Professional Documents
Culture Documents
M.Sc. THESIS
FEBRUARY 2021
ISTANBUL TECHNICAL UNIVERSITY GRADUATE SCHOOL OF SCIENCE
ENGINEERING AND TECHNOLOGY
M.Sc. THESIS
FEBRUARY 2021
İSTANBUL TEKNİK ÜNİVERSİTESİ FEN BİLİMLERİ ENSTİTÜSÜ
ŞUBAT 2021
Abdulkerim Ahmet Kaplan, a M.Sc. student of İTU Graduate School student ID
504181002, successfully defended the thesis entitled “DETERMINATION OF
COMMUTATION SECTOR IN SENSORLESS CONTROL OF BLDC MOTORS
BY MAXIMUM LIKELIHOOD ESTIMATION METHOD”, which he prepared after
fulfilling the requirements specified in the associated legislations, before the jury
whose signatures are below.
v
To my father,
vi
vii
FOREWORD
I thank to my father Fahrettin Kaplan, who passed away two years ago, my mother
Belgin Kaplan, and my elder sister Büşra Kaplan for supporting and valuing me in
every way since my childhood and for always loving me.
This thesis was prepared with both his guidance and moral support. I want to thank to
Asst. Prof. Dr. Derya Ahmet Kocabaş, who is my consultant encouraging me to
succeed in any organization and competition.
In addition, I would like to thank everyone who helped to prepare this thesis at the
institution I work for.
Sincerely yours...
viii
ix
TABLE OF CONTENTS
Page
APPENDIX A ....................................................................................................... 58
APPENDIX B ........................................................................................................ 60
APPENDIX C ......................................................................................................... 62
x
xi
ABBREVIATIONS
AC : Alternaative Current
ADC : Analog Digital Converter
ANN : Artificial Neural Network
BLDC : Brushless DC
BACK-EMF : Back Electromotive Force
CMRR : Common Mode Rejection Ratio
DC : Direct Current
DSP : Digital Signal Processor.
IGBT : Insulated Gate Bipolar Transistor
LSE : Least Square Estimation
MLE : Maximum Likelihood Estimation
MOSFET : Metal Oxide Field Effect Transistor
NextFET : Next Generation Field Effect Transistor
NMOS : N Type MOSFET
OPAMP : Operational Amplifier
PC : Personal Computer
PDF : Probability Density Function
PMOS : P Type MOSFET
ZCPD : Zero Crossing Points Detection
FDAM : Fırçasız Doğru Akım Motoru
xii
xiii
SYMBOLS
C : Capacitance
𝑽 : Applied Voltage
𝒆(𝒕) : Back-EMF Signal
m : Slope of line
t : Time
R : Resistance
L : Inductance
B : Viscous Friction Coefficient
J : Inertia
𝝋𝒔 : Stator Flux Linkage
𝝋𝒇 : Field Flux Linkage
𝝋𝒂 : Armature Flux Linkage
i,I : Current
𝒆𝒂 ,𝒆𝒃 ,𝒆𝒄 : Phase Back-EMF Voltages
S : Surface
T : Torque
𝑻𝒆 : Electrical Torque
𝑻𝒎 : Mechanical Load Torque
P : Power
𝝎 : Angular Velocity
H : Clarke Transformation Matrix
𝒆𝛂 , 𝒆𝛃 : Back-EMF α- β Axis Value
P(x) : Probability Distribution Function
𝝁 : Mean
𝝈𝟐 : Variance
L(x) : Likelihoood Function
l(x) : Log-Likelihood Function
c : Class
𝝉(𝒔) : Mechanical Time Constant
xiv
𝑭(𝜶) : LSE Function
𝜶 : LSE’s Estimate Parameter
𝑹𝑫𝑺 : Drain Source Resistance of MOSFET
xv
LIST OF TABLES
Page
xvi
xvii
LIST OF FIGURES
Page
xviii
Page
xix
DETERMINATION OF COMMUTATION SECTOR IN SENSORLESS
CONTROL OF BLDC MOTORS BY MAXIMUM LIKELIHOOD
ESTIMATION METHOD
SUMMARY
Brushless motors are indispensable in today's technology due to their high power
density and efficiency. Brushless motors have many different applicational areas in
recent years, from electric vehicle production to space applications. As an example,
electrical vehicles use hub motors which are a type of brushless motor that is placed
inside the wheel for traction.
A confusing subject for many researchers in the literature can be stated as the
classification and naming of the brushless motors. These types of motors are separated
into two according to their air-gap flux shape, also to Back-EMF waveforms which
are trapezoidal and sinusoidal. That’s why they are named as brushless direct current
motors or permanent magnet synchronous motors respectively. The reason
determining the shape of Back-EMF waves is not due to the structure of the motor but
to the driving style of the motor. General structures of both permanent magnet
synchronous motors and brushless direct current motors are the same, nevertheless
only driving methods are different.
Brushless motors consist of rotating and non-rotating structural parts called stator and
rotor, respectively. The stators of these motors accommodate classical 3-phase
electrical windings. These windings are fed by an inverter to generate a variable
magnetic field in the motor air-gap. The rotor of the Brushless motors is made of
magnetic materials and permanent magnets that are placed on the rotor. Usage of the
permanent magnets on the rotor side makes these types of motors more efficient than
its counterpart, because there is no rotor copper loss on the rotor side of the machine.
Although their efficiency and power density are higher than that of their equivalents,
the overall system cost of those types of motors' is expensive. The reason for the
problem can be stated as these types of motors cannot be driven directly. Such motors
require complex power electronics circuits that are controlled by a microcontroller,
and expensive sensors for proper control. These sensors increase the cost of the motor
and the system as well as they occupy extra places on the motor. Accordingly, this can
cause the power density of the motor to decrease. Related to this, sensorless control
methods have become inevitable for these types of motors.
Undoubtedly, the most important issue in the sensorless control of brushless motors is
the correct detection of the rotor position. When the position of the rotor is not detected
correctly, it may cause the rotor to be locked when the motor is loaded. When this
happens, very high currents flow instantly over the windings. In this case, the motor
or system may become unusable.
Brushless direct current motors can be stated as broadly used brushless motor type in
the market. Because these types of motor can be controlled with 3 hall sensors which
are the cheapest sensor in the market. According to the information received from these
xx
hall sensors, the commutation of the motor can be realized. Controlling a brushless
motor in this way causes to have a trapezoidal Back-EMF waveform on the stator
windings. Due to that, this kind of control method is called a trapezoidal driving
method. When the literature is deeply examined, most of the sensorless control
methods try to estimate the same information to be received from the hall sensors
values from brushless motor's state variables.
As understood from the examined literature, there are lots of sensorless trapezoidal
driving methods that were proposed and proven in the literature. However, among all
of the methods, one method is more popular than others. The name of this method is
the Back-EMF zero-crossing point detection method. This method differs from others
so that it can be implemented with very simple electronic circuits.
The reason why this method is more popular than the others is that the Back-EMF
zero-crossing method can be also performed without software. In this method, it is
possible to obtain the sensor signals also by comparing each phase voltage with direct
current bus voltage or motor neutral point in comparators.
The Back-EMF zero-crossing point detection method is used today in many places,
from electronic speed control circuits that control the speed of the propellers of
unmanned aerial vehicles to those of spacecraft. Besides its advantages, naturally, it
also has some disadvantages. The most important of these disadvantages is that this
method is sensitive to noise. In addition, this method also cannot be used in wide speed
ranges.
For these reasons, a new sensorless trapezoidal control method was proposed in this
study that can eliminate these disadvantages. The name of this method is the maximum
probability estimation method based motor sector detection method.
The rotor position detection algorithm with the maximum likelihood estimation
method is a statistical-based algorithm. This algorithm helps to obtain the probability
map of the motor by statistically evaluating the data that previously taken from the
motor state variables. In order to obtain this map, firstly, the data received are
classified according to sectors. Then, sector probability distribution functions are
calculated by obtaining the mean and standard deviation values of each sector. Back-
EMF's Clarke transformation operation outputs are used as input to these probability
distribution functions. Then, the sector of the function with the highest probability
value can be determined as the sector in which the rotor is actually located.
The motor sector detection algorithm with the maximum probability estimation
method can be considered as an improvement to Back-EMF zero-crossing point
detection method. Because this method is based on the classification of motor Back-
EMF waveforms by obtaining Clarke transformations. Its general algorithm is similar
to the Back-EMF zero-crossing point detection algorithm.
In this study, driving procedures were carried out by an inverter which is located on
the DRV8301 booster pack card of Texas Instruments. C2000 F28027F launchpad
board was used to control this board which controls the high power side of the
experimental setup. The Back-EMF zero-crossing point detection algorithm and
maximum likelihood estimation based sensorless control of BLDC motor application,
implementations were made by preparing an experimental setup. The data taken from
the experimental setup has been analyzed through Matlab and Simulink to improve the
newly proposed method. In addition, theoretical models of these methods were given.
Later, these methods were compared and their advantages and disadvantages were
xxi
revealed. The mathematical background of the proposed new method, which does not
exist in the literature, was presented.
Examining of the compared results showed that BLDC motor zero-crossing point
detection based control method is sensitive to noise. Abrupt changes in the Back-EMF
signals due to noise lead to the problem of the early commutation process in this
method. The newly proposed maximum likelihood estimation based commutation
algorithm aims to solve this problem.
Consequently, a statistical probability method was merged with the Back-EMF zero-
crossing detection control method of BLDC motors. It was shown that the newly
proposed method is reliable at all speed ranges although the up-to-date Back-EMF
zero-crossing detection methods have drawbacks in commutation control over wide
speed ranges.
xxii
FIRÇASIZ DOĞRU AKIM MOTORLARININ
ALGILAYICISIZ KONTROLÜNDE KOMÜTASYON BÖLGESİNİN
EN BÜYÜK OLASILIK TAHMİNİ YÖNTEMİ İLE TESPİTİ
ÖZET
xxiii
yöntemlerinin çoğu bu hall algılayıcı değerlerini fırçasız motorun durum
değişkenlerinden tahmin etmeye çalışır.
Literatürde önerilen ve kanıtlanmış çok sayıda algılayıcısız trapezoidal sürüş yöntemi
vardır. Bununla birlikte, tüm yöntemler arasında bir yöntem diğerlerinden daha
popülerdir. Bu yöntemin adı, ters-emk sıfır geçiş noktası algılama yöntemidir. Bu
yöntem, bu yöntemin çok basit elektronik devrelerle uygulanabilmesi açısından
diğerlerinden farklıdır.
Bu yöntemin diğerlerine göre daha popüler olmasının nedeni ters-emk sıfır geçiş
yönteminin yazılım olmadan da gerçekleştirilebilmesidir. Bu yöntemde, her faz
gerilimini karşılaştırıcılarda doğru akım bara gerilimi veya motor nötr noktası ile
karşılaştırarak da sensör sinyallerini elde etmek mümkündür.
Ters-emk sıfır geçiş noktası algılama yöntemi, insansız hava araçlarının pervanelerinin
hızını kontrol eden elektronik hız kontrol devrelerinden uzay araçlarına kadar birçok
yerde günümüzde kullanılmaktadır. Bu yöntemin avantajlarının yanı sıra doğal olarak
birtakım dezavantajları da mevcuttur. Bu dezavantajlardan en önemlisi, bu yöntemin
gürültüye duyarlı olmasıdır. Ayrıca bu yöntem geniş hız aralıklarında da
kullanılamaz.
Bu nedenlerle, bu dezavantajları ortadan kaldırabilecek yeni bir sensörsüz trapezoidal
kontrol yöntemi bu çalışmada önerilmiştir. Bu yöntemin adı, maksimum olasılık
tahmin yöntemine dayalı motor sektörü algılama yöntemidir.
En büyük olasılık tahmin yöntemiyle rotor pozisyonu algılama algoritması,
istatistiksel tabanlı bir algoritmadır. Bu algoritma, daha önce motor durum
değişkenlerinden alınan verileri istatistiksel olarak değerlendirerek motorun olasılık
haritasını elde etmeye yardımcı olur. Bu haritayı elde etmek için ilk olarak alınan
veriler sektörlere göre sınıflandırılır. Daha sonra her sektörün ortalama ve standart
sapma değerleri elde edilerek sektör olasılık dağılım fonksiyonları hesaplanır. Ters-
emk'nın Clarke dönüşüm işlem çıktıları, bu olasılık dağılımı işlevlerine girdi olarak
kullanılır. Daha sonra olasılık değeri en yüksek olan fonksiyonun sektörü rotorun
bulunduğu sektör olarak belirlenebilir.
En büyük olasılık tahmin yöntemiyle motor sektörü algılama algoritması, ters-emk
sıfır geçiş noktası algılama yönteminin bir iyileştirmesi olarak düşünülebilir. Çünkü
bu yöntem, motor ters-emk dalga formlarının clarke dönüşümleri elde edilerek
sınıflandırılmasına dayanmaktadır. Genel algoritması ters-emk sıfır geçiş noktası
algılama algoritmasına benzerdir.
Bu çalışmada, Texas Instruments'ın DRV8301 güçlendirici paketi kartında bulunan
evirici ile sürüş işlemleri gerçekleştirilmiştir. Deney düzeneğinin yüksek güç tarafını
kontrol eden bu kartı kontrol etmek için C2000 F28027F fırlatma panosu
kullanılmıştır. Ters-emk sıfır geçiş noktası algılama algoritması ve BLDC motor
uygulamasının en büyük olasılık tahminine dayalı sensörsüz kontrolü, deneysel bir
kurulum hazırlanarak uygulamaları yapılmıştır. Deney düzeneğinden alınan veriler,
yeni önerilen yöntemi geliştirmek için Matlab ve Simulink ortamında analiz
edilmiştir. Ayrıca bu yöntemlerin teorik modelleri de verilmektedir. Daha sonra bu
yöntemler karşılaştırılarak avantajları ve dezavantajları ortaya çıkarılmıştır. Önerilen
yeni yöntemin literatürde yer almayan matematiksel arka planı sunulmuştur.
Karşılaştırma sonucunun incelenmesi Fırçasız Doğru Akım Motorunun (FDAM) sıfır
geçiş noktası tespiti tabanlı kontrol yönteminin gürültüye duyarlı olduğunu
göstermiştir. Ters-emk sinyallerindeki gürültü nedeniyle ani değişiklikler bu
xxiv
yöntemde erken komütasyon süreci sorununa yol açar. Yeni önerilen en büyük olasılık
tahminine dayalı komütasyon algoritması bu sorunu çözmeyi amaçlamaktadır.
Sonuç olarak, bir istatistiksel olasılık yöntemi, FDAM ters-emk sıfır geçiş algılama
kontrol yöntemi ile birleştirildi. Güncel ters-emk sıfır geçiş yönteminin geniş hız
aralıklarında komütasyon kontrolünde dezavantajları olmasına rağmen, yeni önerilen
yöntemin tüm hız aralıklarında güvenilir olduğu gösterilmiştir.
xxv
INTRODUCTION
Brushless Direct Current (BLDC) motors are one of the extensively used electrical motor types
in the last decade due to their some advantages compared with other motor types. These motor
types have various application areas likewise military, aerospace, and automotive, in industry.
The most important advantage of these motor types is its smaller size which is compared to its
equivalents. One of the other remarkable advantages of these types of motor is its high
efficiency thanks to permanent magnets. BLDC motors rotor side have permanent magnets
which produce a natural magnetic field in the air gap. For this reason there is no copper loss in
the rotor [1]. This situation makes BLDC motors more efficient than its counterparts.
BLDC motors can be accepted as the improved version of the Brushed DC motors. Brushed
DC motor’s brush and commutator system lead to a large amount of mechanical losses in the
rotor side of the machine. Due to that, with the development in the permanent magnet industry
brush and commutator systems were left their seat to permanent magnets. This situation gave
way to the development of BLDC motors.
Advantages sorted above make BLDC motors indispensable. However BLDC motors have
some disadvantages. These motors need driver circuits which raises the cost of the overall
system. Also for proper driving rotor position or speed have to be detected [2]. Easiest and
cheapest way to detect rotor position can be realized by hall effect sensors. But these sensors
increase the motor size. Other types of sensors, such as rotary encoders, are too expensive.
Because of these reasons sensorless control is essential for reducing overal system cost and
size.
Maximum likelihood estimation is a widely used method to define parameters of PDF which
belongs to a prespecified class. In this method, different types of likelihood functions to obtain
a PDF from the collected dataset can be used. After using this PDF function the new data point’s
probability of being in the cluster can be calculated. These functions can give to control system
a probabilistic perspective.
1
As it stated above determination of the position is so crucial for sensorless driving . There are
numerous methods to detect the position of BLDC motors for sensorless control. Commonly
used method for sensorless control of BLDC motors is Back-EMF zero crossing detection
method due to its easy implementable algorithm. But these methods have got a major
disadvantage. It is a noise sensitive method.
Purpose of Thesis
In this thesis, A new method called MLE based sector determination from Back-EMF values
regardless the operational speed interval was targetted. MLE based sector determination
method's algorithm and theory of operation will be explained.MLE based sensorless control of
BLDC motor implementation and zero-crossing detection method results are mentioned.
Theoritical results were confirmed by practical study. Results are all given in comparison.
Literature Review
The permanent magnet brushless motors issue has been popular for the last 50 years. This
popularity is sourced by it’s high power density, high efficiency and small size. If this subject
deeply examined, separation of these motors is obligation can be seen, due to their Back-EMF
types. Permanent Magnet brushless motors can be divided into two categories which are named
BLDC and PMSM. Both of these machine’s structures are the same but their Back-EMF types
are different [3]. PMSM motor’s Back-EMF signal changes sinusoidally. But the BLDC motor
Back-EMF signal changes trapezoidally. This situation is sourced by the driving topology of
the motor. Some topologies use trapezoidal commutation control in the literature. Other
topologies use vector control for driving BLDC motors. Vector control of BLDC motor
algorithms uses space vector modulation for pwm generation. But trapezoidal commutation
control algorithms use a six step commutation method for pwm generation. This is one of the
issues that is confused in academy and industry. In the literature review part, trapezoidal Back-
EMF type, sensorless control methods will be focused on..
In the 1960's power electronic circuits developed rapidly. The demand for complex power
electronics circuits became available after this point. As a result of this development, BLDC
motors were started to be used. As stated above their high efficiency and small size made these
motors the leader of high-tech application. In literature there are a lot of control techniques and
2
approaches about trapezoidal sensorless control of BLDC motors. According to Tae-Hyung
Kim, this topic can be divided into four main categories [4]. Each category uses different types
of control methods.
𝑡
𝑒(𝑡) (1.1)
𝑉=∫ 𝑑𝑡
0 𝑘
If the Back-EMF waveform is thought as trapezoidal, the integration equation which is shown
in equation (1.1), can be easily solved like shown in equation (1.2). m denotes the slope of line.
Integral starts from Back-EMF zero crossing point.
𝑚𝑡 2 (1.2)
𝑉=
2𝑘
3
Figure 1.2 : Back-EMF voltage integration method [5].
When reached to a predefined integration limit, commutation of the motor can be realized. m
can be defined according to BLDC motor’s actual speed because Back-EMF highest value has
relationship with motor speed. When the BLDC motor’s speed increases the slope of the Back-
EMF function will be increased. As a result of that m have to be selected as higher valued.
Same logic is valid for the opposite condition. But for the opposite condition m have to be
decreased. The Back-EMF integration method can be stated as an easy implementable method.
However this method is weak at low speeds due to integration errors [6]. Because of this
researchers try to develop new methods in history. Alternative methods have been proposed in
literature for solving this problem.
Secondly Freewheeling diode conduction detection methods can be told, after the Back-EMF
integration technique method. 3 phase Inverters are used to control the BLDC motors. As
known from the topology of inverters have three arms. Each arm has two switches. Each of the
switches of inverters have freewheeling diodes which are parallely connected to switches.
These diodes are placed on an inverter to dissipate the energy, which is stored in the coils. On
each commutation step, different diodes operate. As a result of that, exact commutation time
can be determined using the operating diode conduction time. Due to that, this method uses the
diode conduction time when the switch is off. This conduction time can be determined with
sensing of current flow on the freewheel diodes. Because inductance energy flows through these
diodes when the switch is off. According to diode conduction situation commutation can be
realized. In 2004 Wei Kun and others published an article related to the PWM scheme of this
method Figure 1.3 shows this pwm scheme [7].
4
Method of Freewheeling diode conduction detection has some disadvantages. The most
important one is measurement circuits cost. Inverters have six switches. Due to that, six
measurement circuits have to be used for detecting diode conduction time. In the field of
application, the size of the circuit is a very important factor to implement any method.
Accordingly, the size effect of the sensing circuits has reduced the popularity of this method.
Thirdly, another prominent method for trapezoidal control of BLDC is flux linkage based
commutation. These types of methods try to calculate each phase's flux value. According to the
situation of these flux values, commutation can be decided. Equation (1.3) shows the stator flux
value calculation. In this equation, 𝜑𝑠 denotes the stator flux value. V is the motor’s applied
stator voltage. 𝑖 denotes the current which flows into the motor. R is the resistance of a BLDC
motor.
(1.3)
𝜑𝑠 = ∫(𝑉 − 𝑖𝑅)𝑑𝑡
Motor stator flux can be divided into two main parts. One of the parts can be called as rotor
induction flux linkage which is sourced by permanent magnets placed in rotor and the other one
can be called as armature reaction flux. Equation (1.4) shows the relationship between stator
flux and the others. In equation (1.4) 𝜑𝑓 denotes the rotor induction flux and 𝜑𝑎 denotes the
armature reaction flux linkage.
𝜑𝑠 = 𝜑𝑓 + 𝜑𝑎 (1.4)
5
Armature reaction flux linkage can be calculated directly with the use of equation (1.4) .
Equations have to be calculated for each phase of the BLDC motor. Because this motor is a
type of 3 phase electrical motor. In equation (1.4) 𝐿 denotes the BLDC motor’s inductance
value and 𝑖 denotes the phase current of BLDC motor.
𝜑𝑎 = 𝐿𝑖 (1.5)
After calculating the each phase armature reaction flux linkage, clarke transformation of these
values can be obtained. Using clarke transformation gets the each flux vector’s projection on 𝛼
and 𝛽 axises. Due to that, the number of variables can be decreased. Figure 1.3 shows graph of
𝜑𝑎𝑎 which denotes the 𝛼 axis component of armature reaction flux, versus 𝜑𝑎𝛽 which denotes
the 𝛽 axis component of armature reaction flux.
Graph shows hexagonal shape as seen in above. Each edge of the hexagon corresponds to each
commutation sector of the BLDC motor. Using the corner of the edges commutation of the
BLDC motor can be realized. This topic is explained well in the article which is named as
“Stator Flux Linkage Set for Brushless DC Motor Direct Torque Control”, by Wei Haifeng and
the others [8].
This subject is still workable in the literature. In 2019, explanatory work which is based on flux
linkage based commutation, has been published by Rajesh Nalli and others [9]. Another
prominent publication on this topic was published in 2011 by Salih Baris Ozturk and Hamid A.
6
Toliyat [10]. This publication is related to indirect flux control of BLDC motors. Fourthly,
Back-EMF sensing based methods topics can be divided into two main parts. One of the main
subjects can be called as third harmonic based trapezoidal commutation and the other one can
be called as Back-EMF zero crossing point detection(ZCPD) based commutation method. Both
of the methods sense Back-EMF for commutating the BLDC motor. But these methods use
different features of the Back-EMF waveform.
Third harmonic based methods use the Back-EMFs third harmonic for proper commutation. In
figure 1.3 zero crossing points can be clearly seen. There are a lot of articles about this topic in
literature. In 2006, J. X. Shen and S. Iwasaki published an article which is named “Sensorless
Control of Ultrahigh-Speed PM Brushless Motor Using PLL and Third Harmonic Back EMF”
about this topic [11]. After this publication in 2018 Xinda Song others published an article for
high speed BLDC motors about this topic [12]. In 2019 again Xinda Song and others added
high precision compensation to their method [13]. Back-EMF of the BLDC motor and its third
harmonic as shown in below.
Finally Back-EMF ZCPD method’s general algorithms detect zero crossing points of phase
Back-EMF voltage. This division is more popular than the other division in literature. This type
of method was first proposed by Kenichi Iizuka and others in 1985 [14]. In 2002 Jianwen Shao
and others published a paper which name is “A Novel Direct Back EMF Detection for
Sensorless Brushless DC (BLDC) Motor Drives”, about this topic in STMicroelectronics Power
Systems Applications Laboratory [15]. One year later In 2003, the same research group
7
improved their method [16]. In 2008, Yen-Shin Lai and Yong-Kai Lin published an article
which name is “Novel Back-EMF Detection Technique of Brushless DC Motor Drives for Wide
Range Control Without Using Current and Position Sensors” about to use this method in a wide
speed range [17]. After few years later In 2010 P. Damodharan and Krishna Vasudevan
published an article which name is “Sensorless Brushless DC Motor Drive Based on the Zero-
Crossing Detection of Back Electromotive Force (EMF) From the Line Voltage Difference”
for detecting zero crossing from line voltage difference.
After examining the general sector determination methods, another prominent subject which is
parameter estimation of the BLDC motor, can be told. Because all the methods stated above
use parameters of the BLDC motor for sector determination. Such as resistance, inductance,
viscous friction coefficient and inertia of the rotor. Due to that this topic is so crucial for the
sensorless control of the BLDC motor.
Parameter estimation of the BLDC motor topic has numerous methods. But all of them can be
grouped under two main parts. One of the parts can be named as online parameter estimation
and the other part can be named as offline parameter estimation [18].
Online parameter estimation methods, as understood from its name, estimates parameters of the
motor when the motor is running. These methods are more powerful than the offline parameter
estimation methods. Because online estimation methods detect parameter in running state of
machine.
8
Figure 1.7 : Parameter Estimation Techniques [18].
Least square approximation of BLDC motor parameters is commonly used methods for
parameter estimation. This method gets measurements from the BLDC motor and collects these
data for the estimation dataset. Then using the LS approximation technique parameters of the
motor can be obtained. In this thesis, Least square estimation will be used for identifying the
parameter of the BLDC motor. Table 1 shows the advantages and disadvantages of the
commonly used parameter estimation techniques [18].
In literature, according to review, there is no any publication encountered related with MLE
based zero crossing detection. Due to that, this topic will be new in literature.
9
Table 1.1 : Parameter estimation techniques advantages and disadvantages [19].
10
Hypothesis
The important phase of Sensorless BLDC motor control undoubtedly is the commutation
process. Proper commutation controls the BLDC motor better. As far as understood from the
literature review, the broadly used sensorless BLDC trapezoidal control method is the Back-
EMF zero-crossing detection method. As it stated before this method is sensitive to noise. The
hypothesis of this thesis is to reduce this effect with an MLE based commutation algorithm.
Proposed algorithm will contribute to reduce the effect of noise with power of probability. In
addition to that, a newly proposed method will be earned in literature.
11
BLDC MOTORS
BLDC motors consist of two main parts. One of these parts is named as stator and the
other named as rotor. Stator as understood from the name static part of this motor
consists of windings which generates variable magnetic field in the motor’s air gap.
Rotor as understood from it’s name rotating parts of this motor consist of permanent
magnets. Permanent magnets on the rotor can be placed differently related to
construction of the rotating part.
12
Stator of this motor consists of classical 3-phase windings.These winding is placed
into a rotor with 120 degrees electrical angle differences. In Figure 2.3, 12 slotted 14
pole machine winding schemes are illustrated. A full mechanical angle is 360 degrees.
In a 12-pole motor the pole pitch can be calculated from 360/12 as 30 degrees. If Phase
A winding places on the first slot, Phase B first winding has to be placed on the fifth
slot. Because the fifth slot and first slot have 120 electrical degrees difference. Also
the same calculation is valid for the Phase C. Using this technique all the slots can be
filled with proper windings.
Mathematical Model
13
voltage law to Figure 2.3 electrical side equation of motor can be obtained easily.
Electrical side equations as shown in below [22].
𝑑𝑖𝑎 (2.1)
𝑉𝑎 = 𝑅𝑖𝑎 + 𝐿 + 𝑒𝑎
𝑑𝑡
𝑑𝑖𝑏 (2.2)
𝑉𝑏 = 𝑅𝑖𝑏 + 𝐿 + 𝑒𝑏
𝑑𝑡
𝑑𝑖𝑐 (2.3)
𝑉𝑐 = 𝑅𝑖𝑐 + 𝐿 + 𝑒𝑐
𝑑𝑡
BLDC motor rotating part has ball bearing on its shaft. This ball bearings have viscous
friction. As known from physics, all the rotating parts have inertia. Therefore, BLDC
motor's rotor has inertia. If all of these physical facts take into account, the BLDC
motor’s mechanical equation can be obtained directly. BLDC motor mechanical parts
equations can be given as shown in below. In equation (2.4) 𝐽 denotes viscous friction
coefficient of the motor. 𝐽 denotes the inertia of the rotor. 𝑇𝑚 denotes mechanical
torque. 𝑇𝑒 denotes electrical torque. Finally, the speed of the rotor is denoted by 𝜔.
𝑑𝜔 (2.4)
𝑇𝑒 − 𝑇𝑚 = 𝐵𝜔 + 𝐽
𝑑𝑡
14
These two mathematical parts of the BLDC motor have to connect each other for the
ending the complete model of the BLDC motor. Power of each phase can be obtained
from multiplication of Back-EMF and phase current. Also Torque and speed
multiplication equals power of motor. If these equations combine, the electrical torque
equation of the BLDC motor can be derived. BLDC motor electromechanical parts
equation can be given as shown in below.
𝑃 = 𝑒𝑎 𝑖𝑎 + 𝑒𝑏 𝑖𝑏 +𝑒𝑐 𝑖𝑐 (2.5)
𝑃=𝑇𝜔 (2.6)
𝑒𝑎 𝑖𝑎 + 𝑒𝑏 𝑖𝑏 +𝑒𝑐 𝑖𝑐 (2.7)
𝑇𝑒 =
𝜔
Clarke Transformation
1 1 (2.8)
1 − −
2 2
𝐻= 0 √3 √3
−
2 2
1 1 1
[2 2 2 ]
15
BLDC Motor Driving Circuit
BLDC motors need pre-driver circuits for driving. Like it’s counterparts BLDC motors
also use 3 phase inverter circuits for driving. These inverter circuits consist of six
different switches which constitute three parallel arms. Each of the switches can be
controlled by pulse width modulation signals. Controlling duty cycle of the pulse width
modulation signal controls the conduction time of the switch. High side switch of the
arm and low side switch of the arm can’t be opened at the same time. If this situation
occurs, This can cause a short circuit in this arm. Short circuits can harm the switches.
For preventing short circuit dead time is essential.
16
Figure 2.7: 3 phase inverter.
BLDC motor inverter circuits need high frequency pulse width modulation signals.
Therefore in these applications generally MOSFET and IGBT switches can be used.
There are various types of switches which can operate at higher frequencies than
MOSFET and IGBT but MOSFET and IGBT are more economical than the others.
Characteristics of MOSFET
The MOSFET working region can be divided into three categories. One of the
categories can be named as cut-off region which is the off state of MOSFET. Second
one can be named as the linear region of MOSFET. In general, this region is used for
17
electronic amplification applications. Finally, the third region can be named as
saturation region. This region of MOSFET can be used for switching any device.
Power applications use two regions of MOSFET for proper switching. The linear
region of the MOSFET is not suitable for the power application. Because this region
sets MOSFET output according to linear combination of gate voltage [25].
18
Commutation of BLDC Motor
Commutation of the BLDC motor is the most critical part of BLDC motor driving. If
commutation of the motor realized in the wrong sequence or time, a possible rotor lock
situation can happen. This situation can harm the BLDC motor’s load or driver circuit.
Because when motor locked current decreases to zero abruptly. Derivative of the
current goes to infinity theoretically. Back-EMF of the motor goes infinity as a result
of that. This case can damage our motor’s isolation.
Commutation of BLDC motor consists of six steps. In each step, only two switches
can be in open state. One switch from the high switch of arms, has to be opened and
one switch from the low side. In each step, different combinations of the switch have
to be used. These combinations can be selected according to the hall sensor signal’s
state. Table 3 given below shows the switch states, phase states and induced voltages
on windings according to hall sensor value. H denotes hall sensor value. QxL denotes
the low side switch states. QxH denotes the high side switch state. Px denotes the phase
state and NC denotes the no connection state. GND shows the phase connected to the
ground of the inverter. +𝑉𝐷𝐶 shows the phase is connected to DC voltage source.
1 0 1 0 1 1 0 0 0 +𝑉𝐷𝐶 GND NC
0 0 1 0 1 0 0 1 0 +𝑉𝐷𝐶 NC GND
0 1 1 0 0 0 1 1 0 NC +𝑉𝐷𝐶 GND
0 1 0 1 0 0 1 0 0 GND +𝑉𝐷𝐶 NC
1 1 0 1 0 0 0 0 1 GND NC +𝑉𝐷𝐶
1 0 0 0 0 1 0 0 1 NC GND +𝑉𝐷𝐶
19
Sensorless Trapezoidal Control of BLDC Motor
Sensorless trapezoidal control of BLDC motors can be realized by sensing voltage and
currents of the motor. As a result of that, controller circuits need some instrumentation
parts. In general, these instrumentation parts are realized by the low CMRR valued op-
amps for current sensing. In addition to that, voltage measurements can be obtained
from voltage dividers but these voltage dividers need some filter circuits. Because
directly sensing voltage is too noisy.
Op-amps are complicated electronic devices which are used to sense electrical
quantities from BLDC motors. Each op-amp has three important parameters to define
its quality. These parameters can be sorted as input resistance, gain and output
resistance.
In ideal case, opamp’s input resistance has to be zero and output resistance has to be
infinite. Due to that, selected opamp’s for the measurement circuit have to be close to
these values for non-ideal cases. If these conditions are selected like that, good
measurements can be taken from the measurement circuit. Currents and voltage
measurements are used in the motor model for obtaining Back-EMF of the motor. Each
of the measured variables are used as input to the motor model block. Output of this
motor model block can be stated as BLDC motor’s Back-EMF values. After that sector
determination algorithms come into play. These algorithms may differ depending on
the method used. However, their duties are the same. Each of them try to calculate the
proper sector for switching. Final part of Sensorless trapezoidal control of BLDC
motor is pulse width modulation generation. These control signals can be generated by
the comparison of sawtooth waveform and output of the duty cycle. Generation of the
20
pwm can be realized either analog or digital. The generated signals are sent to the
inverter to control the motor. General closed loop control diagram as shown in Figure
2.12.
The Back-EMF zero crossing method is one of the commonly used sensorless methods
for controlling BLDC motors. In this method basis on the back-emf zero crossing point
detection. But these zero crossing points do not give us direct signals which are the
same with hall sensors output. These zero crossing points have sixty electrical degrees
difference with the hall sensors output. Therefore when zero crossing point based hall
signals are detected, firing of the switches should be realized according to next hall
sensor value.
Software models use Analog Digital Converter (ADC) to detect voltage and current
level. Most important parameter for ADC can be stated as resolution. If systems have
higher ADC resolution, it can detect the voltage level of the signal more sensitively.
Therefore sensitivity of the system is directly related with the ADC resolution.Due to
that, in this thesis, 12 bit ADC resolution is used for sensitive signal level detection.
22
MAXIMUM LIKELIHOOD ESTIMATION
1 (𝑥−µ)2 (3.1)
2) −
𝑃(𝑥; 𝜇, 𝜎 = 𝑒 2𝜎2
𝜎√2𝜋
Let’s make an example about this theorem. With the help of Matlab take random samples which
size is equal 100, with randn() function from population. After that when the histogram of these
samples is plotted, Figure 3.2 can be obtained. With the increasing the size of random samples
histogram which look like a gaussian distribution like Figure 3.3 can be drawn.
23
Figure 3.2: Histogram of randomly selected samples n=100
24
Theory of Maximum Likelihood Estimation
Likelihood word first started to be used in the fourteenth century[30]. After that Fisher
published an article which is named as “An absolute criterion for fitting frequency curves”, in
1912 [31]. In this publication, Fisher introduced the curve fitting criterion according to the
frequency of obtaining data that could be called histogram. Nine years later he named this
criterion as likelihood [30].
In statistics, likelihood can be defined as a measure of goodness of curve fitting with uncertain
parameters. Based on this, maximum likelihood estimation can be defined as finding the
parameters which maximizes the likelihood. Likelihood function is defined as shown in
equations (3.2) and (3.3). Probability distribution function type is selected as gaussian
distribution due to central limit theorem. In equation (3.2) and (3.3) L(x) denotes the likelihood
function of the x variable.
𝑛
(3.2)
𝐿(µ, 𝜎 2 ; 𝑥1 , … , 𝑥𝑛 ) = ∏ 𝑃(𝑥𝑗 ; 𝜇, 𝜎 2 )
𝑗=1
𝑛 2
(3.3)
(𝑥𝑗 −µ)
1 −
𝐿(µ, 𝜎 2 ; 𝑥1 , … , 𝑥𝑛 ) = ∏ 𝑒 2𝜎2
𝑗=1
𝜎√2𝜋
𝑛 (3.5)
𝑛 1 2
2
𝐿(µ, 𝜎 ; 𝑥1 , … , 𝑥𝑛 ) = (2𝜋𝜎 2 )− 2 exp (− 2 ∑(𝑥𝑗 − µ) )
2𝜎
𝑗=1
After this step, the natural logarithm of the likelihood function can be calculated. Because
exponential parts of the equation should be to vanish. This will help us later in calculations.
25
This action will ease the take derivative in the next steps.In literature, calculating the logarithm
of likelihood function is called a log-likelihood function. In general, a logarithm function is
selected for this purpose. Because logarithm function can be increased when likelihood function
increases and also it can be decreased when likelihood function decreases. Logarithm of
likelihood function as shown in below. 𝑙 denotes the log-likelihood function in the below
equations.
𝑛 (3.7)
2
𝑛
2 )− 2
1 2
𝑙(µ, 𝜎 ; 𝑥1 , … , 𝑥𝑛 ) = ln ((2𝜋𝜎 exp (− 2 ∑(𝑥𝑗 − µ) ))
2𝜎
𝑗=1
𝑛 (3.8)
2
𝑛
2 )− 2
1 2
𝑙(µ, 𝜎 ; 𝑥1 , … , 𝑥𝑛 ) = ln ((2𝜋𝜎 ) + ln (exp (− 2 ∑(𝑥𝑗 − µ) ))
2𝜎
𝑗=1
𝑛
𝑛 1 2
(3.9)
𝑙(µ, 𝜎 ; 𝑥1 , … , 𝑥𝑛 ) = − ln(2𝜋𝜎 2 ) − 2 ∑(𝑥𝑗 − µ)
2
2 2𝜎
𝑗=1
𝑛
𝑛 𝑛 1 2
(3.10)
𝑙(µ, 𝜎 ; 𝑥1 , … , 𝑥𝑛 ) = − ln(2𝜋) − ln(𝜎 2 ) − 2 ∑(𝑥𝑗 − µ)
2
2 2 2𝜎
𝑗=1
After several simplifications, it can be clearly stated that problem in maximum likelihood
estimation is maximization of the log-likelihood function. From basics of calculus knowledge,
it can be clearly say that for maximizing or minimizing any function, it’s partial derivative with
respect to parameter which wanted to be maximized, have to equal to zero [32].
max ̂
𝑙(µ, 𝜎 2 ; 𝑥1 , … , 𝑥𝑛 ) = µ, 𝜎2 (3.11)
2
µ,𝜎
𝜕 (3.12)
𝑙(µ, 𝜎 2 ; 𝑥1 , … , 𝑥𝑛 ) = 0
𝜕µ
26
𝜕 (3.13)
2
𝑙(µ, 𝜎 2 ; 𝑥1 , … , 𝑥𝑛 ) = 0
𝜕𝜎
When the log-likelihood function’s derivative is taken with respect to µ which symbolizes the
mean of population, equation (3.16) can be obtained. When this equation is equal to zero, the
population's maximum likelihood mean estimate like in equation (3.18) can be obtained.
𝑛 (3.14)
𝜕 𝜕 𝑛 𝑛 1 2
𝑙(µ, 𝜎 2 ; 𝑥1 , … , 𝑥𝑛 ) = (− ln(2𝜋) − ln(𝜎 2 ) − 2 ∑(𝑥𝑗 − µ) )
𝜕µ 𝜕µ 2 2 2𝜎
𝑗=1
𝑛
𝜕 1 (3.15)
𝑙(µ, 𝜎 2 ; 𝑥1 , … , 𝑥𝑛 ) = 2 ∑(𝑥𝑗 − µ)
𝜕µ 𝜎
𝑗=1
𝑛
𝜕 1 (3.16)
𝑙(µ, 𝜎 2 ; 𝑥1 , … , 𝑥𝑛 ) = 2 ((∑ 𝑥𝑗 ) − 𝑛µ)
𝜕µ 𝜎
𝑗=1
𝑛
(3.17)
((∑ 𝑥𝑗 ) − 𝑛µ) = 0
𝑗=1
𝑛
1 (3.18)
𝜇̂ = ∑ 𝑥𝑗
𝑛
𝑗=1
The same procedure is also valid for the estimation of the population variance. When the
derivative of the log-likelihood function is taken with respect to 𝜎 2 which denotes variance of
a population, equation (3.23) can be achieved. If this equation equalizes to zero, the maximum
likelihood variance estimate of the population can be obtained like in equation (3.24).
𝑛 (3.19)
𝜕 2
𝜕 𝑛 𝑛 2)
1 2
𝑙(µ, 𝜎 ; 𝑥1 , … , 𝑥𝑛 ) = (− ln(2𝜋) − ln(𝜎 − ∑(𝑥𝑗 − µ) )
𝜕𝜎 2 𝜕𝜎 2 2 2 2𝜎 2
𝑗=1
𝑛 (3.20)
𝜕 2
𝑛 1 2 𝑑 1
𝑙(µ, 𝜎 ; 𝑥1 , … , 𝑥𝑛 ) = − − [ ∑(𝑥𝑗 − µ) ] ( )
𝜕𝜎 2 2𝜎 2 2 𝑑𝜎 2 𝜎 2
𝑗=1
27
𝑛 (3.21)
𝜕 2
𝑛 1 2 1
𝑙(µ, 𝜎 ; 𝑥1 , … , 𝑥𝑛 ) = − − [ ∑(𝑥𝑗 − µ) ] (− )
𝜕𝜎 2 2𝜎 2 2 (𝜎 2 )2
𝑗=1
𝑛 (3.22)
𝜕 2
𝑛 1 2 1
𝑙(µ, 𝜎 ; 𝑥1 , … , 𝑥𝑛 ) = − + [ ∑(𝑥𝑗 − µ) ] ( )
𝜕𝜎 2 2𝜎 2 2 (𝜎 2 )2
𝑗=1
𝑛 (3.23)
𝜕 1 1 2
2
𝑙(µ, 𝜎 2 ; 𝑥1 , … , 𝑥𝑛 ) = 2 [ 2 ∑(𝑥𝑗 − µ) − 𝑛]
𝜕𝜎 2𝜎 𝜎
𝑗=1
𝑛
1 2
(3.24)
̂2
𝜎 = ∑(𝑥𝑗 − µ)
𝑛
𝑗=1
Theory of the maximum likelihood estimation is a little bit complicated but the using of
maximum likelihood estimation’s results are effective. If random samples from one set being
have, using the equations (3.18) and (3.24) the best fitted Gaussian probability distribution
function to that set can be found [33].
28
Maximum likelihood estimation based classification is so useful if randomly sampled datasets
being had. Randomly sampled 3 different independent datasets can be seen from Figure 3.4.
Each of the datasets have mean and variance calculations which can be calculated from equation
(3.18) and (3.24). The calculated values can be seen as shown in below. Green colored dataset
will be mentioned as the first dataset and red will be second and blue will be the third one.
𝜇 𝜎
𝑋1 5.0620 𝑋1 0.9858
𝑋2 -5.0501 𝑋2 1.0123
𝑋3 2.0103 𝑋3 0.9988
𝑌1 -0.0217 𝑌1 0.9685
𝑌2 -0.0158 𝑌2 1.0512
𝑌3 -5.9953 𝑌3 1.0198
After using these means and variances values of each dataset, probability distribution function
of each class can be obtained. Probability distribution function of the class can be obtained by
the multiplication of each direction’s probability distribution function.
1 (𝑥−5.0620)2 (3.26)
𝑃1𝑥 (𝑥) = 𝑒− 2∗0.9718
0.9858√2𝜋
1 (𝑦+0.0217)2 (3.27)
𝑃1𝑦 (𝑦) = 𝑒− 2∗0.9380
0.9685√2𝜋
1 (𝑥+5.0501)2 (3.28)
−
𝑃2𝑥 (𝑥) = 𝑒 2∗1.0248
1.0123√2𝜋
29
1 (𝑦+0.0158)2 (3.29)
−
𝑃2𝑦 (𝑦) = 𝑒 2∗1.1050
1.0512√2𝜋
1 (𝑥−2.0103)2 (3.30)
𝑃3𝑥 (𝑥) = 𝑒− 2∗0.9976
0.9988√2𝜋
1 (𝑦+5.9953)2 (3.31)
−
𝑃3𝑦 (𝑦) = 𝑒 2∗1.04
1.0198√2𝜋
Equations from (3.25) to (3.31) can help us to calculate new data point’s probability being in
class or not. Algorithm of this procedure as shown in below. 𝑃(𝑥, 𝑦|𝑐 = 1) can also shown as
𝑃1 (𝑥, 𝑦).
Calculate
𝑃(𝑥, 𝑦|𝑐 = 1)
𝑃(𝑥, 𝑦|𝑐 = 2)
𝑃(𝑥, 𝑦|𝑐 = 3)
Let’s obtain the class probability values of point x=0 y=0. First 𝑃(0,0|𝑐 = 1) value and
secondly 𝑃(0,0|𝑐 = 2) value and finally 𝑃(0,0|𝑐 = 3) will be calculated. Calculations are
as shown in below.
30
𝑃(0,0|𝑐 = 3)=𝑃3𝑥 (0)𝑃3𝑦 (0)=0 (3.33)
Then, the Maximum probability value can be obtained from the second class’s probability
distribution function. Because of that about this data point can be said that should belong to
the second class. The MATLAB M-file for this example can be found in Appendix A.
31
REALIZATION OF PROPOSED COMMUTATION METHOD
Experimental setup of this thesis can be divided into three parts. One of the parts is
named as a microcontroller which controls the inverter. This microcontroller arranges
pwm duty cycle for controlling the inverter. All codes are executed in this part of
experimental setup. Another part of this experimental setup is inverter which is the
high power-side of this experimental setup. This part uses control signals which are
low powered, then converts it into high powered signals. Finally the last part of this
experimental setup is the BLDC motor which is the main plant of this experiment.
Experimental setup of this thesis as shown in Figure 4.1.
4.1.1 Microcontroller
32
not the best way to look at their application to tell the difference between the two. It is
the interior architecture that determines the difference between them. DSPs are
generally optimized to perform high speed and high accuracy crash operations.
4.1.2 Inverter
Inverter circuit is the indispensable part for driving any BLDC motor. In inverter
circuits the most important component is switches. There are several types of switches
in the market as it stated in the second chapter. For efficient drive the most important
parameter of switch is 𝑅𝐷𝑆 which is called drain source resistance for MOSFET.
Because according to value of 𝑅𝐷𝑆 , switch loses can be increase or decrease. Another
important parameter for inverter circuits is switching time. Less switching time can
follow the control signals better. When this is considered these variables best choice
33
in the market can be defined as NextFET’s which is an abbreviation of next generation
MOSFET’s. But these MOSFETs can't be driven directly by control signals due to
floating voltage points between each high and low side on the inverter's arms. They
need a pre-driver circuit or IC for driving. For this purpose DRV8301 IC can be used
to control switches. This IC also has overcurrent and overvoltage protection for safety
operation.
In this thesis, a suitable inverter circuit which name is the DRV8301 booster pack
which includes all of these components has been selected, for sensorless driving of the
BLDC motor. This booster pack also has voltage and current sensing circuits which
includes high precision and CMRR valued op-amps. F28027F launchpad can be
directly connect this inverter circuits via connectors. DRV8301 Booster Pack and
DRV8301 IC as shown in Figure 4.3.
34
Figure 4.4: DRV8301 IC inside [36].
BLDC motor designs and costs vary in the market.A lot of motor types that can be
classified according to their rotor type, can be found in the market. These types of
motor are also used in quadcopter applications. Used BLDC motors in quadcopter
applications need more speed and power density values. In some quadcopter
applications speed values can vary from 3000 rpm to 10000 rpm value. In this thesis,
a surface mounted magnet BLDC motor which can turn 10000 rounds per minute have
used.
Parameter estimation of the BLDC motor is so crucial for the sensorless trapezoidal
control of the BLDC motor. Because most of the model uses these parameters. Such
35
as resistance, inductance, viscous friction constant and inertia. As it stated before, a
commonly used method for this operation is Least Squares Estimation(LSE).
Approximating any constant can be realized by taking the average of the obtained data
from experimental setup. Proof this can be obtained easily. Let’s think of a function
like shown in equation (4.1).
10 (4.1)
2
𝐹(𝛼) = ∑(𝛼 − 𝑦𝑖 )
𝑖=1
Equation (4.1) can be minimized by taking derivative with respect to 𝛼 and after then
equaling the derivative to zero. If these are calculated, equation (4.2) can be obtained.
𝑛
(4.2)
2 ∑(𝛼 − 𝑦𝑖 ) = 0
𝑖=1
𝑛 𝑛
(4.3)
∗)
2 ∑(𝛼 = 2 ∑(𝑦𝑖 )
𝑖=1 𝑖=1
𝑛
1 (4.4)
∗
𝛼 = ∑(𝑦𝑖 )
𝑛
𝑖=1
As understood from the equation (4.4) each parameter of the motor can be calculated
by the taking mean of the obtained experimental data [37].
BLDC motor’s parameters can be calculated using equation (4.4). Obtained results as
shown in Table 4.2.
36
Table 4.2 : BLDC motor parameters [38].
R L J B
In this thesis, A2212 BLDC motor is used for test the algorithms which will be stated
after. This motor will helps us to test both of the methods which will deeply examined.
Parameters of this motor can be found also in article which is named as “Parameter
Estimation of Brushless Dc Motor Using Experimental Methods” which is written by
A.F.Hawary and the others [38]. View of this motor as shown in below.
Software based Back-EMF zero crossing detection algorithms generally use two
different cycles. One of these cycles which can be called as the main cycle,
commutates the motor and the other one of these cycles which can be called the
interrupt cycle, detects the commutation sector. Interrupt cycle includes the BLDC
motor model for calculating the Back-EMF values of each phase. Each interrupt cycle
can be fired according to the pwm counter’s overflow. This gives to the interrupt cycle
periodically calculation ability.
Software based Back-EMF zero crossing detection method needs high speed on
calculating the Back-EMF signals. Because this method calculates the new values of
output of the motor on each ADC interrupt. These calculations include integration and
37
derivation operations. Due to that, speed of the hardware is so crucial for the software
based Back-EMF zero crossing detection method .
After calculation of BLDC motor’s Back-EMF values zero crossing points can be
detected with the if statements directly. Obtained zero crossing points can be used by
shifting sixty electrical degrees for commutating the BLDC motor. General structure
of the algorithm as shown in below.
38
Figure 4.7: ZCPD based commutation main routine.
39
Simulink Model for Data Collecting
F2807F DSP control cards have serial communication interface peripheral inside.
Using this peripheral, data can be transmitted to any host computer over USB bus.
After selecting the true baud-rate, hardware in the loop simulations can be executed
on the host PC. This ability enables us to examine our models. After debugging a visual
copy of the system, the model can easily adapt to the software. General simulink model
which used to collect data as shown in Figure 4.6.
As shown in above this model consists of three different blocks. One of the blocks can
be named as the input stage of the model. Figure 4.7 shows the input block inside.
40
Input section of this block includes the “SCI RCV'' block which is necessary for
obtaining data from serial ports. After getting the data from the serial port, these values
have to be converted to the wanted data type for processing. The Demultiplexer block
in Figure 4.7 is used to separate data from each other. Then the separated voltage and
current data are multiplexed by the multiplexer block of Simulink. Finally this block
subtracts 2048 value for removing DC offset from data.
Secondly, another block which is the heart of the model, can be named as motor
mathematical model. In figure 4.8, the BLDC motor mathematical model is illustrated.
This block calculates the BLDC motor’s Back-EMF values which is obtained from the
motor's voltage and current values.
Finally zero point crossing detection blocks can be illustrated in this model. This block
is the last stage of the model. Obtained Back-EMF values zero crossing points are
41
detected by this block. Comparing the Back-EMF signal with zero is enough for
obtaining the hall sensor's shifted signal.
BLDC motors can’t be triggered directly with sensorless control. Because Back-EMF
waveforms can’t occur at low speeds. Therefore some kind of algorithms are needed
to use at start-up. Generally used algorithm to move a motor at startup can be called a
ramp-up generator algorithm. This ramp-up generator controls the speed of the motor
and generates a suitable duty cycle for BLDC motors soft start-up. After reaching the
minimal speed at which BLDC motor’s Back-EMF waveforms occurred, the
sensorless control algorithm of BLDC motor can be activated. When the Figure 4.6 is
analyzed at the start of the graph due to noise effect it is obvious that hall sensor signals
cannot be obtained truly. When the motor reaches minimal speed, sensor signals can
be obtained from Back-EMF waveforms as shown in Figure 4.13. In general Back-
EMF waveforms are accepted as trapezoidal. But in practical, noise can spoil this
acceptance. If ideal trapezoidal waveforms have been desired to obtain, a two way
limiter can be used for obtaining that waveform. One side of the limiter can be used
to limit the positive side and the other one for the negative side.
42
Clarke transformation of the motor voltage data is important operation for analyzing
data. Because motor BLDC motors are 3 phase system due to that in each coding step
coder have to deal with three different variables. If clarke transformation of the motor
data take, after that,coder can deal with two different variables, which are named as α-
β axis variables of data. This will decrease the complexity of code. In ideal case, when
drawing of the 𝑒𝛼 vs. 𝑒𝛽 graph, is obtained, hexagon shaped graph can be viewed
unlike figure 4.14. But in non-ideal case, 𝑒𝛼 vs. 𝑒𝛽 graph can be obtained as ellipsoidal
due to measurement noise. This graph values can change according to BLDC motor’s
excitation voltage amplitude or load torque value. Shape of the graph can vary
according to motor structure. But in general, But in general, this shape can be called
as an ellipsoid. Graph of the 𝑒𝛼 vs. 𝑒𝛽 as shown in below.
After grouping these Back-EMF values according to their sector, drawing of this graph
with help of MATLAB can be obtained. If the graph is examined, interesting results
can be obtained. There are six sectors for grouping. Each sector has mean and variance
value. If mean and variance values are obtained from the MLE section,classification
of these data for the whole space can be realized. In the ideal case, this shape should
be hexagon but due to the noise effect, the shape of the graph is asymmetrical.
43
Figure 4.14: BLDC Back-EMF Alpha vs Beta values grouped graph according to
hall sensors value.
Central limit theorem shows that, gaussian distribution function can be used for
obtaining class probability distribution function, what if enough samples are being
obtained. This is the general idea, underlying for MLE based commutation logic. Each
sector can be defined with gaussian distribution functions. MLE Based Commutation
of BLDC motors depends on deriving these probability distribution functions from
collected data points. There are six commutation sectors for trapezoidal control of
44
BLDC motors. Each sectors have 𝑒𝛼 and 𝑒𝛽 values. When these values are divided
according to obtained sensor values, each sector’s probability density functions can be
̂2 values have
obtained. For to obtain each sector’s probability density function, 𝜇̂ and 𝜎
to be calculated. Because these estimates reveal the class probability distribution
̂2 values, equation (3.18)
function’s characterization. For proper calculation of 𝜇̂ and 𝜎
and (3.24) can be used. Each sector consists of from one 𝑒𝛼 and 𝑒𝛽 value dataset.
Therefore each sector has two different probability distribution functions. One of them
is derived for 𝑒𝛼 value and the other one is for 𝑒𝛽 value. Intersection of the events
probability, can be found by the multiplication of the events probability. Therefore
Multiplication of these probability density functions can give the sector’s probability
distribution function. After using these functions commutation can be realized
correctly.
45
Sector functions get different values at different inputs as shown in above. In some
cases these functions intersect in some inputs. But this is not a problem. This situation
shows the powerful side of this method. Position of the rotor can’t be determined
directly due to noise. Users can only estimate the rotor probably can be in the which
sector according to probability distribution functions output.
46
Figure 4.18: ADC interrupt state routine.
47
Figure 4.19: Main routine of MLE based commutation.
48
Comparision of the Results
Figure 4.20 shows the closer view of realization results which are obtained from the
oscilloscope, of the Back-EMF zero-crossing point detection based algorithm. The
ideal Back-EMF waveform of the BLDC motor has to be trapezoidal for proper control
of the BLDC motor. As understood from the figure this method in some cycles
commutates motor in the wrong time. The source of the fault can be determined as line
measurement noise. This fault causes torque fluctuations on the BLDC motor. In long-
term usage, these types of faults can reduce the lifetime of the machine.
49
Figure 4.21: Closer view of MLE based commutation algorithm result.
50
CONCLUSION
In this thesis, newly proposed Maximum likelihood estimation based and zero crossing
detection based sensorless control of BLDC motor were realized. General theory of
the methods was given and their detailed algorithms were stated in the previous
chapters. This newly proposed method can be stated as the improved version of the
Back-EMF zero-crossing based methods.
General problem of the zero-crossing detection based method is its being highly noise
sensitive. Because in small power motors signal to noise ratio values are high. As a
result of that, it is possible for the used models to give wrong outputs due to
measurement noise. Obtained results from these models can lock the motor during
operation. If the motor is in full or approximately full load, locking of the rotor cost
can be too expensive for the user. If system health depends on the operation of BLDC
motor working, the whole of the system can be lost due to a system crash.
Comparison of these stated methods can depend also on the operational speed region.
Zero-crossing detection-based methods cannot be used in a wide speed range, while
51
the maximum likelihood estimation-based method can be used. As the number of data
points from different speed ranges increases, it is possible to obtain the probability
map of a BLDC motor. This probability map gives better information about the rotor
position than the classical zero-crossing detection based method with its being less
affected from external noise. This makes the newly proposed method more powerful
than the classical zero-crossing detection based method.
On the other hand, this newly proposed algorithm has some disadvantages when
compared with its counterpart. The most important disadvantages of this method is the
usage of an external function. In embedded systems, exponential functions use floating
point math. These calculations slow down the whole software operation. In the future,
researchers can develop a new speedy exponential calculation method for solving this
problem.
The newly proposed methods can be used in all applications varying from industrial
areas to space applications. One of the most important future works for this method
can be stated as extending the power limits of the control.
52
REFERENCES
[8] Hai-feng, W., Hanpei, W., & Yi, Z. (2016). Stator Flux Linkage Set for Brushless
DC Motor Direct Torque Control.
[9] Nalli, R., Kumar, M., Ramamoorthy, M., & Subbarao, K. (2019). “Sensorless
Control of BLDC Motor using Flux Linkage Based Algorithm”.
International Journal of Engineering and Advanced Technology
Regular Issue, 8(6), 1549-1556. doi:10.35940/ijeat.f8158.088619
53
[10] S. B. Ozturk and H. A. Toliyat, "Direct Torque and Indirect Flux Control of
Brushless DC Motor," in IEEE/ASME Transactions on Mechatronics,
vol. 16, no. 2, pp. 351-360, April 2011, doi:
10.1109/TMECH.2010.2043742.
[11] J. X. Shen and S. Iwasaki, "Sensorless control of ultrahigh-speed PM brushless
motor using PLL and third harmonic back EMF," in IEEE Transactions
on Industrial Electronics, vol. 53, no. 2, pp. 421-428, April 2006, doi:
10.1109/TIE.2006.870664.
[12] X. Song, B. Han, S. Zheng and J. Fang, "High-Precision Sensorless Drive for
High-Speed BLDC Motors Based on the Virtual Third Harmonic Back-
EMF," in IEEE Transactions on Power Electronics, vol. 33, no. 2, pp.
1528-1540, Feb. 2018, doi: 10.1109/TPEL.2017.2688478.
[13] X. Song, B. Han and K. Wang, "Sensorless Drive of High-Speed BLDC Motors
Based on Virtual Third-Harmonic Back EMF and High-Precision
Compensation," in IEEE Transactions on Power Electronics, vol. 34,
no. 9, pp. 8787-8796, Sept. 2019, doi: 10.1109/TPEL.2018.2885031.
[14] K. Iizuka, H. Uzuhashi, M. Kano, T. Endo and K. Mohri, "Microcomputer
Control for Sensorless Brushless Motor," in IEEE Transactions on
Industry Applications, vol. IA-21, no. 3, pp. 595-601, May 1985, doi:
10.1109/TIA.1985.349715.
[15] Jianwen Shao, D. Nolan and T. Hopkins, "A novel direct back EMF detection
for sensorless brushless DC (BLDC) motor drives," APEC.
Seventeenth Annual IEEE Applied Power Electronics Conference and
Exposition (Cat. No.02CH37335), Dallas, TX, USA, 2002, pp. 33-37
vol.1, doi: 10.1109/APEC.2002.989224.
[16] Jianwen Shao, D. Nolan and T. Hopkins, "Improved direct back EMF detection
for sensorless brushless DC (BLDC) motor drives," Eighteenth Annual
IEEE Applied Power Electronics Conference and Exposition, 2003.
APEC '03., Miami Beach, FL, USA, 2003, pp. 300-305 vol.1, doi:
10.1109/APEC.2003.1179230.
[17] Y. Lai and Y. Lin, "Novel Back-EMF Detection Technique of Brushless DC
Motor Drives for Wide Range Control Without Using Current and
Position Sensors," in IEEE Transactions on Power Electronics, vol. 23,
no. 2, pp. 934-940, March 2008, doi: 10.1109/TPEL.2007.915048.
[18] Rafaq, M. S., & Jung, J. W. (2019). A Comprehensive Review of State-of-the-
Art Parameter Estimation Techniques for Permanent Magnet
Synchronous Motors in Wide Speed Range. IEEE Transactions on
Industrial Informatics, 16(7), 4747-4758.
[19] Zhan, X., Zeng, G., Liu, J., Wang, Q., & Ou, S. (2015). A review on parameters
identification methods for asynchronous motor. International Journal
of Advanced Computer Science and Applications, 6(1).
[20] Raj, L., Ganesan, S., & Daniel, M. (01/04/2015). A Review of Novel Brushless
DC Motor and Different Methods of Control. IJSRD - International
54
Journal for Scientific Research & Development, vol. 3, ISSN:2321-
0613, pp. 394-398.
[27] Torres, D. (2011, October 19). Sensorless BLDC Motor Control with Back-EMF
Filtering Using a Majority Function. Retrieved January 13, 2021, from
https://www.digikey.dk/en/articles/sensorless-bldc-motor-control-
with-back-emf-filtering-using-a-majority-function
[28] Le Cam, L. (1986). The central limit theorem around 1935. Statistical science,
78-91.
[29] Murphy, K. P. (2007). Conjugate Bayesian analysis of the Gaussian
distribution. def, 1(2σ2), 16.
[31] Aldrich, J. (1997). RA Fisher and the making of maximum likelihood 1912-
1922. Statistical science, 12(3), 162-176.
[32] Yan, L. (2019, November 8). 20: Maximum Likelihood Estimation. Retrieved
January 19, 2021, from
https://web.stanford.edu/class/archive/cs/cs109/cs109.1206/lectures/
20_mle_blank.pdf
55
[33] Kempthorne, D. (2015, Spring). Parameter Estimation Probability Distributions
Maximum ... Retrieved January 19, 2021, from
https://ocw.mit.edu/courses/mathematics/18-443-statistics-for-
applications-spring-2015/lecture-notes/MIT18_443S15_LEC4.pdf
[37] Nilssen, T. K., & Cai, X. (n.d.). The Method of Least Squares. Retrieved
January 19, 2021, from
https://www.mn.uio.no/ifi/english/services/knowledge/cs/forelesn.kap
5.pdf
[38] Al-Mahturi, Fuad & Samokhvalov, Dmitriy & Bida, Vladislav. (2018).
Parameters identification of a brushless DC motor by specification.
558-561. 10.1109/EIConRus.2018.8317159.
[39] A2212 - 1000KV BLDC Brushless Motor. (n.d.). Retrieved January 13, 2021,
from https://www.electronicscomp.com/a2212-1000kv-bldc-
brushless-motor-india
56
APPENDICES
57
APPENDIX A
clear all
clc
x1=normrnd(5,1,[1,1000]);
y1=normrnd(0,1,[1,1000]);
x2=normrnd(-5,1,[1,1000]);
y2=normrnd(0,1,[1,1000]);
x3=normrnd(2,1,[1,1000]);
y3=normrnd(-6,1,[1,1000]);
figure()
plot(x1,y1,'g o')
hold on
plot(x2,y2,'r o')
plot(x3,y3,'b o')
hold off
figure()
plot(x1,y1,'g o')
hold on
plot(x2,y2,'r o')
plot(x3,y3,'b o')
meanx1=mean(x1);
meanx2=mean(x2);
meanx3=mean(x3);
meany1=mean(y1);
meany2=mean(y2);
meany3=mean(y3);
varx1=sqrt(var(x1));
varx2=sqrt(var(x2));
varx3=sqrt(var(x3));
vary1=sqrt(var(y1));
vary2=sqrt(var(y2));
vary3=sqrt(var(y3));
syms x y
f1=(1/sqrt(2*pi*(varx1^2)))*exp(-1*((x-
(meanx1))^2)/(2*(varx1^2)));
g1=(1/sqrt(2*pi*(vary1^2)))*exp(-1*((y-
(meany1))^2)/(2*(vary1^2)));
P1=f1*g1;
f2=(1/sqrt(2*pi*(varx2^2)))*exp(-1*((x-
(meanx2))^2)/(2*(varx2^2)));
g2=(1/sqrt(2*pi*(vary2^2)))*exp(-1*((y-
(meany2))^2)/(2*(vary2^2)));
P2=f2*g2;
f3=(1/sqrt(2*pi*(varx3^2)))*exp(-1*((x-
(meanx1))^2)/(2*(varx3^2)));
58
g3=(1/sqrt(2*pi*(vary3^2)))*exp(-1*((y-
(meany3))^2)/(2*(vary3^2)));
P3=f3*g3;
prob=[];
for i=-10:0.5:10
for j=-10:0.5:10
a=subs(P1,{x,y},{i,j});
prob(1)=double(a);
a=subs(P2,{x,y},{i,j});
prob(2)=double(a);
a=subs(P3,{x,y},{i,j});
prob(3)=double(a);
index=find(prob == max(prob(:)));
if(index==1)
plot(i,j,'g o')
elseif(index==2)
plot(i,j,'r o')
elseif(index==3)
plot(i,j,'b o')
end
end
end
59
APPENDIX B
60
61
APPENDIX C
62
63
CURRICULUM VITAE
E-Mail : kaplana@itu.edu.tr
EDUCATION :
64