sintered magnets after being
tested in both dry and humid conditions: [6]
Figure 3: Effect of the corrosion in humid conditions on the
magnets’ demagnetization
Coatings:
Because the sintered NdFeB magnets are very susceptible to
corrosion, various coatings have been tested throughout the
years to try and provide the best possible resistance to
corrosion. In reference [6], NdFeB sintered magnets were
coated with several polymeric coatings in order to reveal how
much their corrosion resistance is improved.
Based on the performed tests, uncoated sintered NdFeB
magnets resist corrosion less than 40 minutes while the ones
coated with epoxies and nylon can last up to 123 hours. The
complete test results are revealed in table 2 below:
Table 2: Corrosion test results for plastic coated NdFeB
magnets
The corrosion tests were performed by using a solution with
5% salt spray at the temperature of 35 C
̊ .
3) Bonded NdFeB Magnets:
Permanent magnet materials can be mixed with a bond
while in powder form. These bonds can be either flexible or
rigid. Rubber bonded magnets are widely used in display and
holding services, for example. Rare-earth magnets made in
bonding forms have the advantage of cracking much harder
than unbonded magnets but the disadvantage is that the
magnetic properties are not as good as those of unbonded
magnets. [14]
There is a variety of binders that can be used which includes
thermoplastic and thermosetting resins and elastomers. [10]
Figure 4: BHmax curves of bonded and sintered NdFeB
magnets tested in dry and humid conditions
Coatings:
After coating, the particles can be compression molded into
different shapes. Then they are cured in an oven, cleaned and
sent to the surface coating process. This process can be done
by spray coating or by electrostatic coating (e-coating). The
latter is ideal for flat, angular or irregular shaped magnets. Ecoating has the following advantages:
 Very good finish quality;
 High productivity and cost effectiveness;
 Can be used to coat parts with complex geometries;
 Has a tight tolerance. [10]
In the tables 3 below there is a description of the two
surface coating processes [10]:
Process characteristics of e-coat:
Immersion
Application
method
electrodeposition
epoxy/urethane water
based
Alkaline clean
Pre-treat
process options
Acid etch/passivate
15-25 µm
Thickness
2H-4H
Pencil hardness
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
Spray coating characteristics:
Epoxy, phenolic,
Material
and acrylic
20 µm nominal
Thickness
Good
Acid resistance
Up to 5H
Pencil hardness
-73-204 C
̊
Operating
temperature
Tables 3: Surface coating processes
4) Nanocomposite NdFeB Magnets:
Corrosion and coatings:
The nanocompozite neodymium magnets lack in rare-earth
rich phases; this improves the magnets’ corrosion behavior.
The ones containing also a small quantity of zirconium (Zr)
have significantly smaller weight increase under humid
conditions than the nanocompozite neodymium magnets
without Zr. Basically, the more the rare-earth content of a
magnet, the faster it oxidizes.
Figure 5 shows the weight change per surface unit area of
nanocompozite neodymium magnets and conventional NdFeB
magnets [10]:
4
1) Basic criteria for selecting the right permanent magnet
for the right application
When selecting a certain type of permanent magnet for a
certain type of application, there are some key criteria of
selection:
 The temperature at which the application will be running
 The levels of acids, salts, hydrogen etc. In the
environment
 Device dimensions: size, weight, volume
 Magnet’s sizes and shapes
 The magnet’s integration in the desired device:
 How it will be attached – the method
 Protective coatings/encapsulations – if they are
necessary (and they are in case of NdFeB magnets)
and what type of coating is needed
 Last, but not least – the total cost of the permanent
magnet usage in the selected device [16]
2) Examples of applications with NdFeB permanent
magnets
So far, the most important application of NdFeB permanent
magnets is in DC rotating electric machines and they have
been employed also to promote the evolution of electronically
commutated brushless DC motors. These magnets have been
used particularly for brushless DC spindle motors which are
employed in computer hard disk drives. [8] Figure 6 below
depicts three stator layouts for a DC motor as well as the
improvements in reducing the rotor’s diameter by applying
more powerful permanent magnets: [8]
Figure 6: Permanent magnet stator layouts – DC motor
Figure 5: Weight change per unit surface area (dW) of
compression molded resin bonded magnets
III. APPLICATIONS OF NDFEB PERMANENT MAGNETS
When referring to the implementation of permanent
magnets, it should be stated that they are used in devices
mainly for one reason: in order to increase magnetic flux. This
is because the magnetic flux is necessary for one or more of
the following purposes in that device:
 for detections – like in sensors, for example
 to contribute at creating torque or a force – like in
motors, speakers
 to generate voltage – like in generators [15]
Another major category of NdFeB application is stepper
motors (or simply steppers) which can be controlled digitally
since they rotate in a sequence of discrete steps. A less popular
use for the NdFeB magnets is in synchronous machines,
where the permanent magnet (with its field) and the multiphase stator winding rotate in synchronism.
There is a wide variety of electromechanical devices (they
convert electrical energy in mechanical energy) in which the
NdFeB magnet is used, such as actuators (for example the
hammer mechanism used in printers), [8] and the moving
coil actuators, such as audio speaker systems. More
important, the automobile speaker systems have been
improved over the last years, since the use of NdFeB allowed
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
miniaturization of the headphones and capstan drive motor. [8]
Last but not least, NdFeB permanent magnets have been
employed in electrically controlled functions for automobiles,
such as blowers, starters and radiator cooling fans. [8]
Table 4 below reveals some of the most common
applications which, in the late 1980’s, were containing NdFeB
permanent magnets: [17]
Application group
Acoustic transducers
Permanent magnet
motors and generators,
electromechanical
transducers
Magnetomechanics
Magnetic field and
focusing systems
Examples
Loudspeakers
Headphones
Telephone receivers
Microphones
DC motors
EC motors
Synchronous motors
Stepping motors
Generators
Linear motors
Moving-coil motors
Torque transmitters
Couplings
Separators
Attachment systems
Magnetic conveyors
Bearings
Transport systems
Dipoles
Quadrupoles
Hexapoles
Undulators
Wigglers
Circulators
Mass spectrometers
NMR systems
Table 4: Examples of NdFeB permanent magnet applications
3) Advantages and disadvantages of use for NdFeBs
Advantages:
The NdFeB permanent magnet has the following advantages
of use:
 It is the magnet with the highest energy
 It contains raw materials with a low cost
 There is a large variety of processing methods for them
 Forged, sintered, bonded, die pressed
 Due to a straight line demagnetization curve, parts can be
magnetized before assembly without the need for
keepers and return paths [1]
Disadvantages:
5
The NdFeB permanent magnet has the following
disadvantages when used:
 They need specific design strategies in order to take
advantage of properties, as the direct substitution of
earlier materials is not advised
 Low Curie temperature implies low maximum use
temperature
 They need special coatings and handling due to their high
susceptibility to corrosion
 The magnet’s alloy, when in powder form, is pyrophoric;
this means it ignites when exposed to air
 When milled in chlorinated hydrocarbons (Freon) the
material may explode; special care is therefore needed
 When magnetized, the NdFeB needs very high
magnetizing fields in order to reach its saturation
 It requires a multiple step process when manufactured; it
is difficult to control and automate and a large capital
investment is needed [1]
 When implemented in permanent magnet machines, they
can be accidentally demagnetized by fault currents, like
a short-circuit current, that may be produced by inverter
faults [18]
4) Process advantages and disadvantages
While most advantages & disadvantages refer more or less
to the NdFeB permanent magnet’s performance (coercivity
and remanence) and behavior (for example in a certain degree
of humidity or temperature), there are considerations to mind
regarding the manufacturing type and geometry.
NdFeB permanent magnets – high energy:
 It is an abundant resource of materials
 Inexpensive refining costs
 Well-established manufacturing technology
 Highest output so far among all permanent magnet
materials on the market
 Compromise in energy product – due to high temperature
applications
 Protective coatings required because of the accentuated
tendency to corrosion [16]
Sintered or fully dense, anisotropic NdFeB magnets –
maximum output:
 Maximum energy product for their size and weight
 Limited by process to simple forms of geometry
 They need to be handled with care; they are brittle
The expression ‘fully dense’ refers to the fact that there is
no dilution caused by non-magnetic materials. [16]
Injection molded NdFeB magnets – shape flexibility:
 They have complex geometries
 No finishing operations and small geometric tolerance –
because of the strict dimensions in molding forms
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
 Relatively resistant to chipping
 Assembly costs reduced by insert and over molding
 Many pole configurations available
 Multistep and multicomponent molding for assemblies
production
 Lower energy product by dilution of magnetic phase with
non-magnetic materials – they suffer the biggest
dilution effect
 Isotropic and anisotropic powders provide a large
configuration of magnetic alignments and output
options – this makes them more desirable despite the
dilution effect
 High tooling costs – these magnets are well-suited for a
large volume manufacturing [16]
Compression bonded NdFeB magnets – low cost
manufacturing:
 Compromise in the obtained energy product – not fully
dense magnets, but higher loading than injection
molding
 Limited to simple geometry forms – cylinders, arcs,
rectangles
 Tight geometric tolerance – exception: pressed thickness
 To be handled with care – brittle
 Isotropic powder – many magnetizing patterns
 The greatest advantage of these magnets – the thin wall
cylinder/ring magnets which can be manufactured using
compression bonding
The thin wall cylinder/ring magnets cannot be manufactured
by sintering process because of breakage during grinding. [16]
Heavy
metal
Cu
Zn
Ni
Cr
Pb
Cd
Fe
Mn
B
Total trace in comparison with the contents
of US, Canadian and world soils (mg/kg)
11
7
7
21
16
0.24
1,236
315
32
Table 5: Total trace elements of heavy metals in Denmark’s
soil
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
APPENDIX
So far the environment’s influence on the NdFeB type of
magnets has been discussed. But there is the reverse to
consider as well. Magnets are eventually worn out or simply
replaced. From here on they may be reused in other
applications or worse, be thrown away and become waste, thus
affecting soil and water.
As seen above, NdFeB magnets contain other elements
besides Nd, Fe, and B either in alloys or as coatings. For
instance, Ni is used for coating NdFeB magnets but this metal
is also part of the heavy metals (HMs) which are generally
toxic to animals and plants when becoming waste. [1] It is
used in plating as a replacement of Cd which was banned due
to its very high levels of toxicity. [20]
Ferrous metals represent a net waste disposal of 11 millions
of tons annually, out of which 8.2% represent the final quantity
of waste after recycling. [21] Particularly in Denmark, heavy
metals are found in the soil and their levels of encounter are
represented in the table 3.4 below, adapted from [22]:
6
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
Petrie, Roger – ‘Permanent Magnets in Review’, Electrical Electronics
Insulation Conference and Electrical Manufacturing & Coil Winding
Conference, 1993, Proceedings, Chicago '93 EEIC/ICWA Exposition 47 Oct. 1993 Page(s): 207 – 210
www.webelements.com, Copyright 1993-2007 Mark Winter [The
University of Sheffield and WebElements Ltd, UK].
www.wikipedia.org, Wikipedia® is a registered trademark of the
Wikimedia Foundation, Inc.
http://rare-earth-magnets.com/magnets.htm, © 2002-2007 – National
Imports, LLC. All rights reserved
http://www-ssrl.slac.stanford.edu/lcls/undulator/meetings/2004-1115_llp_review/, Moog, Liz – ‘Magnet Material Choice’ pages 5 and 6,
A.U.S. Department of Energy, Office of Science Laboratory, Operated
by the University of Chicago
Commission of the European Communities – ‘Concerted European
Action on Magnets (CEAM)’, Elsevier Applied Science, 1989
Liu, Yi; Selmyer, D.J.; Shindo, Daisuke – ‘Handbook of Advanced
Magnetic Materials – Volume 1: Nanostructural Effects’, Springer,
2006
Campbell, Peter – ‘Permanent Magnet Materials and their Application’,
Cambridge University Press, 1994
http://rareearth.org/magnets_patents_history.htm, Contents © 20022007 – The Rare-Earth Magnetics AssociationSM. All rights reserved.
Liu, Yi; Selmyer, D.J.; Shindo, Daisuke – ‘Handbook of Advanced
Magnetic Materials – Volume 3: Fabrication and Processing’, Springer,
2006
Magcraft®, Advanced Magnetic Materials – ‘Permanent Magnet
Selection and Design Handbook’, Copyright © 2007 National Imports
LLC, All rights reserved.
Ward, M. And Taylor, J.S. – ‘Magnetic, Mechanical and Environmental
Properties of NdFeB Magnets – A User’s View’, New Permanent
Magnet Materials and their Applications, IEE Coloquium on 9 Jan 1989
Page(s): 5/1 – 5/5
Arenas, M.; Warren, G.W.; Li, C.P.; Dennis, K.W.; McCallum, R.W. –
‘Corrosion and Hydrogen Absorption in Melt Spun NdFeB-Tic Bonded
Magnets’, Volume 33, Issue 5, Part 2, Sept. 1997 Page(s): 3901 – 3903
‘Smithells Metals Reference Book 7th Edition’, edited by E.A. Brandes,
G.B. Brook, 1,792pp, 250 figures – 1997, ISBN: 0750636246
Trout, S.R. – ‘Understanding Permanent Magnet Materials; an Attempt
at
Universal
Magnetic
Literacy’,
http://www.arnoldmagnetics.com/mtc/pdf/strout_univ_mag_literacy.pdf
Constantinides, Steve – ‘Magnet Selection’, Arnold Magnetics Inc.,
October
15-17,
2003,
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
[17]
[18]
[19]
[20]
[21]
[22]
http://www.arnoldmagnetics.com/mtc/pdf/steveC_Gorham_030923_Co
mmented.pdf
Danske Ingeniøres Efteruddannelse – Permanente Magneter, Extract
from ‘Technische Mittelungen Krupp’ 1/1988
Miller, Tim – ‘Brushless Permanent-Magnet Motor Drives’, Power
Engineering Journal, Volume 2, Issue 1, Jan. 1998 Page(s): 55-60
http://www.cirad.fr/en/actualite/communique.php?id=373, ‘A study of
Soil Pollution by Metals in Réunion’, Centre de coopération
internationale en recherche agronomique pour le développement ©
2003-2007 Cirad. Tous droites réservés
European Commission DG ENV. E3 Project ENV.E.3/ETU/2000/0058,
- ‘Heavy Metals in Waste – Final Report’, February 2002, COWI A/S,
Denmark
Timy Katyal and Prof. M. Satake – ‘Environmental Pollution’, Anmol
Publications Pvt. Ltd., New Delhi 110002 (India), Reprint 2001
S.V.S. Rana – ‘Environmental Pollution – Health and Toxicology’,
Alpha Science International Ltd., Oxford, UK, 2006
7