NUMERICAL ANALYSIS OF A BRUSHLESS

PERMANENT MAGNET DC MOTOR

USING COUPLED SYSTEMS

HLA NU PHYU
(B. Eng.(Electrical Power),Y.T.U)

A THESIS SUBMITTED
FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
DEPARTMENT OF ELECTRICAL & COMPUTER ENGINEERING
NATIONAL UNIVERSITY OF SINGAPORE
2004
Acknowledgments

I wish to express my gratitude to my supervisor, Dr. M. A. Jabbar from Depart-

ment of Electrical and Computer Engineering, National University of Singapore

for his guidance, advice, support and encouragement for this research work. I am

grateful to my co-supervisor Dr. Liu Zhejie from Data Storage Institute for his

suggestions and help to this work in all possible aspects.

I am also greatly indebted to Dr. Bi Chao, Research Scientist from the Data

Storage Institute for the experimental set up. I wish to thank Lab officers, Mr.Y.

C. Woo and Mr.M. Chandra from Electrical Machine and Drives Laboratory for

their support and assistance in the Lab where I carried out my research work.

Many thanks to my colleagues: Mr. Nay Lin Htun Aung for his smart ideas and

suggestions concerning with FEM analysis, Ms. Dong Jing for her constant support

and helping hands for programming work, Mr. Krishna Manila for his support,

patience and valuable discussion for both hardware and software implementation

for experiments.

I would like to express my most heartfelt thanks and gratitude to my family,

who have always provided me with constant support, concern and prayers. Finally,

to my husband, San Yu, I express my deepest gratitude. Without his understand-

ing, kindness and sacrifices, the dream would never have come to reality.

i
Contents

Acknowledgement i

Summary viii

List of Figures x

List of Tables xiv

List of Symbols xvi

1 Introduction 1

1.1 Permanent Magnet Motors . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Brushless Permanent Magnet DC Motors . . . . . . . . . . . . . . . 3

1.2.1 Basic Configurations of BLDC motors . . . . . . . . . . . . 4

1.2.2 Characteristics of BLDC Motors . . . . . . . . . . . . . . . . 5

1.3 Magnetic Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.3.1 Hard magnetic materials (Permanent magnets) . . . . . . . 10

1.3.2 Soft magnetic materials . . . . . . . . . . . . . . . . . . . . 14

1.4 Computational Analysis of Electrical Machines . . . . . . . . . . . . 14

1.4.1 Analysis of electrical machines using FEM . . . . . . . . . . 15

1.5 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

1.6 Scope of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

1.7 Outlines of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . 24

ii
 iii

2 Computational Analysis of a BLDC Motor 26

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.2 Finite Element Analysis . . . . . . . . . . . . . . . . . . . . . . . . 27

2.2.1 Mathematical formulations of the physical model . . . . . . 28

2.2.2 Discretization of the problem domain . . . . . . . . . . . . . 31

2.2.3 Derivation of the element matrix equations . . . . . . . . . . 34

2.2.3.1 Galerkin’s formulation for the permanent magnet . 39

2.2.4 Assembling of element matrix equation . . . . . . . . . . . . 40

2.2.5 Imposing the boundary conditions . . . . . . . . . . . . . . . 42

2.2.6 Numerical solution to nonlinear problems . . . . . . . . . . . 46

2.2.7 Solution of the System of Equations . . . . . . . . . . . . . . 50

2.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

3 Time Domain Modelling of a BLDC Motor by Coupled Systems 53

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.2 Modelling Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.3 Mathematical Model of the BLDC Motor . . . . . . . . . . . . . . . 55

3.3.1 Electromagnetic field modelling . . . . . . . . . . . . . . . . 57

3.3.1.1 Modelling of eddy current effect on stator lamination 58

3.3.2 Modelling of electric circuit . . . . . . . . . . . . . . . . . . 61

3.3.2.1 Determination of DC winding resistance and Back-

emf . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.3.2.2 End winding inductance . . . . . . . . . . . . . . . 65

3.3.3 Modelling of the rotor movement equation . . . . . . . . . . 67

3.3.3.1 Consideration of load torque . . . . . . . . . . . . . 68

3.3.3.2 Determination of rotor inertia . . . . . . . . . . . . 69

3.4 Mesh Generation and Rotation . . . . . . . . . . . . . . . . . . . . 71

3.5 Finite Element Formulation in Time Domain . . . . . . . . . . . . . 77

 iv

3.5.1 Galerkin’s formulation of the electromagnetic field equation

in iron core . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

3.5.2 Galerkin’s formulation of field equation in the stator conduc-

tor area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.5.3 The stator circuit equation in Galerkin’s form . . . . . . . . 80

3.6 Time Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

3.6.1 Time discretization of the FEM equation in iron core . . . . 81

3.6.2 Time discretization of the FEM equation in stator conductor

area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

3.6.3 Time discretization of the stator circuit equation . . . . . . 82

3.6.4 Time discretization of the motion equation . . . . . . . . . . 82

3.7 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.8 Coupling the Rotor Movement with the FEM . . . . . . . . . . . . 84

3.9 Solving the Global System of Equation . . . . . . . . . . . . . . . . 85

3.9.1 ICCG algorithm for solving the algebraic equations . . . . . 86

3.10 Determination of Time Step Size for Time Stepping FEM . . . . . . 87

3.11 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

4 Experimental Implementation of the DSP Based BLDC Motor

Drive System 93

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

4.2 Hardware Implementation . . . . . . . . . . . . . . . . . . . . . . . 94

4.2.1 The Variable DC supply . . . . . . . . . . . . . . . . . . . . 94

4.2.2 The voltage source inverter . . . . . . . . . . . . . . . . . . 95

4.2.3 Spindle motor . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4.2.4 Incremental encoder . . . . . . . . . . . . . . . . . . . . . . 98

4.2.5 DS1104 controller board . . . . . . . . . . . . . . . . . . . . 99

4.3 Software Implementation . . . . . . . . . . . . . . . . . . . . . . . . 102

 v

4.4 Measuring Motor Performances . . . . . . . . . . . . . . . . . . . . 104

4.4.1 Rotor position sensing and switching sequence detecting . . 104

4.4.2 Measuring back-emf . . . . . . . . . . . . . . . . . . . . . . 104

4.4.3 Measuring stator current . . . . . . . . . . . . . . . . . . . . 105

4.4.4 Measuring motor speed . . . . . . . . . . . . . . . . . . . . . 105

5 Performance Analysis of the BLDC Motor 109

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

5.2 Steady State Analysis of the BLDC Motor . . . . . . . . . . . . . . 109

5.2.1 Mesh generation . . . . . . . . . . . . . . . . . . . . . . . . 110

5.2.2 Pre-computation using two dimensional magneto-static FEM 110

5.2.3 Computation in time domain by time stepping FEM . . . . 111

5.2.4 Post processing . . . . . . . . . . . . . . . . . . . . . . . . . 112

5.3 Evaluation of Steady State Performances . . . . . . . . . . . . . . . 112

5.3.1 Calculation of stator current . . . . . . . . . . . . . . . . . . 112

5.3.2 Computation of electromagnetic force and torque . . . . . . 113

5.3.3 Determination of torque-speed characteristics . . . . . . . . 119

5.3.4 Computation of cogging torque . . . . . . . . . . . . . . . . 122

5.3.5 Calculation of back-emf . . . . . . . . . . . . . . . . . . . . 123

5.4 Performance Evaluation with and without the Time Steps Adjust-

ment Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

5.5 Transient Analysis of the BLDC Motor . . . . . . . . . . . . . . . . 125

5.5.1 Step voltage variation . . . . . . . . . . . . . . . . . . . . . 127

5.5.2 Step change variation in mechanical load torque . . . . . . . 132

5.5.3 Locked rotor condition . . . . . . . . . . . . . . . . . . . . . 134

5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

6 Application Characteristics of BLDC Motors for Hard Disk Drives137

 vi

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

6.2 Coupling with the Control Loop . . . . . . . . . . . . . . . . . . . . 139

6.3 Analysis of the Starting Process of a HDD Spindle Motor . . . . . . 141

6.3.1 Motor starting without drive limits . . . . . . . . . . . . . . 141

6.3.2 Motor starting with current limits . . . . . . . . . . . . . . . 150

6.3.3 Motor starting with speed limit . . . . . . . . . . . . . . . . 153

6.4 Computational Analysis of the Run-up

Performances of a HDD Spindle Motor . . . . . . . . . . . . . . . . 155

6.4.1 Case I: Motor runs freely under various

stator phase supply voltages . . . . . . . . . . . . . . . . . . 155

6.4.2 Case II: Motor running with current limiter . . . . . . . . . 156

6.4.3 Case III: Motor running with voltage adjusting scheme . . . 160

6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

7 Discussions and Conclusions 165

Bibliography 170

List of Publications 185

A Motor Specification 187

B Newton Raphson Algorithm 188

C Cubic Spline Interpolation 191

D Demagnetization Curve for Permanent Magnet 193

E Specifications of Inverter Circuit Components 194

E.1 MOSFET IRF620 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

E.2 IR2110 high side and low side driver . . . . . . . . . . . . . . . . . 196

 vii

F Specifications of Incremental Encoder 199

G Program Structure for Steady State Analysis 202

Summary

This thesis deals with the modeling, simulation and performance analysis of the

brushless permanent magnet DC (BLDC) motors using numerical methods. The

primary objective is to develop efficient and practical procedures based on numer-

ical techniques to analyze the steady state and dynamic performances of BLDC

motors.

Dynamic model of the BLDC motor is developed using time stepping finite

element method. In this model, nonlinear electromagnetic field, circuit equations

and motion equations are formulated in time domain and solved simultaneously

in each time steps. Due to the direct coupling of the transient electromagnetic

field, circuit and motion, the solutions can take into account the eddy current

effect, the saturation effect, the rotor movement, the non-sinusoidal quantities and

high order harmonics of the electromagnetic fields which are very difficult to include

using analytical approaches and traditional finite element method (FEM). Proposed

dynamic model is used to investigate the transient analysis of the BLDC motor at

step voltage variation, load torque changing and locked rotor condition.

The analysis of the steady state performance of nonlinear electromagnetic

systems using time stepping FEM requires very long computational times. An

improved steady state model is proposed using time stepping FEM combined with

two dimensional FEM. In this model, current fed two dimensional FEM is used as

a pre-computation stage for the time stepping solver. Using the proposed steady

state model, the transient solver can be started with initial conditions quite close

viii
 ix

to the steady state solution and it can reduce the time spent in reaching a steady

state solutions. In addition, the non-sinusoidal quantities, high-order harmonic and

rotor motion which are very difficult to take into account in the traditional steady

state analysis using the FEM can be included. Steady-state model is used for the

calculation of steady-state current, cogging torque and back-emf in time domain

and determination of torque-speed characteristics of the BLDC motor.

BLDC motors cannot work without the electronic controllers. In order to

analyze the motor with a controller as an actual system, a new approach to couple

the time stepping FEM with closed-loop control structure is implemented. Cas-

caded speed and current hysteresis control loop structures is used. By coupling the

control loop features with the time stepping FEM, the stator windings could be

fed with the actual input voltages to the time stepping FEM model. In addition,

motor operations under transient conditions can be controlled instantaneously as

an actual motor-controller system. Using this new scheme, application characteris-

tics of the HDD spindle motors are investigated. Important features of the spindle

motor at starting such as spin-time, starting torque and starting current under no

load and loaded conditions are analyzed. Computational analysis of the run-up

performance of a spindle motor is investigate. It is found that the proposed model

works satisfactorily when it is used to simulate the motor drive under real transient

conditions with voltage, current and speed limits.

In order to determine the accuracy and validation of the proposed dynamic

and steady state model, DSP based BLDC motor test stand is implemented. Simple

and reliable methods of motor performance measurements are presented. A new

approach for detecting the motor starting sequences for controller is developed.

The good agreement of the computational results with the experimental results

indicates that developed numerical models are useful and applicable to analyze the

static and dynamic behaviours of the BLDC motor.

List of Figures

1.1 Typical configurations of a DC motor and a PMDC commutator

motor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Typical configuration of a PMSM . . . . . . . . . . . . . . . . . . . 3

1.3 Multiphase inner rotor BLDC motor configuration . . . . . . . . . . 5

1.4 Exterior rotor BLDC motor configuration . . . . . . . . . . . . . . . 6

1.5 Axial field type BLDC motor configuration . . . . . . . . . . . . . . 6

1.6 Basic components of the BLDC motor drive . . . . . . . . . . . . . 7

1.7 Inverter-fed armature circuits of BLDC motors . . . . . . . . . . . . 8

1.8 Transistor switching sequences and corresponding current waveforms 8

1.9 Demagnetization curve of PM . . . . . . . . . . . . . . . . . . . . . 13

1.10 Characteristic of permanent magnet materials . . . . . . . . . . . . 13

2.1 BLDC motor configuration . . . . . . . . . . . . . . . . . . . . . . . 28

2.2 Characteristic of a permanent magnet material . . . . . . . . . . . . 31

2.3 Typical finite elements (a) One-dimensional (b) Two-dimensional (c)

Three-dimensional . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.4 A triangular element . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.5 FEM mesh of the exterior rotor BLDC motor . . . . . . . . . . . . 35

2.6 Problem domain containing three triangular elements . . . . . . . . 42

2.7 Dirichlet boundary condition for the BLDC motor . . . . . . . . . . 45

2.8 Periodic boundary condition applied to the BLDC motor . . . . . . 45

2.9 Magnetization curve for ferromagnetic material . . . . . . . . . . . 49

x
 xi

3.1 Mechanical structure of the motor . . . . . . . . . . . . . . . . . . . 56

3.2 BLDC motor configuration and power electronic circuit . . . . . . . 56

3.3 Typical input voltage waveforms with respective electrical degrees . 57

3.4 Equivalent circuit for flux flow through the laminations . . . . . . . 60

3.5 Equivalent circuit for flux flow across the thickness of laminations . 60

3.6 The inverter circuit where current flow from phase A to B . . . . . 62

3.7 Circuit representation of a phase winding . . . . . . . . . . . . . . . 62

3.8 Motor geometry for distributed winding inductance calculation . . . 67

3.9 Motor geometry for concentrated winding inductance calculation . . 67

3.10 Rotor part of the BLDC motor . . . . . . . . . . . . . . . . . . . . 70

3.11 Basic stator mesh in including air gap parts . . . . . . . . . . . . . 74

3.12 Basic rotor mesh including air gap part . . . . . . . . . . . . . . . . 75

3.13 FEM mesh at air gap . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3.14 FEM mesh before rotor rotation (1899 nodes, 2828 elements) . . . . 76

3.15 FEM mesh after rotation 1000 steps . . . . . . . . . . . . . . . . . . 76

3.16 Block Diagram of the time stepping solver . . . . . . . . . . . . . . 92

4.1 Photograph of hardware set up in the Laboratory . . . . . . . . . . 94

4.2 Schematic diagram of the hardware equipments . . . . . . . . . . . 95

4.3 Circuit diagram of the voltage source inverter . . . . . . . . . . . . 97

4.4 Hardware set up for motor and encoder . . . . . . . . . . . . . . . . 98

4.5 Overview of DS1104 Feature . . . . . . . . . . . . . . . . . . . . . . 100

4.6 Wye connected stator windings . . . . . . . . . . . . . . . . . . . . 102

4.7 Typical input voltage waveform and switching states . . . . . . . . 103

4.8 Flow chart of switching sequences control program . . . . . . . . . . 107

4.9 Main control program and interrupt service routing flow charts . . . 108

5.1 Input voltage waveform against time . . . . . . . . . . . . . . . . . 113

 xii

5.2 Computed stator current waveforms at no load condition . . . . . . 114

5.3 Experimental and computational results of stator current . . . . . . 114

5.4 Calculated flux plot at static position . . . . . . . . . . . . . . . . . 118

5.5 Calculated flux plot after rotor is rotated 1000 steps . . . . . . . . . 118

5.6 Electromagnetic torques at no load and loaded conditions . . . . . . 119

5.7 Torque-speed curve of the motor . . . . . . . . . . . . . . . . . . . . 121

5.8 Current and torque relationship . . . . . . . . . . . . . . . . . . . . 121

5.9 Cogging torque profiles for 8p 12s spindle motor with different mag-

net strengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

5.10 Calculated back-emf with corresponding rotor angle . . . . . . . . . 124

5.11 Simulated cogging torque with and without step size adjustment

scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

5.12 Simulated load torque with and without step size adjustment scheme 126

5.13 Step voltage change . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

5.14 Speed response during step voltage change . . . . . . . . . . . . . . 127

5.15 Simulated back-emf during step voltage change . . . . . . . . . . . 128

5.16 Stator current transient during step voltage change . . . . . . . . . 128

5.17 Developed torque during step voltage change . . . . . . . . . . . . . 129

5.18 Input simulated step voltage waveform and output back-emf waveform131

5.19 Calculated transient speed and stator current waveform . . . . . . . 131

5.20 Measured transient speed and stator current waveform . . . . . . . 132

5.21 Speed and back-emf transients due to an increase in load torque . . 133

5.22 Current and torque transient due to an increase in load torque . . . 133

5.23 Speed and back-emf transient at locked rotor conditions . . . . . . . 135

5.24 Current and torque transient at locked rotor conditions . . . . . . . 135

6.1 Control system block diagram . . . . . . . . . . . . . . . . . . . . . 141

6.2 Input voltage waveform against time . . . . . . . . . . . . . . . . . 142

 xiii

6.3 Motor starting speed without limits . . . . . . . . . . . . . . . . . . 142

6.4 Back-emf waveform without limits . . . . . . . . . . . . . . . . . . . 142

6.5 Starting current without limits . . . . . . . . . . . . . . . . . . . . . 143

6.6 Starting torque without limits . . . . . . . . . . . . . . . . . . . . . 143

6.7 Motor loaded with one platter . . . . . . . . . . . . . . . . . . . . . 145

6.8 Speed against time waveform when motor is loaded with one platter 145

6.9 Back-emf waveform when motor is loaded with one platter . . . . . 146

6.10 Stator current waveform when motor is loaded with one platter . . 146

6.11 Torque against time graph where motor is loaded with one platter . 146

6.12 Motor loaded with two platters . . . . . . . . . . . . . . . . . . . . 147

6.13 Speed against time waveform when motor is loaded with two platters 148

6.14 Back-emf waveform when motor is loaded with two platters . . . . . 148

6.15 Stator current waveform when motor is loaded with two platters . . 148

6.16 Torque against time graph where motor is loaded with two platters 149

6.17 Calculated motor speed under no load and loaded conditions . . . . 149

6.18 Measured motor speed under no load and loaded conditions . . . . 149

6.19 Voltage comes from the hysteresis controller . . . . . . . . . . . . . 151

6.20 Motor starting current with 1.5A current limit . . . . . . . . . . . . 151

6.21 Motor back-emf waveform when current is limited at 1.5A . . . . . 151

6.22 Speed profile with 1.5A current limit . . . . . . . . . . . . . . . . . 152

6.23 Starting torque profile with 1.5A current limit . . . . . . . . . . . . 152

6.24 Starting speed profile with speed limit . . . . . . . . . . . . . . . . 153

6.25 Motor supply voltage . . . . . . . . . . . . . . . . . . . . . . . . . . 154

6.26 Motor back-emf profile with speed limit . . . . . . . . . . . . . . . . 154

6.27 Stator current profile with speed limit . . . . . . . . . . . . . . . . . 154

6.28 Speed with speed limit and without limits . . . . . . . . . . . . . . 155

 xiv

6.29 Motor speed vs. spin-up time with different supply stator phase

voltages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

6.30 Starting current profiles with different supply stator phase voltages 157

6.31 Speed vs. spin-up time with different supply voltages where current

is limited at 1.5A . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

6.32 Supply voltage vs. spin-up time with and without current limit . . 159

6.33 Comparison of power consumption with and without current limit

at starting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

6.34 Motor transient responses without voltage adjusting scheme . . . . 162

6.35 Motor transient responses with voltage adjusting scheme . . . . . . 163

B.1 Relationship between f (x) and B . . . . . . . . . . . . . . . . . . . 189

B.2 Newton Raphson procedure . . . . . . . . . . . . . . . . . . . . . . 190

B.3 Effect of non-monotonic function on Newton’s method . . . . . . . 190

D.1 Demagnetization curve for bonded NdFeB magnet . . . . . . . . . . 193

E.1 Date sheets of absolute maximum ratings . . . . . . . . . . . . . . . 194

E.2 Thermal and electrical characteristics sheet (1) . . . . . . . . . . . . 195

E.3 Thermal and electrical characteristics sheet (2) . . . . . . . . . . . . 196

E.4 Typical connection diagram . . . . . . . . . . . . . . . . . . . . . . 197

E.5 Functional block diagram . . . . . . . . . . . . . . . . . . . . . . . . 197

E.6 Absolute maximum ratings . . . . . . . . . . . . . . . . . . . . . . . 198

F.1 Photograph of Scancon incremental encoder . . . . . . . . . . . . . 199

F.2 Electrical specifications . . . . . . . . . . . . . . . . . . . . . . . . . 200

F.3 Mechanical specifications . . . . . . . . . . . . . . . . . . . . . . . . 201

List of Tables

2.1 Element contribution . . . . . . . . . . . . . . . . . . . . . . . . . . 42

3.1 Friction coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.2 Material densities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3.3 Densities of permanent magnet . . . . . . . . . . . . . . . . . . . . 71

4.1 Motor specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

4.2 Possible switching sequences . . . . . . . . . . . . . . . . . . . . . . 103

6.1 Power consumptions with different supply phase voltages at no load

condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

6.2 Motor power consumptions with current limits . . . . . . . . . . . . 158

6.3 Spin-time with and without current limits . . . . . . . . . . . . . . 160

6.4 Comparison of power consumption with and without current limits 160

A.1 Motor specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

xv
List of Symbols

H magnetic field intensity

B magnetic flux density

A magnetic vector potential

J current density

Br the remanence flux density

ν magnetic relucitvity

σ electric conductivity

µ0 magnetic permeability in the free space

µ magnetic permeability

Hc the coercivity

Te electromagnetic torque

TL load torque

D, damping coffcient

ω angular velocity

∆t, time step length

θm rotor angle in mechanical degree

xvi
 xvii

E electric field intensity

is stator phase current

S total cross-sectional area of

the stator winding one turn per one coil side

t time

V stator phase voltage

R total stator phase resistance

e backemf

is stator phase current

Lσ end winding inductance

Ω+ , Ω− total cross-section area of ” go “ and

” return “ side of stator windings

Chapter 1

Introduction

1.1 Permanent Magnet Motors

Permanent magnets (PM) have been used in electrical machine applications almost

from the beginning of the development of these machines as replacements for wound

field excitation systems. The availability of high-energy permanent magnets and

advances in power electronics are leading to a large diffusion of permanent magnet

machines in a variety of applications [1]-[2]. In general, permanent magnet motors

are broadly classified into:

• Brushed DC motor (or) PMDC commutator motor: The construction

of a permanent magnet DC motor(PMDC) is similar to a DC conventional

motor with the electromagnetic excitation system replaced by permanent

magnets. A PMDC commutator motor can be compared with a separately

excited DC motor. The only difference is in the excitation flux in the airgap:

for PMDC commutator motor excitation flux is constant whilst for a sepa-

rately excited DC motor’s excitation flux can be controlled. The structures

of a conventional DC motor and a PMDC commutator motor are shown in

Fig. 1.1.

• Brushless permanent magnet motor: Brushless permanent magnet mo-

1
 2

tor falls into two principal classes: Bushless DC motor (BLDC)and Perma-

nent magnet synchronous motor (PMSM).

The PMSM owes its origin to the replacement of the exciter of the wound

synchronous machine with a permanent magnet. PMSMs are fed with three

phase currents in sinusoidal shape and operate on the principle of a magnetic

rotating field. All phase windings conduct current at a time with phase

differences. The structures of a PMSM is shown in Fig. 1.2.

The BLDC owes its origin to an attempt to invert the brush DC machine

to remove the need for the commutator and brush gear. BLDC is fed with

rectangular or trapezoidal shape current waveforms shifted by 120 electrical

degree one from another. Electronic commutation is done by the rotor po-

sition sensors and electronic controller where armature current is precisely

synchronized with the rotor frequency and instantaneous position. Only two

phases are conducting at any given instant of time. Basic configuration and

characteristic of BLDC motor are presented more in detail at next section.

Figure 1.1: Typical configurations of a DC motor and a PMDC commutator motor

 3

Figure 1.2: Typical configuration of a PMSM

1.2 Brushless Permanent Magnet DC Motors

Until recently, conventional DC motors have been the dominant drive system be-

cause they provide easily controlled motor speed over a wide range, rapid accel-

eration and deceleration, convenient control of position, and lower product cost.

However, technical advances in permanent magnet materials, in high power semi-

conductor transistor technology, and in various rotors position-sensing devices have

made using BLDC motor a viable alternative. The developments and applications

of the BLDC motors have been greatly accelerated by improvements in permanent

magnet materials, especially rare-earth magnets. Brushless motors are smaller,

lighter and have higher efficiency and power density compared to traditional DC

motors because of lack of field windings, commutators and brushes. Additionally,

the brushless design offers increased motor speed range because the motor speed is

not limited by the arcing at the commutator as in brushed DC motors. Therefore,

BLDC motors are highly demanded in clean, explosive environments such as aero-

nautics, robotics, electric vehicles, food and chemical industries, and have a wide

variety of applications in the area of HDD drives, servo drives and variable speed

drives.
 4

1.2.1 Basic Configurations of BLDC motors

There are several different configurations of BLDC motors for different applications.

Three basic configurations of the permanent magnet BLDC motor are inner rotor,

outer rotor and axial gap disc designs, with many different winding pattern as well

as many different pole configurations [3]. The magnets may be in strips, arcs or

discs of various shapes and they may or may not be pre-magnetized.

Inner rotor motor configuration is nearly the same as the classical AC synchro-

nous motor or the induction motor. The stator is similar to that of the three-phase

induction motor. The advantage of interior type is its high torque/inertia ratio.

Hence it is widely used in servo systems, requires rapid acceleration and decelera-

tion of the load and the torque/inertia ratio should be as high as possible. Most

inner rotor motors have multiple phases in an effort to reduce the starting prob-

lems associated with single phase motors. The stators may have salient pole or

distributed windings. Fig. 1.3 illustrates a three phase four salient pole inner rotor

type BLDC motor.

If the application requires constant speed at medium to high speed it may

take more sense to use an exterior-rotor configuration with the rotating member on

the outside of the wound stator. This type is used in fans, blowers and computer

hard disk drive spindle motor. Fig. 1.4 shows the cross-section of a typical motor of

exterior- rotor type. The most important application for the exterior-rotor motor

is the spindle motor used in computer hard disk drives. This application requires

a very uniform and constant speed and the high inertia of the exterior rotor is an

advantage in achieving this.

There are other applications such as record players, VCR players, CD players

and floppy disc drives for computers which have a different set of requirements.

These type of motor should be rotated at relatively low speed. It has been common

to design axial-gap or pancake motors for many of these applications. Fig. 1.5 is
 5

axial field type BLDC motor. The main advantages of these motors are their low

cost, their flat shape and smooth rotation with zero cogging.

The choice of motor type is the most fundamental design decision, because

of the relatively high cost of magnets, together with issues related to packaging,

magnet retention, and winding. However, to date, it has not been determined

which configuration should be used to maximize the power density, efficiency and

quietness of a motor [4]. To thoroughly investigate permanent magnet BLDC

motor technology, it is necessary to study the relative merits of each configuration

in terms of the power density, efficiency and noise/vibration levels.

Figure 1.3: Multiphase inner rotor BLDC motor configuration

1.2.2 Characteristics of BLDC Motors

A BLDC motor cannot work without the electronic controller. The terminal volt-

ages on the windings of each phase are controlled by the power electronic switches.

The phase windings are energized in sequence by the switching elements in the in-

verter which are controlled by shaft position sensors. Thus stator magnetomotive

force (mmf) runs ahead rotor mmf keeping a constant angular displacement. The

basic components of the BLDC motor drive system are: a rectifier, an inverter, a
 6

Figure 1.4: Exterior rotor BLDC motor configuration

Figure 1.5: Axial field type BLDC motor configuration

 7

PM motor, rotor position sensors and a controller as shown in Fig. 1.6. Rectifier :

Figure 1.6: Basic components of the BLDC motor drive

The electrical energy can be a DC source, such as a battery, or an alternating

current AC source. Rectifier converts the AC line voltage into DC bus voltage.

Inverter : Inverter includes the power semiconductor switches and their cur-

rent sensors and protection circuitry. The inverter circuit diagram is shown in

Fig. 1.7 for the wye and the delta connections. In square wave operation, there are

normally two transistors conducting at any one time. Transistors in the inverter

receive conduction commands from a system of logic which is synchronized with

the rotor position sensors [3]. Fig. 1.8 shows the sequence of switching transistors

for the corresponding current wave form for wye connection.

Rotor position sensors : Position sensors detect the position of the ro-

tating magnets and send logic codes to a commutation decoder which activates the

firing circuits of semiconductor switches feeding power to the stator winding of the

drive motor. A unidirectional torque is produced via the interaction between the
 8

Figure 1.7: Inverter-fed armature circuits of BLDC motors

Figure 1.8: Transistor switching sequences and corresponding current waveforms

 9

permanent magnets and currents flowing through the winding. Generally, rotor

position sensing in commercial brushless DC motor is done by resolvers, optical

position sensors and hall-effect transducer. Hall sensors are mostly used BLDC

motor drives.

The various shaft position sensors seem straightforward enough in themselves,

and yet in the marketplace there has been concern with the use of brushless motors

in many applications because of the need for these devices. On the other hand,

it is widely stated that there are no brushes or commutator to affect reliability.

However, these must be replaced by a shaft position transducer, with additional

electronic circuitry and an interconnecting cable. These components inevitably add

to the cost, and may decrease the reliability because they are relatively fragile and,

unless they are properly protected, they may be susceptible to damage or mal-

operation for high temperatures, dusts, oil, vibration and shock, etc., and even

from electrical interference. It is not surprising that there has been much effort in

recent years to eliminate the need for the shaft position transducer. This is called

sensorless control. Sensorless motors are applicable in refrigerators, air conditioners

and especially in hard disk drives [5].

PM motor: Different configurations and magnetic excitations for particular

applications are as mentioned in the section 1.2.1.

Controller: The main task of the controller is to decode shaft position

sensor’s input data, to control the supplied voltage to inverter, to operate control

loop such as speed and position. Actually, the power circuit of the electronic

controller is a switchmode circuit. The only means of controlling such circuits

is to control the timing of the gate signals that turn the power transistors on

and off. These low-level timings must be controlled to meet a set of high-level

functional requirements. The state of the art for BLDC motors calls for a single

chip controller that could be used for low-cost applications. Most of the recent
 10

DSP controller board contains all of the active functions required to implement a

full featured open-loop, three or four phases motor control system.

1.3 Magnetic Materials

Good knowledge of materials used in the construction of a permanent magnet

motor is important in the design and operating conditions of the machine. Different

material qualities offer various design possibilities and application choices [6]. In

general, magnetic circuit should be optimized in accordance with the characteristics

of the materials chosen. Generally, two groups of materials are used in the magnetic

circuit:

• Hard magnetic materials (permanent magnet) and

• Soft magnetic materials

The magnetic flux density versus magnetic field strength curve, the hysteresis loop,

is used in the characterization of these materials. A soft magnetic material is one

in which the hysteresis loops are narrow. For narrow loops, the normal magnetiza-

tion curve is a good approximation and is often used to characterize the material.

Another type of material that is characterized by a broad hysteresis loop is called

hard magnetic material.

1.3.1 Hard magnetic materials (Permanent magnets)

Permanent magnets can produce magnetic flux in an airgap with no dissipation of

electric power. The basic operational characteristic of permanent magnet is given

by its demagnetization curve in the second quadrant of B − H plane as shown in

Fig. 1.9. When a permanent magnet has been magnetized, it remains magnetized

even the applied field intensity is decreased to zero. The magnetic field density
 11

at this point is called the remanence flux density Br . If a reverse magnetic field

intensity is applied, the flux density decreases. If the reverse magnetic field is

large enough, the flux density become zero. The field intensity at this value is

called the coercivity Hc . The operating point of the permanent magnet is the

intersection point of a B − H curve of the external magnetic circuit (load line) and

the demagnetization curve of a permanent magnet. The operation point moves

along the demagnetization curve with changes in the external magnetic field. The

absolute value of the product of the flux density B and the field intensity H at each

point along the demagnetization curve can be represented by the energy product

[7] and this quantity is one of the indexes of the strength of the permanent magnet.

For the modern high energy permanent magnets, the demagnetisation curve

at ambient temperature is linear. It only shows a noticeable bent at higher temper-

atures. The result is that the operating point of the modern rare-earth permanent

magnets can easily be designed to be within the linear area of the demagnetisation

curve. This is one of the major advantages of the modern high energy permanent

magnets over the conventional permanent magnets in the designs for engineering

applications. The effect of armature field on the permanent magnets is either

magnetizing or demagnetizing. Since the demagnetizing curve of some permanent

magnets has a knee, there is a limit to the maximal allowable armature field.

Proper selection of magnetic materials is important from both economics as

well as performance considerations. There are basically three different types of

permanent magnet which are used in PM motors:

• Alnico

• Ceramics (Ferrites) and

• Rare-earth materials, i.e Sm-Co and Nd-Fe-B.

Typical demagnetisation characteristics of these three types is shown in Fig. 1.10.

 12

Alnico- The main advantages of Alnico are its high magnetic remanent flux

density, Br and low temperature coefficients. These advantages allow quite a high

air gap flux density and high operating temperatures [2]. But coercive force is very

low and the demagnetization curve is extremely non-linear. When the coercivity

is low and two opposing magnetic poles locate at a close distance, the poles can

weaken each other. Therefore, an Alnico magnet is used after being magnetized

lengthwise. It is very easy not only to magnetize but also to demagnetize this

magnet. Alnico dominated the PM industry from the mid 1940s to about 1970

when ferrites became the most widely used materials.

Ferrite- A ferrite has a higher coercive force than that of Alnico, but at the

same time has a lower remanent magnetic flux density. Temperature coefficients

are relatively high. The maximum service temperature is 400 degree C. The main

advantages of ferrites are their low cost and very high electric resistance, which

means no eddy-current losses in the PM volume. Ferrite magnets are most eco-

nomical in fractional horsepower motors and may show an economic advantage over

Alnico up to about 7.5kW.

Rare- earth permanent magnet materials- The first generation of rare-earth

magnet is based on the composition SmCo5 and invented in the 1960s has been

commercially produced since the early 1970s. Smco5 has the advantage of high

remanent flux density, high coercive force, high-energy product, linear demagne-

tization curve and low temperature coefficient. It is well suited to build motors

with low volume and consequently high specific power and low moment of inertia.

NdFeB is another type of rare earth magnets. It has the highest B*H product

among all the permanent magnet materials currently available. The maximum ser-

vice temperature is 150 degree C. This indicates that the use of NdFeB will result

in the smallest and the most powerful motor design.

 13

Figure 1.9: Demagnetization curve of PM

Figure 1.10: Characteristic of permanent magnet materials

 14

1.3.2 Soft magnetic materials

Desirable properties of the soft magnetic material that can be used in PM motors

are high saturation flux density, high permeability, low coercive field strength and

low specific conductivity. Low coercivity is required to minimize hysteresis losses;

generally, in brushless motors, the armature core experiences an alternating flux as

well as high-frequency flux variation due to PWM. In high frequency applications

the eddy currents are minimized by using thinner laminations and high-resistive

steels. There have been great improvements in the quality of electrical steels over

the last 20 years. This has been made possible by improved manufacturing tech-

niques and a better understanding of factors which control magnetic properties.

For effective use of soft magnetic material, the desire is to have the operating point

in the middle of the magnetization curve. Saturation of the soft magnetic mate-

rial should be avoided, otherwise, an appreciable percentage of the useful energy

provided by the permanent magnet for use in air-gap would be lost.

1.4 Computational Analysis of Electrical Machines

Prediction of electrical machine performance is necessary for the evaluation of ma-

chine designs. Various methods of analysis have been employed to model the electri-

cal machines and to predict their performance accurately at the design stage. Gen-

erally, experimental methods, analytical methods and numerical methods

are used out to predict the performance and design of the electrical machines.

Experimental methods require building hardware prototypes or scaled

models and results obtained from tests and conducted on these models are used to

predict the performance of the actual devices and also to enable hardware changes

to be made in order to meet specifications. These methods prove to be expensive,

cumbersome, and time consuming. Further, often only terminal quantities could
 15

be measured owing to the non-availability of sophisticated sensors and instrumen-

tation.

Analytical methods are computationally quick and simple to use. They

can give a good view over the influence of machine parameters on the performances

for machines with relative ease. However, analytical methods are limited to tech-

niques applicable to linear isotropic media. Only constant values of permeability

and conductivity for magneto static and eddy current problems and constant di-

electric permittivity for electrostatic problems can be used. Several simplifications

of the formulation and associated boundary conditions are required in view of the

limitations on available solution techniques. Modelling of non-linearity is not pos-

sible except for simple problems where transformation methods could be used.

Due to the advent of digital computer and subsequent advances in computing

power and storage devices, it has become practical to use numerical methods

to compute the performance of the electrical machines. For complex geometrical

shapes, with varying materials characteristics and often mixed boundary condi-

tions, numerical methods offer the best and often the most economical solution. In

order to obtain a good method for various engineering problems, many numerical

methods have been developed such as finite difference method, boundary element

method, moment method and finite element method (FEM)[8]. Among these meth-

ods, FEM is widely used in performance analysis of the electrical machines because

it is feasible for a wide rage of application and can be used for solving both linear

and nonlinear problems, static and dynamic problems, 2D and 3D problems.

1.4.1 Analysis of electrical machines using FEM

Electrical machines have their behaviours governed by the electromagnetic fields

which flow in them; the fields, in turn, obey Maxwell’s equations. The solution of

many practical electromagnetic field problems can only be undertaken by applying

 16

numerical methods. Before such a solution can be undertaken, it is important that

a correct mathematical model be established for the problem considered. In the

analysis of electrical machines, different types FEM model are applied based on

the various electromagnetic field behaviour of the problem.

In general, electromagnetic fields in low frequency application can be iden-

tified as: time independent static and quasi-static. Static field is a field which is

excited by DC currents or voltages. If a DC voltage is applied, the DC current

depends on the voltage and DC resistance of the circuit. No induced effects such

as eddy currents or induced voltages are present in the model. In a static model,

the flux is constant in time and thus the reluctivities are constant as well [9]. A

quasi-static field can be divided into time invariant and time-varying fields with or

without involving the eddy currents. In the time invariant cases, the field solution

does not depend on the time-derivative term in the differential equations. It can

be regarded as a static field for a particular instant of time. The calculation of the

field is performed for a certain instant of time and therefore the flux density and the

resulting inductances are calculated for this specific instant of time. In such cases,

magneto-static FEM is used to analyze the behaviour of electromagnetic field.

In the time varying quasi-static field, the eddy currents are involved and

have to be considered, a time derivative term appears in the differential equations.

The varying field generates induced voltages and currents. The eddy currents also

influences the field. Each form of time variation can be modelled either in the

time domain or in the frequency domain. When the field is periodic with one or

a limited number of frequencies, it is more efficient to perform a field calculation

in the frequency domain. For the case in which the field is oscillating at one

frequency and assume all field quantities,imposed sources and boundary conditions

vary sinusoidally with time, time harmonic complex eddy current FEM is used.

Strictly speaking, in the complex eddy current FEM, all material properties are
 17

assumed to be linear and constant in time. However, for non-linear material, an

effective material characteristic could be used [10]-[12]. This characteristic gives a

sinusoidal averaged value in terms of rms value of the field quantity. Indeed, the

non-linear time harmonic analysis seeks to include the effects of nonlinearities like

saturation and hysteresis on the fundamental of the response, while ignoring higher

harmonic content[13].

In the time harmonic complex eddy current model, the simulation based on

the field concepts. The behaviour of the machine is determined directly by the

distribution of magnetic fields and current density. Rotor rotation effect cannot

be included hence motion equations could not be taken into account in the sys-

tem of equations. Non-sinusoidal quantities also cannot be included. Therefore,

this model can apply in the steady state analysis of machines. Many researchers

have studied the performance of BLDC motors using complex eddy current FEM

[14]-[23]. The great majority of machines are used in applications when impor-

tant feature is the steady state performances. Although almost all applications of

machines have definite steady state requirements, an increasing number of appli-

cations place great emphasis on dynamic performance, so that mathematical and

experimental techniques are needed to investigate dynamic features. It frequently

happens that a machine with excellent steady state characteristics cannot be used

because it is unsatisfactory in dynamic operation and vice versa.

In order to determine the transient behaviour of the electromagnetic devices

that are activated by a voltage or current source which may be time dependent, it is

needed to analyze in time domain. In this case, time stepping FEM is appropriate

to use in the analysis of electrical machines [24]-[26]. In the time stepping FEM

analysis, the equations for the stator and rotor fields are written in their own

coordinate systems. The solutions of the two field equations are matched with

each other in the air gap. The rotor rotation is accomplished by rotating the FEM
 18

meshes of the rotor. The circuit equations which describe the supply and control

system and motion equations are directly coupled together to the field equations.

Both the transient performance and the steady state performance of the machine in

time domain can be studied accurately, particularly when non-sinusoidal quantities

and high-order space and time harmonics are to be considered.

Although three dimensional (3D) finite element analysis is required to ac-

curately model the 3D effect such as end winding effects, in practice these are

accounted for analytically, by appropriate modification of 2D finite element pre-

dicted parameter values [27]-[29]. Compared with 3D technique, the 2D FEM has

the advantages of simple mesh generation, short computing time and small com-

puter storage requirement especially in studying of transient conditions where rotor

movement is needed to include. Therefore, in practical applications, it is highly

desirable a 2D model.

1.5 Literature Review

Due to the advent of high energy permanent magnets and sophisticated electronic

controllers, brushless permanent magnet DC motor is getting more and more popu-

lar in a wide range of applications, which include machine tools, robotics, aerospace

generators, actuators, industrial drives and electric vehicles. The majority of work

has been done on the analytical design approach based on the equivalent circuits

model (a-b-c model), state space model and lump parameter model [30]-[39]. Nehl

at el [31] developed a-b-c phase variable model with the back-emf represented

as a Fourier series and P. Pillay [40] also developed a-b-c model to study both

steady state and transient characteristics of the BLDC motor. Y.P. Liu, D. Howe,

T.S. Birch and D.M.H. Matthews [41] proposed the dynamic model of the BLDC

motor with a-b-c reference frame. In these models, the motor voltage equation,
 19

current equations, and torque equation are written in the a-b-c reference frame.

The switching devices are also represented by circuit model. The main advan-

tage of this model is that no assumptions are required regarding constant speed

operation. However, the saturation effect, the eddy current effect and permanent

magnet characteristic cannot be taken into account in the system of solution. M

M Elmissiry and S Chari [42] investigates the dynamic performance of a class of

an axial-flux, permanent magnet brushless DC drive using state space model. In

this study, the motor dynamic equations are written on state-space form. In de-

riving the dynamic equations, the rotor flux density distribution is assumed to be

invariant with the operating conditions but material saturation effect is neglected.

Mathematical model of the drive is presented and two cases of dynamic operation;

motor start-up and the sudden change of the applied voltage are considered.

Based on the above research works, it can be concluded that although the ana-

lytical methods are fast and flexible in computation, they cannot be accurate in the

conditions that involve complex configuration and eddy current effect. Modelling

of nonlinearity was not possible except for simple problems where transformation

methods could be used. Due to the complexity of the BLDC motor geometries,

control techniques and eddy current phenomena in magnetic field of the motor, the

simulation of electromagnetic fields and their effects by numerical models is suitable

as an appropriate engineering tool. Among the numerical methods, finite element

method(FEM) is by far the most efficient and popular method and it has been

extensively used in performance analysis of the BLDC motors. Generally, research

works have been done by FEM modelling in electrical machines can be classified as:

magneto-static approach, frequency domain approach and time stepping approach.

The common parameters and performance indicators of the BLDC motor in steady

states such as calculation of winding inductances [43], cogging torque [44]- [49],

back-emf [50], iron losses [51]-[52] and prediction of the new motor design [53]- [57]
 20

have been done by using magneto-static FEM and the frequency domain FEM.

However, the precision of these two FEM models are limited by the concept of

the equivalent circuit. Non-sinusoidal quantities cannot be included. FEM mesh

cannot rotate according to the rotor positions; hence motion equations cannot be

taken into account in the system of equations. In addition, eddy current effect

cannot be included in magneto-static FEM model. Hence, it is concluded that

magneto-static FEM and complex eddy current FEM model are limited to use in

the steady state analysis of machines.

The advance in time-stepping finite element method in recent years, it is now

possible to overcome these difficulties by coupling the external circuit and rotor

movement with the internal electromagnetic field into the system of equations. The

time stepping FEM is finding increasing popularity in the analysis of induction

machines [58]- [64]. S. L. Ho and H. L. Li [65] used time-stepping FEM in the

modelling of permanent magnet synchronous machines. Jinyun Gan [66] proposed

a new PM brushless DC machine with a unique feature of flux regulation and

analyze the motor performance by using time stepping FEM method. Analysis of

the exterior rotor BLDC motor takes into account the eddy current effect in solid

rotor steel shell by time stepping FEM has been proposed by Seung-Chan Park at

el [67]. Transient analysis of a new outer-rotor BLDC motor at both normal and

flux weakening operations by using circuit-field-torque coupled time stepping finite

element method has been presented at [68]. It can be found that by using time

stepping FEM in machines modelling, the stator current, the load angle, torque

and force can be directly computed. Many assumptions such as neglecting material

non-linearities, eddy current effect, rotor motion and high order harmonics which

are essential when using equivalent circuit model, frequency domain FEM model,

are no longer necessary.

One of the difficulties in the modelling of the time-stepping FEM is developing

 21

the methods of rotor rotation. In the time stepping FEM model, the external circuit

equations and the internal field equations are needed to couple with the rotor

movement and solved simultaneously at each time steps. C. W. Chan and H. L. Li

[69] reported a multislice time stepping FEM model including rotor rotation. They

developed special technique required for the mesh generation for the movement

of the rotor mesh and stator mesh in the multislice model. B. I. Kwon and K.

I. Woo [70] proposed new moving mesh technique of rotor rotation for studying

the steady state and dynamic behaviour of the permanent magnet motors. In this

technique, pseudo-stationary approximation method is used to couple the motion

equations with circuit and field equations. S. J. Salon and M. J. Lee [71] suggested

moving band technique to study the induction motor analysis. They proposed

one moving band between stator and rotor mesh and when the rotor is rotated

according to the time step; only moving band meshes are needed to mesh again.

Chuntin Mi [72] established a new moving mesh technique to calculate the iron

losses of the synchronous motor in transient conditions. In establishing the meshes

for the analysis, the rotor is moved and positioned at each time step such that it

does not disturb the integrity of the mesh structure as it moves. The initial meshes

of the stator and the rotor are generated first. The air gap is divided into two

parts and the first half is for the stator and the other half is for rotor. A stator

mesh and rotor mesh share the same boundary at the middle of the air gap. The

inner stator circumference and outer rotor circumference of the air gap are divided

into equal steps so that their nodes coincide. To provide for rotor movement, the

time step has been chosen so that the angle or length of each step is equal to the

interval between two neighboring nodes along the mid air gap. As a consequence,

the time step cannot be chosen arbitrary. Step size is fixed throughout the whole

process. That is the main disadvantage of this technique.

S. L. Ho, W. N. Fu [73] reported simple and reliable moving mesh technique

 22

in the analysis of induction motor. The air gap is divided into three layers; two

layers belong to the stator mesh and the rest layer belongs to the rotor mesh. The

stator mesh and the rotor mesh are generated separately and connected by the

periodic boundary condition. When the rotor is rotated, only boundary conditions

and node numbers need to be changed. Stator mesh and rotor mesh are unchanged.

In this thesis, moving mesh technique developed by W. N. Fu is adopted in the

simulation of rotor rotation.

BLDC motors cannot work without the electronic controllers and their con-

trol methods are complicated compared with traditional DC motors. The phase

windings are energized in sequence by the switching elements in the inverter which

are controlled by shaft position sensors. The impressed voltages comprised of a

series of pulses of varying widths and the stator phase currents are non-sinusoidal.

High order harmonics in the currents and in the magnetic field will have signifi-

cant effects upon the motor output torque and corresponding motor performances.

In order to simulate precisely both steady state and dynamic performance of the

motor, characteristic of the inverter circuit and its control loop feature must be

included as a coupled system. However, few studies have reported the effects of

control loop feature in the performance analysis of a BLDC motor. Hui Tan [74]

simulated time stepping FEM model to study the steady state analysis of a multi-

pole brushless DC motor including with control loop feature in the system. W. N.

Fu, Z. J. Liu and C. Bi [75] proposed the performance analysis of disk dive spin-

dle motor using time stepping FEM with current loop control and PWM switching

strategies. G. H. Jang [76] proposed nonlinear time-stepping finite element analysis

of the magnetic field considering the switching action of pulse width modulation

(PWM), which controls the average voltage applied to the motor. It can be found

that all of these works which have included of coupling the time stepping FEM

with current loop control were applied only to study the steady state analysis of
 23

the motor. In the BLDC motor, the torque is directly proportional to the wind-

ing currents and speed of the motor can be controlled by measuring the stator

phase voltage applied to the motor. Therefore cascaded current and speed control

scheme is suitable for the purpose of rapid response. In this work, cascaded speed

and current closed loop control is implemented in order to simulate both static and

dynamic performances of the BLDC motor.

1.6 Scope of the Thesis

This thesis deals with the modeling, simulation and performance analysis of a

BLDC motor using numerical methods. The primary objective is to develop an ef-

ficient and practical procedure based on numerical techniques to analyze the steady

state and dynamic performances of BLDC motors. In order to achieve these ob-

jectives, the following modelling techniques, numerical methods and experimental

set up are developed in this thesis.

• Dynamic model of the BLDC motor is developed using two dimensional time

stepping finite element method. In this model, the electromagnetic field equa-

tions, motion equations and circuit equations are formulated in time domain

and solved simultaneously at each time steps. 3D effects i.e end winding ef-

fects are taken into account using empirical formulae. The eddy current effect,

the saturation effect, the skin effect, the rotor movement, the non-sinusoidal

quantities and nonlinear electric circuit systems with semiconductor convert-

ers are included in the system of equations.

• Time stepping finite element method combined with current fed two dimen-

sional FEM is used to develop the steady state model of the BLDC motor.

In this steady state model, current fed two dimensional FEM is used as a

pre-computation stage for time domain model. By using the proposed model,
 24

extensive computation time to reach to steady state using time stepping FEM

alone can be eliminated. In addition, non-sinusoidal quantities, high-order

harmonic and rotor rotation which are very difficult to take into account in

the steady state analysis using the traditional finite element method can be

included.

• A new approach to couple the time stepping FEM with control loop feature is

implemented. Speed loop cascaded with hysteresis current controller is used

in this control loop application. This approach provides simple and efficient

ways to analyze the BLDC motor as an actual motor-controller drive system.

• Experimental test stand for exterior rotor BLDC motor is developed using

DS1104 controller board. Both software and hardware set up for HDD spindle

motor is implemented.

• Performance analysis of the BLDC motor are simulated using proposed steady

state and dynamic models. Application characteristics of the HDD spindle

motor are studied using the time stepping FEM coupled with control loop

feature. Important features of the spindle motor at starting conditions under

no load and loaded conditions are investigated. Optimization of the motor

run-up performances at starting and normal running conditions are analyzed.

Simulation results are verified with experimental results.

1.7 Outlines of the Thesis

Contents of the thesis are organized in the following:

Chapter 1 discusses the basic ideas of the numerical techniques in electri-

cal machines, configuration and characteristic of the BLDC motor and the main

objectives of the thesis.

 25

Chapter 2 presents numerical modelling of the BLDC motor using conven-

tional FEM. Procedures to develop the mathematical models of BLDC motor in

magneto-static FEM are described.

Chapter 3 presents the modelling of a BLDC motor in time domain. Detailed

procedures and formulations by using time stepping finite element method are

presented. Modeling technique to reduce the starting period in studying the steady

state performance in time stepping model is described.

In Chapter 4, software and hardware implementation of the DSP based ex-

perimental set up for exterior rotor BLDC motor-controller system is presented.

Experimental procedures to measure the motor speed with various supply volt-

ages, motor back-emf, stator currents under load and no load conditions and motor

transients are described. Experimental constraints under hardware and software

limitations are discussed.

Chapter 5 deals with the performance analysis of the BLDC motor using

developed numerical models. The steady state performances of a BLDC motor

in no load and load conditions are described. Dynamic conditions of step voltage

variation, load torque changing and locked rotor condition are investigated.

In chapter 6, the application characteristics of the exterior rotor BLDC motors

using developed numerical models are presented. The starting processes of the

spindle motor with voltage, current and speed limits are investigated by coupling

with current-speed cascaded closed loop control structure. Optimization of spindle

motor performance under starting and normal running conditions are analyzed.

Chapter 7 provides the discussions and conclusions of this work.

Chapter 2

Computational Analysis of a
BLDC Motor

2.1 Introduction

It is important to predict accurately the steady state and dynamic performances

of the proposed design of a BLDC motor in order to avoid the design misjudgment

that can prove costly once the motor is manufactured. Traditionally, equivalent

circuits and lumped parameter models have been used to calculate the performance

of BLDC motors and to conduct the design with idealized and simplified physical

representation. However, it is difficult to use the traditional analytical method to

calculate their characteristics taking into account the effect of magnetic saturation,

complex configuration, eddy currents and external circuits at the same time. Basi-

cally, electrical machines are electromechanical converters and their behaviours can

be described by electromagnetic fields. Using Maxwell’s equations, all of the elec-

tromagnetic fields can be expressed by partial differential equations. To solve these

partial differential equations, numerical methods have been recognized as practical

and accurate tools. Among numerical methods, finite element method has been

proved to be an accurate tool to compute the electromagnetic field distribution and

has been extensively used for solving field related problems [77]-[81].

In this chapter, the finite element analysis for the two-dimensional electro-

26
 27

magnetic field problems is presented. The finite element model for static magnetic

field calculation is derived. Proposed static FEM model is used to develop the

steady state model of a BLDC motor.

2.2 Finite Element Analysis

In the finite element analysis, the partial differential equations modelling the field

problems are transformed into energy-related functionals. Approximate solutions

are then sought to the field problems that minimize these functionals. By this

procedure, a detailed modelling of the geometry of the field region is possible, and

the results obtained are found to be accurate. In brief, general procedures of the

finite element method are as follows [82]-[85]:

1. Define the boundary value problem by partial differential equations (Mathe-

matical formulations of the physical model)

2. Represent the variational formulation of the partial differential equations in

terms of energy-related functionals

3. Discretize the problem domain with sub-regions called ’elements’

4. Choose the trial function for each elements in terms of nodal values of the

elements and interpolating functions

5. Derive the element matrix equations using the principle of weighted residual

or the variational principle

6. Assemble the element matrix equations

7. Impose the boundary conditions into the system of equations

8. Solve the system of equations to obtain the discretized function values on

every node
 28

Detailed descriptions of the procedures listed above will be discussed in next

sections. These procedures are leading to the solution of the electromagnetic prob-

lems of BLDC motors. In this work, under-slung type hard disk drive spindle

motor (exterior rotor type BLDC) is used as a working example to study the finite

element analysis of the BLDC motor. Fig. 2.1 shows the physical model of the

BLDC motor to be analyzed.

Figure 2.1: BLDC motor configuration

2.2.1 Mathematical formulations of the physical model

The nonlinear magneto-static field in an electric machine can be described by the

following Maxwell’s equations

∇×H =J (2.1)

B = µH (2.2)

The magnetic vector potential A is defined as

B =∇×A (2.3)

Substituting equations (2.1) into (2.2), we get

B
∇× =J (2.4)
µ
 29

1
∇× (∇ × A) = J (2.5)
µ

∇ × ν(∇ × A) = J (2.6)

where A is the axial component of the magnetic vector potential, ν is the

reluctivity of the material and J is the current density. In the analysis of electric

machines, it is common practice to evaluate the field solution in two dimensions

only so that the current density and magnetic vector potential have only z-direction

components. This analysis is valid for most cases because the air gap between the

rotor and the stator in an electric machine is so small that over a considerable range

in the axial direction of the machine, except the end regions, the magnetic field

is virtually two-dimensional. Furthermore, two dimensional FEM requires much

less computing power, storage and computer time usage compared to the three

dimensional FEM [86]. In this mathematical model, it is assumed that the eddy

current on the stator laminations and rotor back iron are neglected. Therefore,

for a problem domain or region, the equation (2.6) in two dimensional X-Y planes

becomes,
   
∂ ∂Az ∂ ∂Az
ν + ν = −Jz (2.7)
∂x ∂x ∂y ∂y
The above equation (2.7) is Poisson equation considering only the current source

excitation. In this For permanent magnet BLDC motor, the mathematical model

for permanent magnets is required. There are two possibilities of the modeling of

permanent magnet materials:

• Magnetisation vector model and

• Current sheet approach.

Although these two models have different starting points, they both result in the

same set of equations [10]. In this work, magnetization vector model is used.
 30

Magnetization vector model

According to the magnetization phenomenon of the permanent magnet, the

characteristic of a permanent magnet can be shown as shown in Fig. 2.2 where Br

is the remanence flux density and Hc is the coerecitivity. The relationship between

magnetic flux density and magnetic field intensity can be given by

B = µ0 (H + M ) (2.8)

B = µ0 {(1 + χ)H + Mr } (2.9)

B = µ0 {µr H + Mr } (2.10)

B = µH + µ0 Mr (2.11)

Hence

H = ν (B − µ0 Mr ) (2.12)

we have

∇×H =J (2.13)

Substitute equation (2.12) into equation (2.13),

∇ × {ν (B − µ0 Mr )} = J (2.14)

∇ × (νB) − ∇ × νµ0 M = J (2.15)

Therefore, the field equation for two-dimensional magneto-static analysis of per-

manent magnet is

∇ × (νB) = J + ∇ × νµ0 Mr (2.16)

∇ × (νB) = J + Jm (2.17)

where Jm = ∇ × νµ0 Mr , which is the equivalent current density of the permanent

magnet.
 31

According to the equations (2.7) and (2.17), the mathematical model of the

prototype BLDC motor can be represented as

   
∂ ∂A ∂ ∂A
ν + ν = −J + Jm (2.18)
∂x ∂x ∂y ∂y

where ν is the reluctivity of the material, J0 is the applied source current

density, Jm is the equivalent current density of the permanent magnet, ω is the

angular frequency, σ is the conductivity of the material. The first term of the

right hand side of the equation represents for the applied source region (the stator

conductor region)and the last term represents for the permanent magnet region.

After we got the mathematical formulation of the physical model, according to the

FEM procedures as mentioned before, it is needed to discretize the physical model

into small sub-regions or finite elements to derive the finite element equations for

elements using variational principle or method of weighted residuals.

Figure 2.2: Characteristic of a permanent magnet material

2.2.2 Discretization of the problem domain

Domain discretization is one of the most important steps in FEM. Most of the time

necessary for the finite element method is spent for domain discretization or mesh

generation. The size of the elements, in other words the dimension or number of

the elements decides length of computing time and storage of computer memory.
 32

Since the fineness of the mesh has major effect on the accuracy of the solution.

Understanding the nature of the field is useful in determining the mesh densities of

the various places in a machine. Generally where the field is changing rapidly,the

mesh density should be higher. The shapes, sizes and the configurations of the

elements can be of many different types. Fig. 2.3 shows examples of various shapes

of elements that can be used. Triangular elements are widely used because these

elements are easy to fit in the domain sub-divisions.

Figure 2.3: Typical finite elements (a) One-dimensional (b) Two-dimensional (c)

Three-dimensional

After dividing the problem domain into finite elements, an interpolation or

shape function for the distribution of the unknown function in each element is de-

rived. The choice of the geometry of the element and the form of the approximating

function to represent the behaviour of field variables within each element are both

extremely important. Mostly, a polynomial series is used to represent the shape of

the element since the polynomial is easy to manipulate. Let us consider a triangle
 33

element with vertex i, j and k as shown in Fig. 2.4. The vertices are nodes at which

Figure 2.4: A triangular element

the unknown vector potentials will eventually be calculated. Assume the poten-

tial varies linearly in the element. With this approximation, a linear polynomial

function, as a trial function, at any point in the triangle can be expressed as

A = C1 + C2 x + C3 y (2.19)

where C1 , C2 and C3 are constants to be determined. At vertex i, j and k of the

triangle as shown in Fig. 2.4, we have x = xi , xj , xk and y = yi , yj , yk . At these

points, A must be equal to Ai , Aj and Ak so that

C1 + C2 xi + C3 yi = Ai (2.20)

C1 + C2 xj + C3 yj = Aj (2.21)

C1 + C2 xk + C3 yk = Ak (2.22)

Solving equations (2.20), (2.21) and (2.22) by using Kramer’s rule, we have

(ai + bi x + ci y)Ai + (aj + bj x + cj y)Aj + (ak + bk x + ck y)Ak

A= (2.23)
2∆

where ∆ is the area of the triangle, ai = xj yk − xk yj , bi = yj − yk and ci = xk − xj .

Equation (2.23) can be represented as,

m
X
A= Ni (x, y)Ai (2.24)
i=1
 34

Writing the nodal potential in the element in the matrix form,

 
Ai
A = (Ni , Nj , Nk )  Aj  (2.25)
 

Ak

1
Ni = [(xj yk − xk yj ) + (yj − yk ) x + (xk − xj ) y] = (ai + bi x + ci y)/2∆ (2.26)
2∆
1
Nj = [(xk yi − xi yk ) + (yk − yi ) x + (xi − xk ) y] = (aj + bj x + cj y)/2∆ (2.27)
2∆
1
Nk = [(xi yj − xj yi ) + (yi − yj ) x + (xj − xi ) y] = (ak + bk x + ck y)/2∆ (2.28)
2∆

where m are nodes in the element and Ni are the shape functions for the

triangular element because they only depend on the position of the nodes and di-

mensions or ”shape” of the element and have nothing related with the node values.

These shape functions are linear. These functions have the following important

properties:

1. Ni = 1 at node i and is 0 at all other nodes.

2. The sum of all shape functions at any point in the triangle is 1.

FEM mesh of the exterior rotor BLDC motor geometry is as shown in Fig. 2.5.

2.2.3 Derivation of the element matrix equations

The variational approach and the weighted residual approach are used to derive

the element matrix equations. Both principles force the error between the real

and the approximate solution to approach zero. The variational method is based

on the equivalent functional of the governing equation. The weighted residual

method is based on the operator equation directly. In this respect, the weighted

residual method is a more general and more universally applicable method than

the variational approach because to apply it the variational principle need not be

known. In fact, there are a number of practical cases in which the variational

expression does not exist and the weighted residual method can be applied. Based
 35

Figure 2.5: FEM mesh of the exterior rotor BLDC motor

 36

on the choice of different weighting functions, basic functions and the approach to

discretization, different weighted residual methods are formulated. In this work,

the Galerkin’s method is employed for the FE formulation. This is a particular

weighted residual method for which the weighting functions are the same as the

shape functions. The method of weighted residuals can be stated simply as follows.

Our objective is to approximate the solution to an operator equation. An operator

equation L(x) = 0; in the region Ω with boundary condition C. If we substitute an

approximate solution x̂ into an operator equation, the residual R can be obtained

because x = x̂.

L(x̂) = R (2.29)

Choose the weighing function W and set the integral of the multiplication of weigh-

ing function and a residual to zero.

ZZ
WR = 0 (2.30)
Ω

According to the Galerkin’s method, substituting an approximation  for A into

the mathematical model of the BLDC motor (equation (2.18)), the residual R is
" # " #
∂ ∂ Â ∂ ∂ Â
R= ν + ν + J0 − Jm (2.31)
∂x ∂x ∂y ∂y

if we multiply with the weighting function W and set the integral to zero over the

domain area Ω, !
1 ∂ 2 Â 1 ∂ 2 Â
ZZ
W + dxdy
Ω µ ∂x2 µ ∂y 2
ZZ ZZ (2.32)
− W J0 dxdy = W Jm dxdy
Ω
Ω

For the sake of clarity and simplicity, Galerkin’s formulation of the permanent

magnet will be discussed in next section. In this section, we are concentrated on

Galerkin’s formulation of the first three terms of the left hand side of the equation

(2.32). Rewrite the equation (2.32) eliminating the term represents the permanent
 37

magnet as follow.
!
1 ∂ 2 Â 1 ∂ 2 Â
ZZ ZZ
W + dxdy = W J0 dxdy (2.33)
Ω µ ∂x2 µ ∂y 2
Ω

Integrating the first term by parts,

! !
1 ∂ 2 Â 1 ∂ 2 Â
ZZ ZZ I
1 ∂W ∂ Â ∂W ∂ Â 1 ∂ Â
W 2
+ 2
dxdy = + dxdy− W dc
Ω µ ∂x µ ∂y Ω µ ∂x ∂x ∂y ∂y C µ ∂ n̂
(2.34)

Substitute the equation (2.34) into equation (2.33), we get

ZZ ! I
1 ∂W ∂ Â ∂W ∂ Â 1 ∂ Â
+ dxdy − W dc
Ω µ ∂x ∂x ∂y ∂y C µ ∂ n̂
ZZ (2.35)
= W Jo dxdy
Ω

We can break the surface integrals into summations over small areas. First order

triangular element is used to discretize the whole region. Domain discretization and

derivation of element shape functions have been shown in section (2.2.2). Hence,

we get
X  1 Z Z ∂W e ∂Ae ∂W e ∂Ae  1 ∂A
I
e
+ dxdy − e W e dc
M
µ Ω ∂x ∂x ∂y ∂y µ ∂ n̂
ZZ c (2.36)
= Jo W e dxdy
Ωe

where M is the number of triangular elements. The line integral of the above

equation (2.36) needs to be evaluated over elements which have a side in common

with the boundary of the problem. Normally, this integral is simply set to zero
∂ Â
implying that ∂ n̂
= 0 [10].

According to Galerkin’s method, the weighting function is the same as the

element shape function, hence,

 
Nie
W e =  Nje  (2.37)
 

Nke
Taking derivatives of equation (2.23) with respect to x and y
 
Âi
∂ Â 1 e e e 
= (bi , bj , bk )  Âj  (2.38)

∂x 2∆
Âk
 38

 
Âi
∂ Â 1 e e e 
= (ci , cj , ck )  Âj  (2.39)

∂y 2∆
Âk
Taking derivatives of equations (2.26), (2.27) and (2.28) with respect to x and y
 
e
b i
∂W e 1  e 
=  bj  (2.40)
∂x 2∆ e
bk
 
e
c
∂W e 1  ie 
=  cj  (2.41)
∂y 2∆ e
ck
Substitute equations (2.38), (2.39), (2.40) and(2.41) into the first term on the

left hand side of the equation (2.36),

!
∂W e ∂ Âe ∂W e ∂ Âe
ZZ
1
+ dxdy =
µe Ωe ∂x ∂x ∂y ∂y
(2.42)
  
b2i + c2i bi bj + c i c j bi bk + c i c k Âi
1 
 bi bj + c i c j b2j + c2j bj bk + cj ck   Âj 
 
e
4µ ∆
bi bk + c i c k bj bk + c j c k b2k + c2k Âk
The above coefficient matrix is sometimes referred to as the stiffness matrix

S in the finite element analysis. Its size is N xN , where N is the number of nodes.

The material of the elements and element structure determines the items in the

stiffness matrix.

The forcing function (right hand side of the equation (2.36))becomes

ZZ ZZ
e (ai + bi x + ci y) (ai + bi x̄ + cy ȳ)
J0 W dxdy = J0 dxdy = J0 (2.43)
Ωe 2∆ 2∆
Ωe

where x̄ and ȳ are the coordinates of the triangle.

1
x̄ = (xi + xj + xk ) (2.44)
3

Substituting the values we found for ai , bi and ci ,

1
ȳ = (yi + yj + yk ) (2.45)
3

Assuming that J0 is constant over the triangle we get

ai + bi x̄ + ci ȳ ∆
= (2.46)
2 3
 39

Therefore, the element matrix equations for the equation (2.36) is

    
b2i + c2i bi bj + c i c j bi bk + c i c k Âi 1
1  2 2 e∆ 
 bi bj + c i c j bj + c j bj bk + cj ck   Âj  = J0  1  (2.47)
  
4µe ∆ 3
bi bk + c i c k bj bk + c j c k b2k + c2k Âk 1

2.2.3.1 Galerkin’s formulation for the permanent magnet

Using Galerkin’s method, the field equation (2.32) of two dimensional magneto-

static analysis of permanent magnet can be represented as,

!
1 ∂ 2 Â 1 ∂ 2 Â
ZZ ZZ
W + dxdy = W Jm dxdy (2.48)
Ω µ ∂x2 µ ∂y 2 Ω

where Jm = ∇ × νµ0 M , hence we get

ZZ ZZ
(∇ × v(∇ × A)).W dxdy− (∇ × (vµo M ).W dxdy = 0 (2.49)
Ω Ω

Rearranging the equation (2.49)

ZZ
∇ × (v∇ × A − vµo M ) · W dxdy = 0 (2.50)
Ω

As before, we choose the weighting function W to be the same as the element shape

function, N . We now integrate by parts. Using the vector identity

(∇ × F ) · G = ∇ · (F × G) + F · (∇ × G) (2.51)

and with

F = v∇ × A − vµo M (2.52)

and G = N . Therefore,
ZZ
(∇ × (v∇ × A) − vµo M ) · N dxdy =
ΩZ Z

(v∇ × A − vµo M ) · (∇ × N )dxdy (2.53)

ZZ
+ ∇ · ((v∇ × A − vµo M ) × N )dxdy

The last term of the equation (2.53) can be written as a line integral using the

stroke’s theorem.
ZZ I
∇ · (v∇ × A − vµo M ) × N dS = {(v∇ × A − vµo M ) × N } · n̂dC (2.54)
C
 40

Ignore the line integral, the equation (2.54) becomes

ZZ ZZ
ν(∇×A) · (∇ × N )dxdy = νµo M (∇ × N )dxdy (2.55)
Ω Ω

For two dimensional case, A and J have only z components and M has only x and

y components. Substituting the curl of the weighting function in the first term on

the right hand side of equation (2.55) we obtain

ZZ   ZZ   
∂A ∂N ∂A ∂N ∂N ∂N
ν + dxdy = νµo Mx − My dxdy (2.56)
Ω ∂x ∂x ∂y ∂y Ω ∂y ∂x

The term representing the permanent magnet is

ZZ  
∂N ∂N
νµo Mx − My dxdy (2.57)
Ω ∂y ∂x

But
∂Ni bi
= (2.58)
∂x 2∆
∂Ni ci
= (2.59)
∂y 2∆

Substituting equations (2.58) and (2.59) into equation (2.57), we get

    
ZZ ci bi
νµo 
Mx  cj  − My  bj  dxdy (2.60)
   
Ω 2∆
ck bk

Therefore, according to the equations (2.47) and (2.60) the final global finite ele-

ment matrix equations for the BLDC motor is

  
b2i + c2i bi bj + c i c j bi bk + c i c k Âi
1 
 bi bj + c i c j b2j + c2j bj bk + c j c k   Âj 
 
4µe ∆
bi bk + c i c k bj bk + c j c k b2k + c2k Âk
       (2.61)
1 ci bi
∆   νµo 
= J0e  1  + M x  c j  − M y  b j 
   
3 2∆
1 ck bk

2.2.4 Assembling of element matrix equation

All element equations of the problem domain are needed to be assembled to form a

system of matrix equations. The principle for assembling the element matrices into
 41

a system matrix is based on the property of compatibility. It means that at every

node the nodal values of the unknown function of each element are the same no

matter where the node belongs to. In the assembly process all the elements matrices

are simply added to form the global matrix. We go through each element and add

the (i j) terms of the stiffness and mass matrices to the corresponding (i j) terms

in the global matrix. The resulting global matrix will be sparse, symmetric and

singular. The procedures for assembling the system matrix of the problem are as

follows. For simplicity, the problem domain containing three elements is considered

as shown in Fig. 2.6. If is coefficient matrix for element e. The contribution of

each element can be summarized as shown in Table 2.1. After assembling, the final

global matrix [K] will be

 
K11 K12 K13 K14 K15
K21 K22 K23 K24 K25
 
 
 

 K31 K32 K33 K34 K35 

K41 K42 K43 K44 K45
 
 
K51 K52 K53 K54 K55
  (2.62)
1 1 1
K11 K12 K13 0 0
1 1 2 3 1 2 3 2 3
K12 K22 + K22 + K22 K23 + K23 K24 K25 + K25
 
 
 
= 1 1 2 1 2 2
 K13 K23 + K23 K33 + K33 0 K53 

3 3 3
0 K24 0 K44 K45
 
 
2 3 2 3 3
0 K25 + K25 K53 K45 K55

In practical case, one node can only connect with the other limited nodes

through several elements so that there exists a great number of zero elements in

the matrix. Hence, the resulted stiffness matrix is symmetric and sparse. The

diagonal elements of the matrix are positive and dominant, and all the values of

the sequent sub-determinant are positive. Therefore, the system matrix is a positive

definite matrix.
 42

Figure 2.6: Problem domain containing three triangular elements

Table 2.1: Element contribution

Element No Contribution nodes Kiie e

Kjj e
Kkk Kije e
Kik e
Kjk
1 1 1 1 1 1
Element 1 1,2,3 K11 K22 K33 K12 K13 K23
2 2 2 2 2 2
Element 2 2,5,3 K22 K55 K33 K25 K23 K53
3 3 3 3 3 3
Element 3 2,4,5 K22 K44 K55 K24 K25 K45

2.2.5 Imposing the boundary conditions

Specific boundary conditions are needed to be imposed upon the system of equa-

tions. In most applications the electromagnetic field is an infinitely extending

space. In these applications, boundaries are used to simplify the finite element

model and approximate the magnetic vector potential at nodal points. In fact, the

correct and appropriate application of the boundary condition is the key to define

the field problems and to allow an accurate solution in an efficient way. For exam-

ple, rotary electrical machines have identical pole pitches or sometimes half pole

pitches. Boundaries due to symmetry greatly reduce the size of the finite element

matrix. In General, boundary conditions can be classified into three types.

• Dirichlet boundary condition

• Neumann boundary condition and

• Cyclic or periodic boundary condition.

 43

Dirichlet boundary condition sets the unknown function to a known function on

the boundary of the differential problem. This boundary condition is imposed by

assigning the specific values to the vectors on the boundary. The most common use

of Dirichlet boundary conditions is to define magnetic vector potential, A = 0 along

a boundary to keep flux from crossing the boundary. Fig. 2.7 shows in a typical

example the application of the Dirichlet boundary condition of an electromagnetic

field of a BLDC motor. It is sometimes called the boundary condition of first kind.

Neumann boundary condition is the next important boundary condition for

field problems. Here the known value of the derivative of the unknown function in

the normal direction of the boundary is prescribed. If the derivative in the normal

direction is constant, lines of constant potential can pass the outer boundary of the
∂A
studied domain. Usually, ∂n
= 0 is defined along a boundary to force flux to pass

the boundary at exactly a 90 degree angle to the boundary. The most important

property of this type of boundary condition is that by knowing symmetries of the

field, and applying the Neumann boundaries there, the numerical model can be

reduced to obtain the same solution of the problem. In this way, the problem size,

the time to prepare the field problem and the computational efforts can be reduced

significantly. On the other hand, if less than the complete geometry has to be

defined and discretised, a higher accuracy is achievable for the overall solution of

the problem without extra effort. Therefore, particular attention must be paid to

this boundary condition.

Periodic boundary condition can be applied in the rotating electrical ma-

chines, not only using symmetries in the geometry but also symmetry in the mag-

netic field distribution. This type of periodic boundary condition has the form:

kA1 (x) + A2 (x) = m (2.63)

Under load conditions, the air gap field of an electrical machine repeats itself every

pole pitch. This field periodicity can be used to define another type of boundary
 44

condition to reduce the size of the numerical model. The local potentials in such

boundaries depend on the solution of the field problem and thus inherently occur

always in pairs. One boundary is computed and the opposite one is linearly linked

to this value. In Fig. 2.8 the dotted lines indicate the pairs of the boundaries.

After determining which boundary conditions have to be used for a particular

problem, treatment of boundary conditions can be included in the formulation of

the system of equations. For example, if the problem subject to the Dirichlet

boundary condition, the nodal value of those nodes located on the boundary are

known. The known value must be removed to the right hand side of the matrix

equation. The system matrix and the column vector of the right hand side of the

equation are needed to modify as follows. Consider the following the global matrix

equation.     
k11 · · · k1m · · · k1N A1 p1
 . .. .. .. ..  .   .. 
 .. . . . .   ..   . 
    
 km1 · · · kmm · · · kmN   Am  =  pm  (2.64)
    
 . .. .. .. ..  .   .. 
 .. . . . .   ..   . 
    
k N 1 · · · kN m · · · k N N AN PN
Assuming that the node m is on the Dirichlet boundary and its potential value is

A0. Its means that the equation of m row in the simultaneous equations is . Hence,

let and is set to . Where N is 1 to i. Diagonal element of mth row is replaced by 1

and other element in the mth column are filled by zero. Then the matrix equation

(2.64) is changed to
    
k11 ··· 0 · · · k1N A1 p1 − k1m A0
 . .. .. .. ..  .   ..
 ..   ..  

 . . . .    . 

 0 ··· 1 ··· 0   Am  =  A0 (2.65)
    

 . .. .. .. ..  .   .. 
 .. . . . .   ..   . 
    
kN 1 · · · 0 · · · kN N AN P N − k N m A0

If there are several nodes on the Dirichlet boundary, the same procedure will be

needed to carried out. Above description shows, the dimension of the column ma-

trix is not reduced by introducing the known potential values of the boundary and
 45

the size of the matrix is unchanged. Consequently, the advantage of this method is

that the construction of the system matrix is unchanged and the computer program

is independent of the specific geometry.

Figure 2.7: Dirichlet boundary condition for the BLDC motor

Figure 2.8: Periodic boundary condition applied to the BLDC motor

 46

2.2.6 Numerical solution to nonlinear problems

After enforcing the appropriate boundary conditions and re-assembling the element

matrix equations, the set of simultaneous system equations are to be solved. In

this BLDC model, the ferromagnetic properties of iron and the demagnetization

properties of permanent magnets as well as the voltage versus flux characteristic of

the resulting network elements are non-linear. This result in a system of non-linear

equations and this system cannot be solved in a closed form. A numerical iteration

scheme has to be used to obtain a field solution. Newton Raphson method is widely

used in solving the non-linear finite element equations. Newton Raphson algorithm

is described in Appendix B. This method is based on derivatives of the function

or quantity to be updated, and the procedure yields rapid convergence, with the

error in a given step decreasing as the square of the error in the previous step.

Therefore, it is one of the quadratic methods.

The application of Newton Raphson’s to a nonlinear magnetic finite element

problem is as follow. Let us consider two-dimensional nonlinear Poission equation

for the magnetic vector potential. The stiffness matrix for a two-dimensional first

order triangle is
    
sii sij sik Ai Je
ν   ∆
 sji sjj sjk   Aj  =  Je  (2.66)
 
4∆ 3
ski skj skk Ak Je

This procedure consists of finding a change in the potential as a result of

a change in the reluctivity, which again varies a function of the potential. Since

the permeability of the iron and the permanent magnet may be non-linear and

field dependent, one cannot obtain the solution to the field problem by solving the

matrix equations for the discretized field region only in one iteration. Further, the

solution to the problem is piecewise continuous, we cannot determine a priori the

permeability in the different subregions or finite elements spanning the iron parts.
 47

Let us assume  
Ai
ν h i
 Je ∆
F1 = sii sij sjk  Aj − (2.67)

4∆ 3
Ak
 
Ai
ν h i
 Je ∆
F2 = sji sjj sjk  Aj − (2.68)

4∆ 3
Ak
 
Ai
ν h i
 Je ∆
F2 = ski skj skk  Aj − (2.69)

4∆ 3
Ak
For simplicity, we consider the equation (2.67) only to derive the Newton

Raphson form because the formulation procedures are the same for the rest equa-

tions (2.68) and (2.69). The first step is to find out the derivatives necessary for

the Newton Raphson method.

∂F1 ν 1 ∂ν
= sii + [sii Ai + sij Aj + sik Ak ] (2.70)
∂Ai 4∆ 4∆ ∂Ai

∂F1 ν 1 ∂ν
= sij + [sii Ai + sij Aj + sik Ak ] (2.71)
∂Aj 4∆ 4∆ ∂Ai

∂F1 ν 1 ∂ν
= sik + [sii Ai + sij Aj + sik Ak ] (2.72)
∂Ak 4∆ 4∆ ∂Ai

The Newton Raphson form is

∂F1 ∂F1 ∂F1

∆Ai + ∆Aj + ∆Ak = −F (Ai , Aj , Ak ) (2.73)
∂Ai ∂Aj ∂Ak

Substituting equations (2.70), (2.71) and (2.72) into the equation (2.73) we

get
ν
[sii ∆Ai + sij ∆Aj + sik ∆Ak ] +
4∆  
1 ∂ν ∂ν ∂ν
[sii Ai + sij Aj + sik Ak ] ∆Ai + ∆Aj + ∆Ak (2.74)
4∆ ∂Ai ∂Aj ∂Ak
ν Je ∆
=− [sii ∆Ai + sij ∆Aj + sik ∆Ak ] +
4∆ 3
 48

The rest of the equations (2.71) and (2.72) can derive the same procedures as

mentioned above.
∂ν
In the equation (2.74), the term ∂A
can be represented by using the chain

rule as
∂ν ∂ν ∂B 2
= (2.75)
∂A ∂B 2 ∂A
∂B 2
Where the the term ∂A
can be calculated as follows:

The flux density B can be calculated by using the magnetic vector potential

values at the element nodes. For first order elements, we have

ai + b i x + c i y aj + b j x + c j y ak + b k x + c k y
A= Ai + Aj + Ak (2.76)
2∆ 2∆ 2∆
∂A Ai c i + Aj c j + A k c k
Bx = = (2.77)
∂y 2∆
∂A Ai b i + Aj b j + Ak b k
By = − =− (2.78)
∂x 2∆
q
B = Bx2 + By2 (2.79)
2 2
(Ai bi + Aj bj + Ak bk )2 + (Ai ci + Aj cj + Ak ck )2
 
2 ∂A ∂A
B = + = (2.80)
∂x ∂y 4∆2
∂B 2 2bi (Ai bi + Aj bj + Ak bk ) + 2ci (Ai ci + Aj cj + Ak ck )
= (2.81)
∂A 4∆2
∂ν
The value of ∂B 2
can be calculated from the typical magnetization curve as

shown in Fig. 2.9. For the realization of a numerical solution, the numerical values

of the non-linear characteristics of the magnetization curve of the ferromagnetic

magnetic material, specified as discrete sets of points, is stored in a file in a com-

puter. Since Newton Raphson method requires a continuous representation of the

variables, interpolation procedures are used to generate the required values. Nu-

merous ways of modeling magnetic property curves have been tried out. Different

methods of interpolation are available. In the simplest case, one can use a stepwise

linear interpolation. The main disadvantage of this method is the discontinuity of

the equations required for the evaluation of the Jacobian elements, which adversely
 49

affects the convergence of the Newton Raphson’s method. In addition, the devia-

tion of the approximated value from the real function is considerable. For greater

accuracy, polynomials of higher order can be used. If the first priority is on the

continuity of the first and higher derivatives of the equations, a rigid interpolation

polynomial of a spline function is used. The main reason for the interest in splines

is that they result in a simple formulation. They interpolate exactly at the given

data points and have a continuous first and second order derivative. The contin-

uous first derivative makes the method suitable for the Newton iteration. A great

advantage of using cubic spline is the fact that this linear system of equations has

to be solved only once to obtain the values of the second derivative. For the finite

element method, this system of equation has to be solved only once independently

of the number of finite elements used in the model of the magnetic circuit. For the

practical used of the method a field of 20-25 data samples is sufficient to represent

a non-linear ferromagnetic material characteristic. In this work, cubic spline in-

terpolation is used to represent the B-H curve of the stator and rotor irons of the

BLDC motor. The development of cubic spline algorithm is shown in Appendix C.

Figure 2.9: Magnetization curve for ferromagnetic material

 50

2.2.7 Solution of the System of Equations

Generally, the methods for solving the linear algebraic equations can be classified

into direct methods and iterative methods. Direct methods are supposed to give the

exact solution from the theoretical point of view. And the computation time can

also be estimated. However, due to the rounding errors produced by computation,

the results are still approximate.

Iterative methods have the advantage that the sparseness of the coefficient

matrix is utilized. If a problem can be solve by iterative method, the solutions

are not affected by the rounding errors. However, the computation time is difficult

to be estimated as it is dependent on the initial values used. The over-relaxation

iteration and conjugate gradient method are two well known schemes using it-

erative methods. The matrices resulting from the finite element formulation are

sparse and positive definite. Although the number of equations may be large, each

equation will have relatively few terms. An efficient method for solving sparse ma-

trix equations resulting from the discretization of boundary value problems is the

preconditioned conjugate gradient method. This method combines useful aspects

of iterative methods with the desirable characteristics of direct methods. In this

work, Incomplete cholesky conjugate gradient method (ICCG) is used to solve the

system of equations.

From the mathematical point of view, solution of the system matrix is not

a difficult problem. However, in practical cases, as the number of un-knowns are

normally in thousands or even in hundreds of thousands, the size of the stiffness

matrix is huge. Therefore an effective storage method plays an important role in

solving the global system of matrix equations. In the finite element computation,

an effective storage scheme means,

• It can store all the information of the stiffness matrix.

 51

• It can use small storage to store the matrix information.

As the stiffness matrix is symmetric, banded and non-zero terms are clustered

around the main diagonal, utilizing this characteristics, non-zero storage methods

are mostly used in finite element analysis. In this work, one type of the non-zero

storage methods called index storage method is used. The index storage method

is suitable for solving the linear system of equation by using ICCG algorithm. The

structure of the proposed non-zero storage method is as follows.

Consider the mth row of the following global matrix [K].

 
k11 k12 k13 k14 0
 k21 k22 0 k24 0 
 
 
K=  k31 0 k33 k34 k35 
 (2.82)
 k41 k42 k43 k44 k45 
 
0 0 k53 k54 k55

In this method, upper half of the matrix with non-zero terms is needed to store.

Detailed procedures are as follows.

1. Arrange the non-zero element column by column in the matrix [A].

2. The corresponding row numbers of each element are stored in the matrix

[IC].

3. Meanwhile, the position of each diagonal element in the matrix [A] is stored

in matrix [ID].

For example, for the global matrix [K] from the equation (2.82),
h i
[A] = k11 k12 k13 k14 k22 k24 k33 k34 k35 k44 k45 k55 (2.83)
h i
IC = 1 1 1 1 2 2 3 3 3 4 4 5 (2.84)
h i
ID = 1 5 7 10 12 (2.85)

In the ICCG iteration, retrieve the element position by using the these two matrices

[IC] and [ID]; and solve the linear equations according to the algorithm.
 52

2.3 Conclusion

This chapter described two dimensional finite element analysis of the prototype

BLDC motor. Mathematical models of the BLDC motor were developed using

Maxwell’s equations. Basic procedures and modelling techniques of two dimen-

sional finite element method were presented. Proposed FEM model will be used as

a pre-computation stage for developing the steady state model of the BLDC mo-

tor. In the steady state model, the solutions got from this FEM model are used as

initial values for time stepping solver in order to reduce the extensive computation

time to reach to the static conditions in time stepping FEM. Modelling of BLDC

motors using time stepping finite element will be presented in the next chapter.
Chapter 3

Time Domain Modelling of a

BLDC Motor by Coupled Systems

3.1 Introduction

Electrical machines are electromechanical converters and their behaviours can be

described by the electromagnetic fields. In turn, electrical machines receive power

form external sources through electric circuit. This requires the modelling of elec-

tromagnetic field inside the machine to be coupled with electric circuit. In addition,

electrical machines are moveable devices. It is important to take into account the

interaction of electromagnetic field with mechanical forces and motions. Therefore,

a comprehensive modelling of field, circuit and motion of an electrical machine sys-

tem should be considered together. Moreover, if the machine is in its transient

state, transient fields are varied with time. So it is needed to develop the model in

time domain.

A method for proper coupling of transient fields, circuits and motion must be

such that [86]-[87]:

1. Only terminal voltage (or total terminal current) applied to the device is

required as a known input quantity, and total terminal current (terminal

voltage) is calculated as an unknown;

2. The transient external circuit equations that represent the electrical sources

53
 54

and circuit components are coupled to the finite element field equations; and

3. Equations for motion are coupled to the finite element field equations.

The most rigorous way is to allow for time variation of magnetic non-linearity,

harmonic effects, rotor rotation and coupling with the external circuit in the time

stepping approach. The following sections describe a non-linear time stepping

method to analyze the performance characteristics of BLDC motors. A system

composed of a BLDC motor, a semiconductor inverter and the control equipment

is considered. The proposed mathematical model includes the finite element equa-

tions of the electromagnetic field in time domain, the equations which define the

connections of windings and inverter elements, the motion equation, and the rela-

tions which describes the control systems. A HDD spindle motor (exterior rotor

type BLDC motor) is used as a typical example to describe the principles of the

proposed method.

3.2 Modelling Techniques

In the time stepping FEM model, the electromagnetic fields represented by Maxwell’s

equations are coupled with the stator circuit equations and the rotor motion equa-

tions and solved simultaneously at each time step. The inputs are stator phase

voltages, rotor position, motor geometries and material characteristics; other vari-

ables such as magnetic vector potentials for each node of the FEM mesh, stator

currents, back-emf, motor torque and the speed are directly calculated. The mag-

netic field equations of the stator and the rotor are written in their own coordinate

systems in time domain. The solution of the two field equations are matched with

each other in the air gap. The rotor part of the FEM mesh is rotated at each time

step by an angle determined from the motion equation. Hence, the solution of the

time stepping model can take into account the complexity of the geometric model,
 55

non-linearity, induced eddy currents, the material saturation and the rotor rotation

which are very difficult to be considered using analytical methods.

3.3 Mathematical Model of the BLDC Motor

An underslung type hard disk drive spindle motor is used in this model. This

is the exterior rotor type permanent magnet BLDC motor. Underslung motors

are generally designed for low capacity and low profile hard disk drives. These

are mostly used in PC computers. Fig. 3.1 shows the mechanical structure of the

underslung type disk drive spindle motor. The configuration of the motor and

power electronic circuit to be modelled are shown in Fig. 3.2. Motor specifications

are listed in Table A.1. Demagnetization curve of the permanent magnet used in

this motor is given in Appendix D. In the proposed model, the terminal voltages

on the stator windings of each phase are controlled by power electronic switches

according to the control scheme. In this star connected motor, at any moment,

there are two phase windings in conducting state and one phase winding in non-

conducting state. Typical switching states and input voltage waveforms are shown

in Fig. 3.3. Modelling of the BLDC motor in the time stepping FEM can be divided

into three main parts:

1. Electromagnetic field modelling

2. Circuit modelling

3. Rotor rotation

Each part yields its own matrix equations. They are coupled and solved simulta-

neously at each time step.

 56

Figure 3.1: Mechanical structure of the motor

Figure 3.2: BLDC motor configuration and power electronic circuit

 57

Figure 3.3: Typical input voltage waveforms with respective electrical degrees

3.3.1 Electromagnetic field modelling

Time-dependent electromagnetic field of the BLDC motor can be represented as

∇ × (ν∇ × A) = J (3.1)

where A is the magnetic vector potential, ν is the reluctivity of the material and

J is the current density. J and A are assumed to be z direction and independent

of z. According to formulation equation (??), in the area of the stator conductor,

the electromagnetic field equation can be represented as

is
∇ × (ν∇ × A) + =0 (3.2)
S

where is is the stator phase current and S is the total cross-sectional area of the

one turn on one coil side.

In the solid rotor iron core and the laminated stator, there have eddy current

effect, the electromagnetic field can be expressed as

∂A
∇ × (ν∇ × A) + σ =0 (3.3)
∂t

There is no induced electric field produced by time varying magnetic flux in the

air gap and the term σ ∂A

∂t
is zero. Therefore, the electromagnetic field equation in
 58

these areas can be simply represented as

∇ × (ν∇ × A) = 0 (3.4)

There are two models which are commonly used to represent permanent magnets;

a magnetization vector model and an equivalent current sheet model. Magnetizing

vector method is used in this modelling. Detailed procedures and formulation

of magnetization vector method has been given in section (2.2.1.1). According

the proposed method, the permanent magnet can be represented as an equivalent

magnet current source as

∇ × (ν∇ × A) = ∇ × (νBr ) (3.5)

3.3.1.1 Modelling of eddy current effect on stator lamination

Most of the stator iron cores of BLDC motors are built up out of thin lamina-

tions to reduce the eddy current effects. There are many methods and approach to

model eddy current effect on stator laminations [88]-[89]. One possible method is

to model each discrete lamination and insulation between laminations in the finite

element geometry. An alternative is to treat the laminated material as a contin-

uum and derive bulk properties. Bulk lamination modelling is applied to model

the stator lamination in this thesis because it is simple, reliable and easy to imple-

ment in FEM calculation. David Meeker, developer of Free FEM software named

Finite Element Method Magnetic (FEMM) has implemented this bulk approach

[13]. Detail modelling technique is as follow.

Consider that the flux can flow through the laminations in a combination of

two ways: along the laminations and across the thickness of laminations as shown

in Fig. 3.4. Assume the permeability of the iron is isotropic in deriving the bulk

lamination model. For the condition where the flux flow along the laminations, the

circuit model can be produced as shown in Fig. 3.4. There are two reluctances in
 59

the model: one for flux that flows through the iron part of the laminations Rf e and

another reluctance for flux that flows through the air between laminations, Rair .

L
Rf e = (3.6)
µr µo cW
L
Rair = (3.7)
µo (1 − c)W

where L and W are the length and width of the path traversed and c is the fraction

of the path filled with iron. Adding these two reluctances in parallel yields

L
Rez = W (3.8)
((1 − c) + cµr )µo

Since L and W are arbitrarily chosen, the bulk permeability of the section is

µez = ((1 − c) + cµr )µo (3.9)

For the condition where the flux flow across the thickness of the laminations, circuit

model can be described as shown in Fig. Fig. 3.5. In this case, the circuit is two

reluctances are in series, as the flux has to cross the insulation and the lamination.

These reluctances are

cL
Rh = (3.10)
µr µo W
(1 − c)L
Ra = (3.11)
µo W

Adding these two reluctances together in series yields

 
c + (1 − c)µr L
Rhard = (3.12)
µr µo W

Since L and W are arbitrarily chosen, the bulk permeability of the condition is

µr µo
µhard = (3.13)
c + (1 − c) µr

In equation (3.2), the value of bulk permeabilities are substituted. Hence, the

electromagnetic field equation (3.2) takes into account the eddy current effect in

stator laminations.
 60

Figure 3.4: Equivalent circuit for flux flow through the laminations

Figure 3.5: Equivalent circuit for flux flow across the thickness of laminations
 61

3.3.2 Modelling of electric circuit

In BLDC motors, the stator windings are connected in delta or star. In both cases,

the power electronic circuit and the switching sequence of the transistors remain

the same. Since the settling time of the semiconductor devices are shorter than the

time step length of the time stepping FEM, the switching devices can be simply

represented by the nonlinear resistances corresponding to the forward conduction

resistance in the ‘ON’ state and a high resistance to simulate the ‘OFF’ state [90]-

[94]. In this model, one of the high speed switching MOSFET model-IRF620 is

used as the switching devices. BLDC motor operates on square wave supply and

two phases are in operation. Commutation (changing switching states between

two phases) takes place every 60 electrical degrees. During this interval the third

phase carries no current and is idle. For example, during interval when phase A

and phase B are conducting at the same time as shown in Fig. 3.6, the conducting

current flows from phase A to phase B and semiconductor switches S1 and S4 will

be ’ON’.

The stator winding for one phase can be represented by total stator phase re-

sistance, end winding inductance and induced back-emf voltage as shown in Fig. 3.7.

It is assumed that the coil ends are treated as concentrated resistance and induc-

tances elements in this two dimensional model. Hence, the stator phase circuit

equation can be expressed in general form as

di
V = Ris + Lσ +e (3.14)
dt

where R is the sum of ’ON’ state semiconductor resistance and armature

winding resistance, V is the stator phase voltage, Lσ are ending winding induc-

tances and e is the induced voltage (back-emf). In this equation, stator phase

voltage is input and the output stator current will be calculated by coupling the

circuit equation (??) with the field equations in the time stepping model. Required
 62

parameters such as winding resistance, end winding inductance and back-emf can

be determined as describe in next section.

Figure 3.6: The inverter circuit where current flow from phase A to B

Figure 3.7: Circuit representation of a phase winding

3.3.2.1 Determination of DC winding resistance and Back-emf

The electromagnetic field in an electric machine is governed by Maxwell’s equations:

∂B
∇×E =− (3.15)
∂t

with the constitute relationship:

J = σE (3.16)
 63

But

B =∇×A (3.17)

Substituting equation (3.15) to equation (3.17) yields

∂ (∇ × A) ∂A
∇×E =− = −∇ × (3.18)
∂t ∂t

or
 
∂A
∇× E+∇× =0 (3.19)
∂t

According to the Poincare’ Theorem, the curl of a gradient is zero, hence we can

write
∂A
E+ = −∇V (3.20)
∂t
∂A
E=− − ∇V (3.21)
∂t

where V is a scalar potential. Substituting equation (3.21) into equation (3.16),

we get
 
∂A
J = σE = σ + ∇V (3.22)
∂t

In the equation (3.22), the first term of right hand side represents induced electric

field intensity per unit length of stator conductor produced by electromagnetic

induction (back-emf) and the second term is caused by the effect of charge build

up at the conductor end.

If we consider a current carrying conductor of length l, its positive terminal

is point a and a negative terminal is point b. If the difference in electric scalar

potential between point a and b is Vab , we can write

Zb Zb
Vab = Va − Vb = Edl = − ∇V dl (3.23)
a a

In two-dimensional problem ∇V is considered to be constant in z-direction, then

equation (3.23) becomes

Vab = −∇V · l (3.24)

 64

Vab
∇V = − (3.25)
l

Therefore, the current density for the current carrying conductor for two dimen-

sional field is
∂A Vab
J = −σ +σ (3.26)
∂t l

Integrating the equation (3.26) over its corresponding cross-sectional area S, we

get the total current flowing through the conductor, which is

ZZ
I= JdxdyJ
S
ZZ ZZ
∂A Vab
= −σ dxdy + σ dxdy (3.27)
∂t l
ZSZ S
∂A S
= −σ dxdy + σ Vab
∂t l
S

Divided by σ S1 Va b on both sides of the equation (3.27), the terminal voltage vab

across a Z-directed conductor with length l and cross-sectional area S is

ZZ
l l ∂A
Vab = I+ dxdy (3.28)
σS S ∂t
S

In the equation (3.28), the first term of the right hand side is the voltage drop

across the stator conductor and the last term represents the induced voltage (back-

emf) of the conductor. Therefore, DC winding resistance for the conductor can be

calculated by
Nl Nl
R= = (3.29)
σS 4σπd2

where N is number of winding per phase, d is the diameter of the wire.

Stator windings are made up by conductors connected in many turns. If we

consider a ’go’ and ’return’ loop of current carrying conductors, back-emf for one

phase can be represented as

 
ZZ ZZ
Nl  ∂A ∂A
e= dxdy − dxdy  (3.30)
S ∂t ∂t
S+ S−
 65

3.3.2.2 End winding inductance

End winding inductance is difficult to calculate accurately with simple formulas

because the conformation of end winding is complex and difficult to characterize

mathematically in simple terms. Determination of the end winding inductances

are different from different winding structures, end winding layers, rotor type etc.

Probably the easiest way to determine the end winding inductance for distributed

winding type and concentrated winding type are by the following analytical meth-

ods. Validity of the analytical results has been verified by experiment and finite

element analysis in [3] and [95]-[98].

Determination of end winding inductance for distributed winding:

The inductance is estimated by assuming that the end-windings from the

two ends taken together form a circular coil whose diameter d is equal to the chord

spanned by the coil, evaluated at the radius of the bottoms of the slots, rs as shown

in Fig. 3.8. But

D
rs = − ds (3.31)
2

And coil pitch is τ slot pitches, the coil span in radians is

2π
σ= τ (3.32)
Ns
σ
d = 2rs sin( ) (3.33)
2

Therefore, the end winding inductance of one coil Lσ is calculated from the formula

µ0 dNc2
 
4d
Lσ = ln −2 (3.34)
2 dGM D

where Ns is number of complete coils, Nc is the number of turns per coil, dGM D is the

geometric mean distance between the conductors in the coil cross section. If the coil
√
cross-section is assumed square with the area A, then dGM D is taken to be 0.447 A.

The total contribution from the phase windings to the phase inductance can be

obtained from the series/parallel combination of the end-winding inductances of

 66

the number of coils per phase, each one of which is given by the above equation

(3.34). The geometric mean distance is modified to include all conductors of one

coil which are assumed to be perfectly coupled in the end windings. With this

changes, end winding inductance of the distributed winding structure is

Ns µ0 dNc2
 
4d
Lσ = 2 . ln −2 (3.35)
a 2 dGM D

where a is the number of parallel paths.

Determination of end winding inductance for concentrated winding:

Most of the hard disk drive spindle motor has used concentrated type winding

due to their short end turns and simple structure suitable for high-volume auto-

mated manufacturing. The basic idea to determine the end winding inductance for

concentrated winding is the same as the distributed winding as mentioned above.

However, there a little modification is needed to apply because of the different

winding structures. To calculate the end winding inductance of one coil, it is as-

sumed that the end windings are the circular arc as shown in Fig. 3.9. If the stator

iron core between the end-coils is removed, the two end-coils from the same coil

form a circular loop as shown in Fig. Here, the diameter Dc is approximately equal

to the length of the arc segment of the middle of the two adjacent slots. We have

2π
α= (3.36)
Ns

Hence

Dc = r × α (3.37)

where Ns is the total number of slots. The end winding inductance for one coil is

µ0 Dc Nc2
 
4Dc
Le = ln −2 (3.38)
2 dGM D

where Nc is the number of turns per coil and dG M D is the geometric mean distance

between the conductors in the coil cross-section. The total end winding inductance
 67

per phase is obtained by multiplying the number of coils per phase Ncoil and di-

viding the a2 , where a is the number of parallel paths as below.

Ncoil µ0 Dc Nc2
 
4Dc
Le = 2 · ln −2 (3.39)
a 2 dGM D

Figure 3.8: Motor geometry for distributed winding inductance calculation

Figure 3.9: Motor geometry for concentrated winding inductance calculation

3.3.3 Modelling of the rotor movement equation

Rotor motion governed by the mechanical equation is given by

dω
Jm = Te − TL + Dω (3.40)
dt
 68

where Jm is the total moment of inertia this includes the inertia of the transmission

system, the load referred to the motor shaft and the inertia of the rotor; ω is the

rotor speed; Te is the electromagnetic torque, D is the damping coefficient and

TL is the torque required to drive the load referred to the motor shaft including

the external load torque, windage torque and friction loads (load caused by the

bearings, the system inefficiencies and stiction force caused by the read/write head

with the platters in working model BLDC motor). Load torque TL , damping

coefficient D and moment of inertia Jm are necessary to determine prior before

the coupling the motion equation with field and circuit equations.

3.3.3.1 Consideration of load torque

It includes friction torque, stiction torque and external load torque. It is known

that friction torque occurs from the relative motion between surfaces and it is

mainly due to in bearings, lead screws, gear-boxes, slideways, etc. In the prototype

motor, stiction force between read/write head and the platters at starting condition

is needed to be taken into account. In general, the force required to overcome the

surface friction, Ff , for a normally applied load, W , is given by [99]

Ff = γW (3.41)

where γ is the coefficient of friction; typical values of friction coefficients are given

in Table 3.1. The value of coefficient of friction depends on both surfaces. The

motor used ball bearings, and therefore friction caused by ball bearings effect is

necessary to be considered. The manufacturer can supply a complete data set, but

as an indication, the friction torque, Tb for a bearing can be determined using the

generally accepted relationship

Tb = 0.5BL dµb (3.42)

where d is the shaft diameter, µb is the coefficient of friction for the bearings (Table

3.1) and BL is the bearing load computed from the radial load, Fr and the axial
 69

load, Fa in the bearings; BL is given by

1/2
BL= Fr2 + Fa2 (3.43)

Table 3.1: Friction coefficients

Material Friction coefficient Mechanism Friction coefficient

Steel on steel 0.58 Ball bushing < 0.001

Steel on steel (greased) 0.15 Linear bearing < 0.004

Aluminium on steel 0.45 Dove tail slides 0.2

Copper on steel 0.30 Gibb ways 0.5

Brass on steel 0.35 − −

Plastic on steel 0.15 − 0.25 − −

3.3.3.2 Determination of rotor inertia

The following method for calculating rotor inertia is based on the assumption that

all parts of the rotor rotate around the center of the shaft. For the proposed motor,

we shall calculate the inertia of each part separately as shown in Fig. 3.10, then

add the parts together for the total rotor inertia. The inertia of the shaft is given

by
 4
Dshaf t
Jshaf t = 2πρLshaf t (3.44)
2

The inertia of the solid rotor iron core is given by

" 4  4 #
Dro Drm
Jiron core = 2πρLc − (3.45)
2 2

The inertia of the permanent magnet is given by

" 4  4 #
Drm Dri
Jpm = 2πρLc − (3.46)
2 2
 70

There are two parts in the aluminium hub; one part has a bigger diameter

as shown in Fig. 3.10. Therefore these two parts are considered separately in the

calculation of inertia. The inertia of the first part (larger diameter) which is fixed

with the permanent magnet and the solid rotor iron core is given by
" 4  4 #
Drm Dri
Jpm = 2πρLc − (3.47)
2 2

The inertia of the other part which is the outermost part of the rotor is given by
" 4  4 #
Dro DAl
JAl out = 2πρLAl out − (3.48)
2 2

For the above equations ρ is the density of the material used respectively. Density

of the aluminium is 2700kg/m3 and the steel is 7750 kg/m3 (Table 3.2). Densities

of the common permanent magnets are also shown in Table 3.3 for reference. In

this model, NdFEB magnet is used.

Figure 3.10: Rotor part of the BLDC motor

 71

Table 3.2: Material densities

Material Lb/in3 g/cm3

Aluminium 0.098 2.7

Brass 0.3 8.3

Bronze 0.295 8.17

Copper 0.322 8.91

Plastic 0.04 1.11

Steel 0.28 7.75

Hardwood 0.029 0.8

Table 3.3: Densities of permanent magnet

Material Lb/in3 g/cm3

Bonded ferrite 0.134 3.71

Sintered ferrite 0.177 4.9

Alnico 0.264 7.31

Samarium cobalt 0.302 8.35

NdFeB 0.271 7.5

3.4 Mesh Generation and Rotation

Mesh generation for the time stepping FEM should be simple, robust and rotor

meshes should be allowed to rotate easily. In this work, simple and reliable moving

mesh technique developed by W. N. Fu at el [100] is used for meshing and rotation of

the problem domain. The moving mesh technique adopted in this work is described

below.

The FEM mesh of the cross-section of the BLDC motor is divided into two

parts; the stator and the rotor, with each including a part of the air gap. The air

gap is divided into three layers: two upper layers belong to the rotor and the rest

layer belongs to the stator. Meshes of the two parts are then generated separately.
 72

Detailed procedure of basic mesh generation composes of the following steps:

1. Generation of the boundary mesh by using the algorithm of reduction of

nodes [100].

2. Identification to insert the internal points according to the relative area of

elements by using element quality factors.

3. Improvement on the mesh connection and quality by Delaunay method [101]-

[102].

In dividing the mesh, triangular elements are used. The algorithm of reduction of

point is to reduce the points one by one from every edge. The main idea is the

identification of the point which is to be eliminated. To find a suitable point in

line L, a new element is created. From the edge L, try to find out the points so

that the newly created element has the best quality factor. The quality factor of

the triangular elements can be determined by the following formula [103].

√
4 3∆
Q2 = 2 (3.49)
(a + b2 + c2 )

where a, b and c are the lengths of the sides of the triangle; and δ is the area of

the triangle. The quality factor of a triangle is always greater zero. An equilateral

triangle has the best quality factor of 1.

After identification of the points to be reduced, it is needed to justify the new

created element that it is reasonable. The following procedures are carried out for

justification:

• Check whether the new element is inside the problem domain

• Justify whether any line already exists across the new element

• Make sure it is the new element

• Whether the new element includes any point.

 73

If it satisfies the above conditions, the point can be reduced and a new element is

created. If there is no point that satisfies the above procedures, one has to start

again from the next edge. Detailed program formulation can be found in reference

[100].

For quality improvement, it is needed to create internal points based on the

geometrical considerations. It is also necessary to justify where to insert points

and how to insert points. In this technique, two approaches have been used.

1. Calculate the element index n as shown below and if the index n is larger

than the average value of all the elements, a point is inserted.

∆k
Index, n = N
(3.50)
1
P
N
∆i
i=1

2. Check the quality factor of each elements and if the quality factor of element

is small compared to the average value of the qualification factor of all the

element, a point is inserted.

The final step of automatic mesh generation is done by Delaunay method. Further

detailed of this method is in references [104]. Delaunay triangulation provides a

mesh that is optimal for a given set of points. However, it cannot avoid thin

triangles because all points have been fixed. Therefore, a step for mesh smoothing

has to be carried out. This is accomplished by replacing each internal node at a

new position given by [104]:

N
n ω X
0
x = (1 − ω)x + xi (3.51)
N i=1

N
n ω X
0
y = (1 − ω)y + yi (3.52)
N i=1

where (x0 , y 0 ) and (xn , y n ) is the old and new positions of a given node respectively

and (xi , yi ) are the neighboring node locations. The relaxation factor ω is in the
 74

interval of [0, 2] and N is the number of neighboring nodes. This process is activated

for several steps until the all of the element quality stop increasing.

Fig. 3.11 and Fig. 3.12 shows separately generated the basic stator mesh and

rotor mesh including air gap parts. As shown in Fig. 3.13, the air gap is divided

into three layers and the upper two layers belong to the rotor part and the rest layer

belongs to the stator part. The inner-most nodes of the rotor mesh and the outer

most nodes for the stator mesh are connected by the periodic boundary condition

as shown in Fig. 3.14. When the rotor is rotated according to the time step, the

shape of the mesh for both the stator and rotor can be kept constant and only the

coordinates of the rotor mesh and the periodic boundary condition on the interface

are needed to be changed. Therefore, in this moving mesh technique, the stator

mesh and the rotor mesh are needed to generate only one time. FEM mesh after

rotor is rotated 1000 steps is shown in Fig. 3.15.

Figure 3.11: Basic stator mesh in including air gap parts

 75

Figure 3.12: Basic rotor mesh including air gap part

Figure 3.13: FEM mesh at air gap

 76

Figure 3.14: FEM mesh before rotor rotation (1899 nodes, 2828 elements)

Figure 3.15: FEM mesh after rotation 1000 steps

 77

3.5 Finite Element Formulation in Time Domain

In this proposed model, the Galerkin’s method is employed for the finite element

formulation. This is the particular weighted residual method for which the weight-

ing functions are the same as the shape functions. According to the Galerkin’s

method the magnetic vector potential A at each node of the triangular element can

be represented as
3
X
 = N i Ai (3.53)
i=1

and the element shape functions, Ni are equal to the weight functions, Wi .

W i = Ni (3.54)

where i = 1, 2, 3.

Galerkin’s formulations for the permanent magnet has been described in sec-

tion (2.2.3.1). Galerkin’s formulations of the magnetic field in iron core and stator

conductor region; circuit equations and the motion equation for the proposed mod-

elling of the BLDC motor in time domain are discussed in the following sections.

3.5.1 Galerkin’s formulation of the electromagnetic field

equation in iron core

The eddy current effect in the iron core of the solid rotor and the laminated stator

of the motor cannot be ignored. The field equation (3.3) of the iron core is
   
∂ ∂A ∂ ∂A ∂A
ν + ν +σ =0 (3.55)
∂x ∂x ∂y ∂y ∂t

According to Galerkin’s method, the field equation is multiplied by the weighting

function W , choose the approximation solution  and set the integral to zero. The

equation (3.55) becomes,

ZZ ! ! ! ZZ
∂ ∂ Â ∂ ∂ Â ∂A
W ν + ν dxdy + Wσ =0 (3.56)
∂x ∂x ∂y ∂y ∂t
 78

Integrating the left side of the above equation by parts,

ZZ ! I ZZ
∂W ∂ Â ∂W ∂ Â ∂ Â ∂A
ν + dxdy − νW dc + Wσ dxdy = 0 (3.57)
∂x ∂x ∂y ∂y ∂n ∂t

These surface integrals may be broken into summations over small areas. The

whole domain is replaced with small sub-domains so called finite elements, and

integration can easily be performed over the individual elements.

 
N Z Z e e e e e Z Z Z e
X  ∂W ∂ Â ∂W ∂ Â ∂A ∂ Â 
νe + dxdy − νe Wie dC + σ W dxdy = 0
i=1
 ∂x ∂x ∂y ∂y ∂ n̂ ∂t 
C
(3.58)

where N is the total number of elements in the domain. The line integral in equa-

tion is only employed over elements which have a side common with the boundary
∂A
C. By imposing the natural boundary condition ,i.e ∂ n̂
= 0 this line integral is

vanished. Therefore, the equation (3.58) becomes

∂W e ∂ Âe ∂W e ∂ Âe ∂ Âe

ZZ ZZ
+ dxdy + σ W dxdy = 0 (3.59)
∂x ∂x ∂y ∂y ∂t

According to the equation (3.54) , where weighting function W is equal to

element shape function N , the equation (3.59) becomes

∂N e ∂ Âe ∂N e ∂ Âe ∂ Âe

ZZ ZZ
+ dxdy + σ N dxdy = 0 (3.60)
∂x ∂x ∂y ∂y ∂t

In matrix form,
" ( )e #
X n oe ∂ Â(t)
ν [S(t)] Â(t) + σ [T ]e
e
=0 (3.61)
e
∂t

∆e
 

 i = j 
where Te,ij =
RR
Ni Nj dxdy = 6
 ∆e

i=6 j

12
RR  ∂Ni ∂Nj ∂Ni ∂Nj  b bj +ci cj
Se = ∂x ∂x
+ ∂y ∂y dxdy = i 4∆ e

and ∆e is the triangular area of the element.

 79

3.5.2 Galerkin’s formulation of field equation in the stator

conductor area

According to the electromagnetic magnetic field equation of the stator conductor,

equation (3.2), the magnetic diffusion equation of the stator conductor area can be

represented as
   
∂ ∂A ∂ ∂A is
ν + ν + =0 (3.62)
∂x ∂x ∂y ∂y s

Multiplying by the weighting function, W , and choosing the approximation solu-

tion, and setting the integral to zero, we get

ZZ       ZZ
∂ ∂A ∂ ∂A is
W ν + ν dxdy + W dxdy = 0 (3.63)
∂x ∂x ∂y ∂y s

Integrate the first term of the equation (3.63) by parts and neglect the line integral

as mentioned in above section, we get

ZZ " # ZZ
∂W ∂ Â ∂W ∂ Â is
ν + dxdy + W dxdy = 0 (3.64)
∂x ∂x ∂y ∂y s

Now break the surface integrals into summations over small areas. First order

triangular elements are used to discretize the desired motor geometry. The integral

over the entire domain is represented with the summation of the integral over the

individual triangles. we can get

N ZZ
" #
e e e e ZZ
X
e ∂W ∂ Â ∂W ∂ Â is
ν + dxdy + W e dxdy = 0 (3.65)
i=1
∂x ∂x ∂y ∂y s

where N is the number of triangular elements. Substituting the equations (3.53)

and (3.54) into equation (3.65), the Galerkin’s form of the electromagnetic field

equation in the stator conductor area is

  ! ! 
 3
P 3
P 

 ∂ N j Aj ∂ N j Aj 
X ZZ  
  ∂Ni j=1 ∂N i j=1



ZZ
is
ν + dxdy + Ni = 0
e   ∂x ∂x ∂y ∂y  s
  


 

(3.66)
 80

Alternatively, in matrix form,

X is

ν [Se ] {A(t)}e + {Qe } =0 (3.67)
e
s

Ni dxdy = ∆e
RR
where Qe = 3
and

RR ∂Ni ∂Nj ∂Ni ∂Nj 
bi bj +ci cj
Se = ∂x ∂x
+ ∂y ∂y dxdy = 4∆e
.

3.5.3 The stator circuit equation in Galerkin’s form

Before combining the stator circuit equation into the global matrix equation, it is

necessary to formulate the circuit equation into Galerkin’s form. The stator circuit

equation as shown in equation (??) is

 
ZZ ZZ
l ∂A ∂A  di
V =  dΩ − dΩ + Ri + Lσ (3.68)
s ∂t ∂t dt
Ω+ Ω−

But in Galerkin’s form

3
∂A X ∂Ai
= Ni (3.69)
∂t i=1
∂t

Therefore, the stator circuit equation in Galerkin’s form is as shown below:

      
l T ∂A T ∂A dis
V = {Q} − {Q} + Ris + Lσ (3.70)
s ∂t Ω+ ∂t Ω− dt

3.6 Time Discretization

The field equation in iron cores (3.61), the circuit equation (3.70) and the motion

equation (3.31) are time dependent. These equations are needed to discretize in

the time domain. The method of time-discretization used is based on the following

equation:
t+∆t t
{A}t+∆t − {A}t
 
∂A ∂A
β + (1 − β) = (3.71)
∂t ∂t ∆t

Basically, there are three different types of time discretization.

1. Forward difference type, where the constant value of (β = 0)

 81

2. Backward difference type, where (β = 1) and

3. Crank-Nicholson Method, where (β = 1/2).

In this proposed method, backward difference type is used to discretize the time

domain because it has a good convergence rate. Therefore, equation (3.71) becomes
t+∆t
{A}t+∆t − {A}t

∂A
= (3.72)
∂t ∆t

3.6.1 Time discretization of the FEM equation in iron core

The Galerkin formulation of the equation of the iron core

 
∂A
ν[s] {A(t)} + σ [T ] =0 (3.73)
∂t

Substituting back ward difference time discretization equation (3.72), into equation

(3.73), " #
{A}t+∆t − {A}t
ν [s] {A}t+∆t + σ [T ] =0 (3.74)
∆t

Rearranging the equation (3.74) yields the following equation in time discretizing

model,
σ [T ] {A}t+∆t σ [T ] {A}t
+ ν [s] {A}t+∆t = (3.75)
∆t ∆t

3.6.2 Time discretization of the FEM equation in stator

conductor area

Galerkin’s form of the field equation in stator conductor area is

is
ν [S] {A(t)} + {Q} =0 (3.76)
s

applying the time discretization equation (3.72),

[Q]
ν[S][∆A]t+∆t + [∆is ]t+∆t = 0 (3.77)
S
 82

3.6.3 Time discretization of the stator circuit equation

In order to discretize the stator circuit equation in time domain, recall the stator

circuit in Galerkins form equation (3.70) as follow,

      
l T ∂A T ∂A dis
V = {Q} − {Q} + Ris + Lσ (3.78)
s ∂t Ω+ ∂t Ω− dt

Rearranging the equation (3.78)

   
l ∂A l ∂A dis
{Q} − {Q} + Lσ = V − R σ is (3.79)
s ∂t Ω+ s ∂t Ω− dt

applying the time discretization equation (3.72),

t+∆t
{is }t+∆t − {is }t

dis
= (3.80)
dt ∆t
t+∆t
{A}t+∆t − {A}t

dA
= (3.81)
dt ∆t
Substituting equations (3.80) and (3.81)into equation (3.79), we obtain the stator

circuit equation in time discretization model as

! !
l {A}t+∆t − {A}t l {A}t+∆t − {A}t
{Q} − {Q}
s ∆t s ∆t
Ω+ Ω− (3.82)
t+∆t t
{is } − {is }
+Lσ = Vs − Rt it+∆t
s
∆t

3.6.4 Time discretization of the motion equation

The motion equation (3.31) is discretized according to equation (3.72):

t+∆t
ω t+∆t − ω t

∂ω
= (3.83)
∂t ∆t

dω t+∆t
J = Te − Tf − Dω (3.84)
dt
ω t+∆t − ω t
J = Te − Tf − Dω (3.85)
∆t
 83

3.7 Linearization

In this model, field equations are nonlinear; linearization of these equations are

required before they can be combined with other equations of the system in a

global matrix equation. In this work, Newton-Raphson procedure is applied to

linearize the nonlinear system equations. Cubic spline interpolation algorithm is

used to represent the magnetization curves of the ferromagnetic materials.

Newton-Raphson form of the stator conductor field equation after time dis-

cretization is given by,

[Q] [Q]
ν[S][∆A]t+∆t
k+1 + [∆is ]t+∆t t+∆t
k+1 = −ν[S][A]k − [∆is ]t+∆t
k (3.86)
S S

in air gap,
t+∆t
ν[S][∆A]k+1 = −ν[S][A]kt+∆t (3.87)

in laminated stator iron core and solid rotor iron core,

σ[T ] σ[T ]
ν[S][∆A]t+∆t
k+1 + [∆A]t+∆t t+∆t
k+1 = −ν[S][A]k − [∆A]t+∆t
k (3.88)
∆t ∆t

in permanent magnets, the equation is

ν
ν[S][∆A]t+∆t t+∆t
k+1 = −ν[S][∆A]k + (Brx [ci ] − Bry [bi ]) (3.89)
2

and the stator circuit equation is given by

t+∆t Lσ [Q] [∆A]t+∆t

k+1
k+1 R + [∆is ]t+∆t
k+1 + l
∆t S ∆t (3.90)
L σ [Q] [A]kt+∆t Lσ t [Q] [A]t
= −[is ]t+∆t
k R − [i s ] t+∆t
− l + V + [is ] + l
∆t k S ∆t ∆t S ∆t
Hence, the final global equation in matrix form is

 [Q] 
ν [S] S
 ν [S] + σ [T
∆t
]
0 h
t+∆t t+∆t
i
 ∆Ak+1 [∆is ]k+1 (3.91)
 

 ν [S] 0 
[Q] l [Lσ ]
S ∆t
[R] + ∆t
 84

 
[Q]
−ν[S][A]t+∆t
k − [∆I]kt+∆t

 S 

 
 
σ[T ]
−ν[S][A]kt+∆t − [∆A]t+∆t
 
 k

 ∆t 
=



 ν 
 −ν[S][∆A]t+∆t
k + (Brx [ci ] − Bry [bi ]) 
2
 
 
 
 t+∆t

[Q] [A]k [Q] [A]t
−[is ]t+∆t
k R− Lσ
[i ]t+∆t
∆t s k
− S
l ∆t +V + Lσ
[i ]t
∆t s
+ l
S ∆t

3.8 Coupling the Rotor Movement with the FEM

After time discretization the motion equation can be represented by

 
k k−1 Te − TL
ω =ω + × ∆T (3.92)
J

and rotor displacement can be determined by the following equations:

∆θm = ω k × ∆T (3.93)

k k−1
θm = θm + ∆θm (3.94)

where θm is the rotor angle. This step is carried out in the post processing process.

The procedure is as follows;

• The electromagnetic torque, Te is calculated using Maxwell stress tensor

method.

• The motion equation (3.92) is used after time discretization to calculate the

rotor speed.

• The rotor displacement (rotor angle)is calculated according to equations

(3.93) and (3.94).

The rotor FEM mesh is moved according to the rotor angle θm at every time steps.
 85

3.9 Solving the Global System of Equation

The resulting global matrix equations are symmetric, positive definite and non-

zeros terms are clustered around the main diagonal. Hence, for iterative solver,

only the upper triangular coefficient matrix with nonzero elements is stored. Index

storage method describe in section (2.2.7) is used in this model. At each iteration

cycle of Newton Raphson method, the ICCG algorithm is used to solve the large

system of equations. Detailed formulations of solving methods are as follow.

The general form of global system of equation (3.91) can be represented as

∂X
AX + B =P (3.95)
∂t

where X is the unknown vector to be determined. For this proposed model, the

unknown vector X includes the magnetic vector potentials and stator currents.

A and B are sparse coefficient matrices. P is a column vector associated with

the input voltages. After time discretization using Backward Euler method, the

equation (3.95) becomes

Bk Bk
 
k k k
A + X = P + X k−1 (3.96)
(∆t)k (∆t)k

we can represent the equation (3.95) in Newton Raphson method as

k k
 
Bn−1 Bn−1
Jnk (∆X)kn = k
Pn−1 + k
X k−1 k
− An−1 + k
k
Xn−1 (3.97)
(∆t) (∆t)

where Xnk = Xn−1

k
+ (∆X)kn ; J is the Jacobian matrix and n is the iteration number

of the Newton Raphson method.

The initial calue of the solution X k can be predicted by the backward method
 k−1
dX
X0k =X k−1
+ (∆t)k (3.98)
dt

The error of the Newton Raphson method at the nth iteration cycle is

k k
   
1  k B n−1 k−1 k B n−1 k

(eN R )n = Pn−1 + X − A n−1 + Xn−1
 (3.99)
Nx  (∆t)k (∆t)k 
 86

where eN R is the number of the total unknown values. The iteration of the Newton

Raphson method will stop if

(eN R )n 6 eN R−allowed (3.100)

where eN R−allowed is the error tolerance of the Newton Raphson method. Within

each cycle of the Newton Raphson method, the ICCG algorithm is used to solve

the large system of algebraic equations.

3.9.1 ICCG algorithm for solving the algebraic equations

At each iteration cycle of the Newton Raphson method, the ICCG algorithm is used

to solve the set of algebraic equations. In the ICCG algorithm, it is necessary to

determine the approximate factorization of the coefficient matrix. The incomplete

Cholesky factorization is used in this method in order to reduce the storage and

computing time for factorization the matrix. For iterative solver, only the upper

triangular coefficient matrix with nonzero elements is needed to be stored. The

coefficient matrix of the algebraic equation set can be expressed as

J = LLT + E (3.101)

Where E is the error matrix and L is the lower triangular matrix of the incomplete

Cholesky factorization. Since the nonzero values of the coefficient matrix are mainly

distributed near the matrix diagonal, the conjugate gradient method is used to solve

the equation

L−1 Jnk L−T LT (∆X)kn

( " # )
k k (3.102)
D D
= L−1 Pn−1 k
+ n−1k X k−1 − Cn−1
k
+ n−1k Xn−1
k
(∆t) (∆t)

The step by step iteration procedures are carried out according to the following

formulations.

−1
Q1 = LLT r0 (3.103)
 87

T

−1
rm−1 LLT rm−1
βm−1 = (3.104)
T
rm−2 (LLT )−1 rm−2

−1
Qm = LLT rm−1 + βm−1 Pm−1 (3.105)

T

−1
rm−1 LLT rm−1
αm = T
(3.106)
Qm JQm
The iteration process is controlled by a vector r and it can be represented as
" #
k k
D D
k n−1 k−1 k n−1 k
Jn (∆X)kn
 k
r = Pn−1 + k
X − Cn−1 + k
Xn−1 − (3.107)
(∆t) (∆t)

The iteration error of the ICCG algorithm can be determined as

k k
   
1  k Bn−1 k−1 k Bn−1 k k k

(eICCG )n = Pn−1 + X − A n−1 + Xn−1 − J n (∆X)n
Nx  (∆t)k (∆t)k 
(3.108)

If (eICCG )n 6 eICCG−tolerance the iteration of the ICCG algorithm is termi-

nated. The block diagram of the solution process is shown in Fig. 3.16.

3.10 Determination of Time Step Size for Time

Stepping FEM
∂X ∆X
In the time stepping method, ∂t
(X is unknown variable) is substituted by ∆t

approximately. The system equation solved at times t1 , t2 , .....(tk = tk − 1 + ∆t)

step by step. The computation precision is dependent directly on the step size ∆t.

If the step size ∆t is too small, a lot of computing time is required. If the step size

∆t is too large, the error will be unacceptable. The choice of the step size is thus

critical in reaching a compromise between CPU time and acceptable errors [105].

Hence the best way is to adjust the step size automatically within the program.

The criteria developed for controlling the time steps depends on a knowledge of the

magnitude of the local truncation error [106]. It is necessary to estimate the local
 88

truncation error from a knowledge of the numerical solution only and to use this

information to control the step selection [107]-[108]. The general form of global

system of equation (3.91) can be represented as

∂X
AX + B =P (3.109)
∂t

where X is an unknown vector to be evaluated. A and B are coefficient matrices.

Equation (3.109) can be further expressed in a more simplified form as

∂X
= f (X, t) (3.110)
∂t

Integrating the above equation from tk−1 to tk gives

Z tk
Xk = Xk−1 + f (X, t)dt (3.111)
tk−1

If Backward Euler’s method is used, the approximate solution at the k th step is

Xk∗ = Xk−1 + (∆t)f (Xk∗ , tk ) (3.112)

Equation (3.111) can be expressed by using Talyor’s expansion at tk as

(∆t)2 00 ∗
Xk−1 = Xk − (∆t)f (Xk , tk ) + X (ξ ) (3.113)
2

where Xk is the exact solution and X 00 is the second derivative of X with respect

to t, tk−1 6 ξ ∗ 6 tk . Assuming f (Xk∗ , tk ) is equal to f (Xk , tk ) and subtracting the

equation (3.112) from (3.113), the local truncation error of X by using Backward

Euler’s method at the k th step is given by

(∆t)2 00
ε∗ = Xk − Xk∗ = − X (ξ) (3.114)
2

On the other hand, if Euler’s method is used at the k th step and its approximate

solution X # is,

Xk# = Xk−1 + (∆t)f (Xk−1 , tk−1 ) (3.115)

Using as above steps, the local truncation error is

(∆t)2 00 #
ε# = Xk − Xk# = X (ξ ) (3.116)
2
 89

where Xk is the exact solution at tk−1 6 ξ # 6 tk . Subtracting equation (3.114)

from (3.115) gives

(∆t)2 00 # (∆t)2 00 ∗
Xk∗ − Xk# = X (ξ ) + X (ξ ) (3.117)
2 2
00
Assuming X (t)) is continuous, there certainly exists a ξ between ξ # , ξ ∗ or
 

 ∗ #
ξ , ξ , hence
(∆t)2 00 # (∆t)2 00 ∗
X (ξ ) + X (ξ ) = (∆t)2 X 00 (ξ) (3.118)
2 2
00
Assuming X (t) changes very slowly between a small area ξ # , ξ ∗ or ξ ∗ , ξ # , one
   

00 00
obtains X (ξ ∗ ) ≈ X (ξ). Substituting equation (3.118), (3.117) into (3.114), the

local truncation error of X using Backward Euler’s method is

ε∗ ≈ (Xk# − Xk∗ )/2 (3.119)

The norm of the vector X = [x1 , x2 , ...., xN ]T is defined as follows;

1
kXk = (|x1 | + |x2 | + · · · + |xN |) (3.120)
N

where N is the number of unknowns. Thus the norm of ε∗ is defined as

ε = Xk# − Xk∗ /2 (3.121)

where ε is the error indicator.

According to the equations (3.119) and (3.114), the accuracy can be improved

by modifying Xk∗ as

Xk = Xk∗ + ε∗ ≈ (Xk# − Xk∗ )/2 (3.122)

Substituting equations (3.115), (3.112) into equation (3.122), the improved solution

is
∆t
Xk∆ = Xk−1 + [f (Xk−1 , tk−1 ) + f (Xk , tk )] (3.123)
2

The process of the adjusting step size is summarized as follows:

At the first step

 90

1. Input an initial step size (∆t)0 .

2. Use Euler’s equation (3.115) to obtain the solution X 00 .

3. use the Backward Euler’s equation (3.112) to obtain the solution X ∗ . This

is the approximate solution of X at this step.

4. Check the error ε by (3.121). If ε is larger than the allowed value εallowed ,

reduce the step size according to the following equation, then go back to

process 2.
r
0 εallowed
∆t = 0.8 (∆t) (3.124)
ε

Where 0.8 is a typical safety factor to ensure the step size is smaller than the

allowed value.

5. Calculate f (X1 , t1 ) for the next step according to equation (3.112) as

Xk∗ − Xk−1
∗


f (Xk , tk ) = (3.125)
∆t

At the k th step

1. Estimate the step size according to the error at the last step. Assume
00 00
X (ξ ∗ )k ≈ X (ξ ∗ )k−1 . From equation (3.114), we get
r
εallowed
(∆t)k = 0.8 (∆t)k−1 (3.126)
ε

2. Use Euler’s equation (3.115)to obtain the solution Xk# .

3. Use Backward Euler’s equation (3.112) to obtain the solution X ∗ .

4. Check the error by equation (3.121). If error is larger that the allowed value

εallowed , reduce the size according to the equation (3.126) and then go back

to the process 2.

5. Calculate f (Xk , tk ) for the next step using the equation (3.125).
 91

If the improved solution given in equation (3.123)is used as the approxi-

mate solution of X at each step,Xk# has to be obtained by solving the Euler’s

explicit equation (3.115). Simulated result with or without using time step adjust-

ing scheme will be shown in performance evaluation of the motor.

3.11 Conclusion

In this chapter, numerical modelling of the BLDC in time domain was presented.

Circuit-field-motion coupled time stepping finite element method was used. Detail

procedures of mathematical modelling of the BLDC motor in time varying electro-

magnetic field, coupling with the non-linear circuit equation and rotor movement

equations were described. Circuit equation included real time model of the inverter

switches. Although the switches are assumed to be ideal devices, the circuit model

developed is flexible enough to incorporate turn ’ON’ and ’OFF’ times.

An efficient method to generate the FEM mesh and rotate the rotor mesh was

presented. Using proposed mesh rotation technique, stator mesh and rotor mesh

are needed to generate only one time. When rotor is rotated, minor modification

to the nodes on the interface is needed in order to generate the topology of the

whole mesh. Hence, the computing time is greatly reduced.

Finite element formulations of mathematical equations of the time varying

magnetic field, circuit and motion using Galerkin’s method were described. Time

dependent FEM equations are discretized by Backward difference method. Since

the electromagnetic field of the problem domain are modelled in time varying mag-

netic field; the eddy current effect, the non-sinusoidal quantities and high order

harmonics can all be taken into account in this proposed model. Linearization of

the non linear electromagnetic field equations were performed by Newton’s Raphson

method coupled with cubic spline interpolation. Hence, saturation characteristics

of the ferromagnetic materials can be included into the system of equations. De-
 92

tailed procedures to solve the global system of equations using ICCG algorithm

were developed.

START

Time step k=k+1

Rotate FEM

Generate the coefficient matrix

X n = X n −1 + ∆X m

Calculate Newton Raphson

error tolerance, eNR
( eICCG )n ≤ eICCG −toleranc

No

( eNR )n ≤ eNR _ min

Calculate eICCG

Yes

Post processing
ICCG algorithm

Yes

No
Terminate the time
stepping?

END

Figure 3.16: Block Diagram of the time stepping solver

Chapter 4

Experimental Implementation of
the DSP Based BLDC Motor
Drive System

4.1 Introduction

The experimental set up for BLDC motor drive especially for sensorless micro mo-

tor type spindle motor is another challenging subject. Owing to certain inherent

features of permanent magnet motor with electronic controllers, the standard DC

motor test procedures are no longer suitable for BLDC drive system. A simple

and reliable test stand for exterior rotor BLDC motor type, HDD spindle motor

has developed in this chapter. The test equipments include a 8 poles 12 slots

spindle motor, dSpace1104 controller board, six steps IRF620 MOSFET voltage

source inverter, incremental encoder and rectified variable DC power supply. Soft-

ware implementation includes control program algorithm and drive functions based

on the DS1104 real time interface. A simple approach to determine the starting

switching sequences of the BLDC motor with respective rotor positions has devel-

oped. Measuring the motor performances in both steady state and transient are

presented.

93
 94

4.2 Hardware Implementation

The system hardware essentially consists of a spindle motor, a voltage source in-

verter, rectified variable DC supply, current sensors, an incremental encoder, a

DS1104 controller board and a host PC computer. Fig. 4.1 shows the hardware

set up in the laboratory and Fig. 4.2 shows the schematic diagram of the hardware

equipment. The individual functions and characteristic features of the hardware

components are briefly described in following subsections.

Figure 4.1: Photograph of hardware set up in the Laboratory

4.2.1 The Variable DC supply

The variable DC source is used for supplying the motor and the driving voltage

source for the inverter. The spindle motor requires variable power supply ranges

from 5V to 12 V whereas the inverter requires 12 V fixed DC supply for driving

the MOSFETs and 5V fixed DC supply for MOSFET’s drivers.

 95

Figure 4.2: Schematic diagram of the hardware equipments

4.2.2 The voltage source inverter

It is well known that an ideal semiconductor device for power switching should

exhibit a low forward-voltage drop to reduce forward-conduction losses and a high

switching speed to reduce switching losses to a minimum. In addition, the switch

should ideally have a zero voltage drop during the on-state while providing infinite

resistance in the off-state. In this experimental set up, N-channel 200V, 0.8 ohm

static drain-source on resistance, 6A drain current (at 25 C, continuous) IRF620

MOSFET is used to implement the voltage source inverter for the prototype BLDC

motor. In the switching of high voltage floating devices (such as MOSFET) within

a three phase bridge configuration, the driving voltage to a device must typically be

higher than the drain or collector voltage. If a device is referenced to the high side,

the required voltage will be greater than the supply. The driving signal is supplied

from the logic circuits, which are normally referenced to the ground potential;

hence the signals need to be level shifted to the source or emitter of the device.

As the voltage across the lower device switches between the two rail voltages, a

high-performance level changer and isolation is required. Therefore, in order to

ease the design tasks, IR2110 high and low side drivers are used to control one arm
 96

of the a MOSFET bridge circuit where the high side is floating with respect to the

low side.

The key feature of driver IR2110 is their suitability for applications up to

500V, with a 2A peak output capability and a 25ns switching speed. To protect

the system, the device will shut down if an under-voltage is detected on the logic

supply; this is critical for safe operation during switching of the circuit. Circuit

diagram of hardware implementation of voltage source inverter is shown in Fig. 4.3.

The voltage for the high-side driver is provided by the network capacitor C2 and

and diode D1. For example, when the MOSFET S1 is on, capacitor C2 is charged

via D1. The diode prevents the capacitor from discharging when the device is off

and when the voltage of the collector approaches the supply voltage, Vs. The time

during which the current can be supplied to the high side driver is limited by the

current drain and by the size of the capacitor. A typical value of 1000pF and a

current drain of 5mA can be sustained at an operating frequency of up to 10kHz.

Detailed specifications of hardware equipment used in this inverter circuit are given

in Appendix E.

4.2.3 Spindle motor

An in-hub type 8 pole 12 slots spindle motor is used in this experiment. Principal

motor specifications are listed in Table 4.1. Motor is mounted with an incremental

encoder to sense the rotor position. Disc platters are used as loads for the motor

and mounted on the motor hub the same way used in the hard disk drive. Fig. F.1

shows the hardware set up of the spindle motor with the encoder. Electronic

commutation is carried out according to the rotor position and switching sequences

of the controller.
 97

Figure 4.3: Circuit diagram of the voltage source inverter

 98

Figure 4.4: Hardware set up for motor and encoder

Table 4.1: Motor specifications

voltage 12V

No of poles 8

No of slots 12

Rated speed 7200rpm

Stator outside diameter 28mm

Rotor outside diameter 30mm

Stack length 4.5mm

Air gap 0.2mm

Magnet N dF eB

4.2.4 Incremental encoder

In hub-type disk drive motors are mostly designed to rotate only two or three disc

platters. Therefore, an encoder mounted with this type of spindle motor should

have relatively very small holding torque and inertia. In this type, SCANCON
 99

incremental encoder, mirco hollow shaft type 2MCH model is used. It has a resolu-

tion of 1024 lines per revolution; mass moment of inertia is 0.25gcm2 and starting

torque is < 0.005N m. Detailed specifications are attached in Appendix F. The

output from the encoder is in pulse form and it cannot provide the rotor position

at starting. Encoder is the incremental type therefore it is needed to detect the

index or home signal. DS1104 encoder interface provides the feature to detect

the index signal. Adding an index channel to an incremental encoder allows the

encoder to provide absolute rotor position.

4.2.5 DS1104 controller board

DS1104 controller board is specifically designed for the development of high speed

multivariable digital controllers and real time simulations in various fields. It is a

complete real-time control system based on a 603 PowerPC floating point processor

running at 250MHz. For advanced I/O purposes, the board includes a slave-DSP

subsystem based on the TMS320F240 DSP microcontroller. For rapid control pro-

totyping, specific interface connectors and connector panels provide easy access to

all input and output signals of the board. The DS1104 controller board is designed

like a PC expansion card so that it can be inserted into the PCI slot of the PC,

together with the PC video card or sound card. Fig. 4.5 shows the overview of the

architecture and the functional units of the DS1104. The controller board provides

the following features:

Master PPC which is the main processing unit of the board and consists of

• A PowerPC 603e microprocessor with running at 250 MHz, containing 16

kilo-bite L1 data cache and L1 instruction cache

• An interrupt controller

• A 32 MB synchronous DRAM controller

 100

• Timers and

• A PCI interface (5V, 32 bit, 33 MHz).

Figure 4.5: Overview of DS1104 Feature

The master PPC controls the I/O features of analog to digital (ADC) unit, digital to

analog (DAC) unit, bit I/O unit, incremental encoder interface and serial interface.

These features can be fully programmed from the real time interface by simulink

or hand coded C program. The ADC units convert the analog feedback signals of

10 V into digital format so that the DSP can read the measured quantity. In this

experiment, motor input voltages and stator currents are measured via the ADC

converters. Alternatively, DAC units are used for the digital to analog signals of

the range of 10 V so that all these signals can be displayed on the oscilloscope.

Bit I/O unit can be used for digital I/O to and from the experiment and real

time interface of the DS1104 control board. It contains 20 bit I/O pins. The I/O

pins are called IO0.. IO19. The direction of each bit can be selected individually by

software control program. In this experiment, bit I/O unit is used as an interface
 101

for control program and inverter switching input signals. Bit IO0 to IO5 are used to

provide the required six input signals for inverter circuit to switch on the respective

MOSFET.

There are two input channels for two digital incremental encoders (Inc1 and

Inc2) in the master PPC of the DS1104. Both of theses two encoder interfaces

can support single ended TTL and differential signals. It can provide 1.65 MHz

maximum encoder line count frequency that means it can measure up to 1,650,000

encoder lines per second. Since the DS1104 perform internally a fourfold subdi-

vision of each encoder line, the board can therefore handle count frequencies of

up to 6.6 MHz. The DS1104 is equipped with a 24 bit position counter. Due to

the fourfold subdivision of each encoder line, the counter allows to measure up

to 222 encoder lines in the range -221...+221-1. The count direction depends on

the encoder’s rotation direction. The counter can be reset by the encoder’s index

pulse. Each encoder channel provides an index input. The input is connected to

the DS1104 interrupt control unit. The index signal can be extracted and written

the position information to the position counters immediately when an index is

found. Motor speed can be calculated via encoder interface. In this experiment,

“Inc1” incremental encoder interface is used to measure motor speed and sense

the rotor position. Interrupt controller provides various hardware and software

interrupts. In this work, encoder index signal is extracted by encoder interrupt.

Host interface is used for setting up the DS1104, downloading the programs and

transferring runtime data to and from the host PC.

The DS1104 is equipped with two memory: global memory and flash memory.

Global memory provides 32 MByte synchronous DRAM for experiment. Flash

memory provides of 8 Mbyte which is divided by the 4 blocks of 2 Mbyte each. The

application for this experiment is very simple therefore global memory is enough

to use for real time implementation.

 102

4.3 Software Implementation

Spindle motors are operated in two phases on operation. In order to run the motor

in correct direction, it is necessary to switch on the correct and exact two phases

of the motor. Therefore, rotor position should be known a prior and then it is

required to determine the switching sequences. If the switching sequences are not

correct, motor may not run properly. Fig. 4.6 and Fig. 4.7 show typical input phase

voltage waveforms and switching states for the motor with a wye connected stator

winding. Possible switching sequences are shown in Table 4.2. In practical cases,

the motor may start at any of these switching conditions.

Based on these consequences, simple and reliable control program is imple-

mented in this experiment. Firstly, detect the initial rotor position by encoder

interface of the DS1104. And then switch on any predefined two phases (for ex-

ample phases A and B of switching sequence no(1)) and check whether the motor

starts to run or not. If the motor rotates as the desired direction, suppose that the

proposed switching sequences are the correct sequence to run the motor. If not,

move on to another two pairs of switching states and check again. It is needed to

test only six steps and one can decide the switching sequence very easily.

Figure 4.6: Wye connected stator windings

 103

Figure 4.7: Typical input voltage waveform and switching states

Table 4.2: Possible switching sequences

sequence sequence sequence sequence sequence sequence

no.1 no.2 no.3 no.4 no.5 no.6

AB AC BC BA CA CB

AC BC BA CA CB AB

BC BA CA CB AB AC

BA CA CB AB AC BC

CA CB AB AC BC BA

CB AB AC BC BA CA

Bit I/O unit is used to operate the inverter circuit driver ,IR2110 to switch

on the respective MOSFET (from S1 to S6). In this experiment, bit IO0 is used to

provide the switching signal for high end MOSFET, S1, IO1 is used to provide the

switching signal for low end MOSFET, S2 and so on. In other words, if the value

of bit IO0 is 1, S1 will be turn on and if bit IO0 is 0, S1 will be turn off. Program

flow chart for switching sequence control is given in Fig. 4.8. Main control program

and interrupt service routine program flow charts are shown in Fig. 4.9. The entire

control program is written in C language.

The execution processes for developed C program into DSP real time interface

are as mentioned below. Real time RTILib is used to implement the hand-coded C
 104

program. Downloading and starting applications on the real time hardware can be

done by the Platform Manager as a batch client. After downloading the program

into DSP, the users can easily modify the process variables and control parameters

via on line real time interface.

4.4 Measuring Motor Performances

Motor performance can be measured using DS1104 controller board in real time

interface and an oscilloscope. The following subsections describe how to measure

the motor performances in both transient and steady state conditions.

4.4.1 Rotor position sensing and switching sequence de-

tecting

Rotor position is measured every sampling period of the DSP control program via

encoder interface. Measured rotor position serves as an input signal for the control

program. Outputs from the control program are switching signals for the inverter.

These are sent to the inverter via bit I/O unit. Hence, switching states can be

detected by measuring the output of bit I/O unit. It is required to check the

switching states before connecting the motor with the inverter circuit to protect

malfunction of the inverter switching sequences.

4.4.2 Measuring back-emf

Measurement of the back-emf in BLDC motor mostly uses prime mover such as DC

motor or stepper motor which is directly mounted with the test motor. The test

motor is allowed to run the rated speed and then measures the induced voltage

within the motor winding (back-emf) via oscilloscope or other data acquisition

units. In most cases, HDD spindle motors used in PC computers are designed
 105

to run two disc platters. Mounting both encoder and prime mover in this type

of spindle motor may lead to improper running condition because encoder has its

own mass moment of inertia and prime mover motor has holding torque itself.

Motor will be overloaded and sometimes it may not start because of high holding

torque. In order to overcome these shortcomings and difficulties, a simple method

is implemented in this experiment. Firstly, motor is allowed to run the rated speed

to few minutes and then switch off the motor power supply. Although there is

no current in the stator winding, the motor will be run few milliseconds at this

instant because of the rotor high inertia. Therefore, induced voltage in stator

winding which is called Back-emf can be measured at this instant. When the

motor speed goes down to zero, the measured bacemf signals will be distorted and

go down to zero. However, only a few cycle is enough for measuring back-emf since

it is repeated for every 360 electrical degree cycle.

4.4.3 Measuring stator current

Sufficiently accurate measurements of the motor phase currents are essential to

ensure high-performance in the BLDC drive system. In this experimental set-up,

two simple and reliable current sensing methods: (a) measuring by using clamp

meter and (b) by using shunt resistors, are utilized to measure the phase-A and

phase-B currents. In current sensing by clamp meter, the measured currents from

the clamp meter are converted into voltage signals scaled to the range 10V by

using resistors of appropriate resistance values. These voltage signals are then fed

into the A/D converters of the DS1104.

4.4.4 Measuring motor speed

Motor speed is measured using the encoder interface of the DS1104 controller board.

The speed is calculated by the encoder line count. Both transient and steady state
 106

conditions can be measured easily. For disk drive motor, starting time to reach

the rated motor speed is one of the important parameters for drive precisions and

implementation. In this experiment, motor speed vs. time under the following

conditions

• motor starting with different supply voltages

• motor starting with different rotor inertias

• motor starting with drive limits.

 107

START

First step?
No
Yes

Switch on phase AB
(Sequence no.1)

Yes -Use this squence from now on

Rotor start to rotate?
-Note down sequence no = 1

No

Switch on phase AC
(Sequence no.2)

Yes -Use this squence from now on

Rotor start to rotate?
-Note down sequence no = 2

No

Switch on phase BC
(Sequence no.1)

Yes -Use this squence from now on

Rotor start to rotate?
-Note down sequence no = 3

No

Switch on phase BA
- Check sequence no
(Sequence no.1)
- Switch on according to
the predefined sequences

Yes -Use this squence from now on

Rotor start to rotate?
-Note down sequence no = 4

No

Switch on phase CA
(Sequence no.1)

Yes -Use this squence from now on

Rotor start to rotate?
-Note down sequence no = 5
No

Switch on phase CB
(Sequence no.1)

Yes -Use this squence from now on

Rotor start to rotate?
-Note down sequence no = 6

END

Figure 4.8: Flow chart of switching sequences control program

 108

START

- Initialize DSP registers and memory

- Initialize digital bit I/O units and encoder
- Initialize control variables
START
Interrupt Service Routine

Start sampling rate timer

- Start ADC units
- Measure voltages and currents
by ADC

Timer interrrupt No
arrive?
- Read rotor position from encoder
- Calculaterotor speed
Yes - Use low pass filter to eliminate
the noises of the speed signals
Interrupt serives routine

start switching sequence control program

- Generate the switching signals
No - Send the signals to the inverter by bit I/O
Reset DSP?

Yes EXIT
Interupt Service Routine
END

Figure 4.9: Main control program and interrupt service routing flow charts
Chapter 5

Performance Analysis of the

BLDC Motor

5.1 Introduction

Detailed formulations of mathematical models and experimental methods for ex-

terior rotor BLDC motor have been developed in the previous chapters. In this

chapter, the investigations of steady state operations and dynamic behaviours of

the BLDC motor under no-load and loaded conditions are presented. In the time

stepping FEM simulation, time step sizes are adjusted as mentioned in Section

3.10. In order to validate the numerical models, experimental results are described

and compared with simulation results obtained from FEM analysis.

5.2 Steady State Analysis of the BLDC Motor

Normally, the transient solver begins with a flat start by initializing all state vari-

ables to zero. This results in a large DC components of current and several hundred

integral time steps are needed before reaching its steady state [109]-[110]. While

only the steady state solutions may be of interest, it may take many cycles of

computation before starting transient disappear. This excessive computation time

required to reach the steady state solution has prompted an interest in the tra-

ditional current fed two dimensional magneto-static FEM. One motivation in this

109
 110

work is that the steady state solution from the two dimensional magneto-static

FEM model is used as an initial condition for the time stepping solver. The steady

state model of the BLDC motor includes four main parts.

1. Mesh generation

2. Pre-computation by two dimensional magneto-static FEM

3. Computation in time domain by time stepping FEM

4. Post processing

Detailed descriptions for each of these parts are discussed in the next sections. The

program structure is given in Appendix G.

5.2.1 Mesh generation

FEM mesh is generated according to the procedures which have been described in

Section 3.4. For the static analysis, the rotor is assumed to be psuedo-stationary.

Stator mesh and rotor mesh are generated separately and connected by periodic

boundary condition. After mesh generation, pre-computation by two dimensional

magneto-static FEM is started.

5.2.2 Pre-computation using two dimensional magneto-static

FEM

Pre-computation of initial conditions for the time domain solver is carried out by

using current fed two dimensional magneto-static FEM which has been developed

in Chapter 2.

Recall the non-linear electromagnetic field equation (2.18), which is

   
∂ ∂A ∂ ∂A
ν + ν = −J0 + Jm (5.1)
∂x ∂x ∂y ∂y
 111

where ν is the reluctivity of the material, J0 is the applied source cur-

rent density and Jm is the equivalent current density of the permanent magnet.

Galerkin’s method is employed for finite element formulation. Detailed procedures

of Galerkin’s formulation for the above two-dimensional eddy complex eddy current

equation (5.1) have been discussed in Chapter 2. Hence, the final global system

element matrix equation for the electromagnetic field equation (5.1) is

  
b2i + c2i bi bj + c i c j bi bk + c i c k Âi
1 
 bi bj + c i c j b2j + c2j bj bk + cj ck   Âj 
 
4µe ∆
bi bk + c i c k bj bk + c j c k b2k + c2k Âk
       (5.2)
1 ci bi
e∆   νµo 
= J0  1  + M x  c j  − M y  b j 
   
3 2∆
1 ck bk
Newton Raphson’s method is used to linearize the non-linear equations. ICCG

algorithm is used to solve the global system of equations. After obtaining a solu-

tion (magnetic vector potential A) from magneto-static FEM, an instant of time is

chosen and the time stepping solver is started.

5.2.3 Computation in time domain by time stepping FEM

In this step, time domain models of the BLDC motor which have been developed

in chapter 3 are used. Computational analysis in time stepping FEM includes

modelling of the electromagnetic field, circuit and motion equations in time domain,

finite element formulation, time discretization, linearization of the nonlinear field

equations, simulation of rotor rotation and solving the global system of equations

as discussed before. At the first time step, the initial values are the solutions got

from the previous magneto-static FEM.

 112

5.2.4 Post processing

The post processing is the process to compute the desired parameters of motor

performances. Post processing includes computation of magnetic flux density, sta-

tor current, torque, force, motor speed etc., with various time frames. Evaluations

of steady state performances in a BLDC motor using the proposed steady state

model are discussed in the next section.

5.3 Evaluation of Steady State Performances

This section describes the methods of computation of important steady state per-

formance indicators such as determination of electromagnetic torque, back-emf,

stator current, torque-speed characteristics and computation of cogging torque us-

ing proposed steady state FEM model. Experimental results are given to compare

with the computational results in order to verify the validity of the proposed steady

state model.

5.3.1 Calculation of stator current

In time stepping model, the field equations, the circuit equation and the motion

equation are solved simultaneously. The inputs are the stator voltages and rotor

positions whereas the outputs are magnetic vector potentials and stator currents.

Hence, no extra step is needed to calculate the stator current. Fig. 5.1 shows the

input stator voltage waveforms and Fig. 5.2 is the corresponding output stator

current waveform at no load condition computed in time domain. Fig 5.2 shows

the computation and experimental results of the stator current waveforms at no

load condition. It can be seen that the simulation and experimental results are in

good agreement. According to the Fig. 5.2, the transient time stepping solver can

also be started with the initial conditions quite close to the steady state solution.
 113

There is still a transient at the first cycle due to the fact that the 2D static FEM

does not correctly account for eddy current effect and ignores the motion effects.

Figure 5.1: Input voltage waveform against time

5.3.2 Computation of electromagnetic force and torque

To predict the performance of an electrical machine, precise knowledge of force

and torque is important. Electromagnetic forces and torques can be computed in

a variety of ways. The three basic methods often employed with finite element

analysis are Lorentz force, Maxwell’s stress tensor and Energy methods [111]-[112].

Each method has its own merits. In this work, Maxwell’s stress tensor method is

used to calculate the torque and force of the BLDC motor. Maxwell’s stress tensor

method is based on the fact that the field distribution inside a closed surface in

along the air gap remains unchanged even if the external sources are removed and

replaced by a distribution of currents [113]. For a ferromagnetic material of volume

 114

Figure 5.2: Computed stator current waveforms at no load condition

Figure 5.3: Experimental and computational results of stator current

 115

v, the volume current Jv can be defined by

∇×M
Jv = (5.3)
µ0

And then the force density can be found by

fv = Jv × B (5.4)

The force density can be represented as follow by using Ampere’s law ∇ × H = J

and constitutive relationship B = µ0 H , therefore

 
B
fv = ∇ × ×B (5.5)
µ0

After performing a few mathematical calculations, the force density can be ex-

pressed as the divergence of the Maxwell tensor σ,

fv = ∇.σ (5.6)

where the Maxwell tensor σ is

1 1
B 2~n


σ= (B.~n) B − (5.7)
µ0 2µ0

Since

B = Bn~n + Bt~t (5.8)

and

B 2 = Bn2 + Bt2 (5.9)

Substitute equations (5.8) and (5.9) into equation (5.7) yields

1 1
(Bn Bt ) ~t + Bn2 − Bt2 ~n


σ= (5.10)
µ0 2µ0

where Bn and Bt are the normal and tangential components of flux density respec-

tively.

From Maxwell formulation, the electromagnetic force acting on the rigid body

can be calculated by
Z Z
F~ = f~v dv = ∇.σdv (5.11)
v v
 116

According to the Divergence theorem, the volume integral can be changed into

surface integral as
I
F~ = σ.dS (5.12)
s

where S is the surface surrounding the ferromagnetic material where forces are ex-

erting. Substitute the tensor equation (5.10) to the electromagnetic force equation

(5.12) we get,

I  
1 1
F~ = (Bn Bt ) ~t + 2 2


Bn − Bt ~n dS (5.13)
µ0 2µ0
S

When equation (5.13) is applied to the calculation of the torque of a BLDC

motor, a closed integration surface that surrounds the stator in free space along

the air gap must be chosen. For a two dimensional case, the surface integral is

reduced to a line integral along the air gap. If a circle of radius r is taken as the

integration path, the force is obtained from the following:

Z2π  
1 1
F~ = (Br Bθ ) θ~ + 2 2


Br − Bθ ~r rdθ (5.14)
µ0 2µ0
0

Since, for the two dimensional case,

Br = Bn (5.15)

Bθ = Bt (5.16)

dS = rdθ (5.17)

If the solution would be exact, the force would be independent of the integration

radius r when r varies within the air gap. However, the calculated force depends

greatly on the choice of the integration radius, and to achieve satisfactory accuracy,

the force is calculated over the closed integration circular path that surrounds the

rotor in free space along the air gap [112]. Due to the fact that the true force is
 117

independent of the radius we obtain by integrating (5.22)

Zrs Z2π  
1 1
F~ (rs − rr ) = (Br Bθ ) θ~ + 2 2


Br − Bθ ~r rdθdr
µ0 2µ0
rr 0
Z   (5.18)
1 1
(Br Bθ ) θ~ + 2 2


= Br − Bθ ~r dS
µ0 2µ0
Sag

Z  
1 1 1
F~ = (Br Bθ ) θ~ + 2 2


Br − Bθ ~r dS (5.19)
(rs − rr ) µ0 2µ0
Sag

where rr and rs are the outer and inner radii of the circular ring; Sag is the

cross-sectional area of the integration path; Br and Bθ are the components of the

flux density respectively. In the two dimensional electromagnetic field, the electro-

magnetic force for the rotation of rotor is generated by the tangential component

of force. Hence the electromagnetic force acting on the rotor is

Zrr X
M Z2π  
1 1
Fθ = lm Br Bθ rdθdr
rr − rs m=1
µ0
rs 0
(5.20)
M ZZ
1 X
= lm Br Bθ dS
µ0 (rr − rs ) m=1
Sag

where M is the triangular elements on the integration path of the air gap, lm is the

axial length of the machine.

Finally, the electromagnetic torque can be obtained by integration along the

circular path
Zrr X
M Z2π
1
Te = lm (rBr Bθ )ds (5.21)
µ0 (rr − rs )
rs m=1 0

Calculated flux plot at static and after rotation 1000 step are given in Fig. 5.4

and Fig. 5.5. The calculated electromagnetic torque computed using Maxwell’s

stress tensor method at no load and loaded conditions at each time step is shown

in Fig. 5.6.
 118

Figure 5.4: Calculated flux plot at static position

Figure 5.5: Calculated flux plot after rotor is rotated 1000 steps
 119

Figure 5.6: Electromagnetic torques at no load and loaded conditions

5.3.3 Determination of torque-speed characteristics

BLDC motors are used in many constant and adjustable speed control applications,

because their speed and torque can be easily controlled. Hence, the best way to

gain an understanding of the performance characteristics of a BLDC motor is to

study the speed and torque relationship. This curve can represent the steady state

capability of the motor in driving various types of loads. The most basic functional

requirement is to ensure that the motor has enough torque at all speeds, from zero

to full speed, to accelerate the load from standstill and maintain full speed without

exceeding any thermal or electrical limits.

Torque vs. speed characteristic of the proposed BLDC motor with 4V and

6V supply voltage is presented in Fig. 5.7. It can be seen that the torque-speed

curve is merely straight line; when torque is increased, the speed is decreased. This

phenomena can be explained by the following steady state equations of the motor.

The equivalent circuit equation of the BLDC motor in steady state condition is

V = IR + e (5.22)

Where V is the supply voltage, I is the stator current, R is the stator conductor
 120

resistance and e is the back-emf and can be represented as

e = kφω (5.23)

Motor torque in steady state can be expressed as

T = kφI (5.24)

In equations (5.15) and (5.16), kφ is the back-emf constant and torque constant. If

the motor is operating at no load, very small current is drawn from the supply and

generally it is neglected for analysis. According to the equation (5.12), if there is no

voltage drop in the stator resistance and the motor will accelerate until the back-emf

equals to the supply voltage. This occurs at the no load speed which is the highest

speed of the motor. Torque speed characteristic of the proposed BLDC motor is

shown in the Fig. 5.7. When load torque is applied, current is drawn from the

supply, resulting in a voltage drop in the stator conductor resistance. This voltage

drop is possible only if the back-emf, falls to the value (V − IR), and according

to the equation (5.13), the speed must fall. The drop in e is proportional to the

current, and therefore to the torque because motor torque is directly proportional

to the current according to the equation (5.22).

The results of torque-speed characteristic curve (Fig. 5.7) and current vs.

torque curve (Fig. 5.7) show the relationship between motor speed and the torque

is linear and motor speed is directly proportional to the supply voltage in this type

of BLDC motor. In addition, torque and current are directly proportional to each

other. Based on these results it can be seen clearly that controlling of these two

parameters (torque and speed) is straightforward. Motor speed can be controlled

just simply by varying the supply voltage where the back-emf is kept constant. On

the other hand, motor torque can be controlled by varying both supply voltage

and back-emf. This results are used as the application of simple torque and speed

cascaded control for BLDC motor drive system in the next chapter.
 121

Figure 5.7: Torque-speed curve of the motor

Figure 5.8: Current and torque relationship

 122

5.3.4 Computation of cogging torque

In a slotted permanent magnet motor, cogging torque is one of the key components

of torque pulsations. High cogging causes not only variations in motor speed and

torque but they are also a source of motor vibration and noises [114]-[116]. It is well

known that cogging torque is a magneto-static effect. It is caused by the air-gap

permeance variation due to slotting effect; the rotor magnet is attracted to certain

positions where the effective permeance is maximum. This torque component is

typically independent of stator current excitation. The precise determination of

the cogging torque is of great importance for the design of a slotted permanent

magnet motor especially for disk drive spindle motor where constant motor torque

is highly demanded [117].

Generally, cogging torque is computed according to the magnetic field com-

putations when there are no armature currents and no magnetic field excitation,

respectively. However, this method cannot take into account the effect of the sat-

uration of iron.

In this work, a new approach has been used for computation of cogging torque

including the effect of iron saturation. The proposed algorithm to compute the

cogging torque synchronously with the time stepping solver is: in the time domain

solver, after the normal time stepping computation has been done, the reluctivity of

the materials are fixed and the currents in stator winding are set to zero. And then,

the time stepping computation is simulated again. Cogging torque is computed by

Maxwell’s stress tensor method and the same function block is used for computing

the electromagnetic torque. However, the motor current is zero within cogging

torque computation process. Hence, cogging torque obtained by this method has

included saturation effect. No other field computation is required. This method

can be applied for back-emf and inductance computation.

Fig. 5.9 shows the results of computation of cogging torque profiles with var-
 123

ious magnet strengths. It is found that varying the magnet strength with the

remanent flux density (7.0T, 6.5T, 6.0T, 5.5T, 5.0T, 4.5T as shown in Fig. 5.9)

affects only the peak value of cogging torque. Since there is no change in the phys-

ical geometry of the motor, the shape of cogging torque profiles remain unchanged.

Although reducing the magnet strength lowers the cogging torque, it can also de-

grades the performance of the motor at the same time because the electromagnetic

torque of a motor is directly proportional to the amount of flux produced by the

permanent magnets.

Figure 5.9: Cogging torque profiles for 8p 12s spindle motor with different magnet

strengths

5.3.5 Calculation of back-emf

A good estimate of back-emf is required because it is an important indicator of

the ability of the motor to produce torque and speed. It also requires for accurate

simulation of the motor operating with its controller, and also for the determination

of the current waveform for the correct control strategy. In time stepping solver,

back-emf is computed while stator current is set to zero at each time step as
 124

mentioned in Section 5.3.4. The back-emf as shown in equation (3.30) is given by

ZZ
lN ∂A
e=− ds. (5.25)
S ∂t
S

And the back-emf equation after time discretization is given by equation (3.92)
! !
l {A}t+∆t − {A}t l {A}t+∆t − {A}t
e = {Q} − {Q} . (5.26)
s ∆t +
s ∆t −
Ω Ω

According to the equation (5.26), back-emf can be calculated after the magnetic

vector potential values at each time step have been determined in each nodes of

the elements. Fig. 5.10 shows the computed back-emf values where input voltage is

4V and motor speed is 7200 rpm. It can be seen that there is not much difference

between the magnitude of input voltage and motor back-emf when the motor is

in its steady state condition. This is one of the spindle motor characteristics, it

requires very small current at normal running conditions. The voltage drop across

the motor resistance is very small and the supply voltage is nearly the same as

its induced voltage of stator winding when the motor is running in steady state

condition.

Figure 5.10: Calculated back-emf with corresponding rotor angle

 125

5.4 Performance Evaluation with and without the

Time Steps Adjustment Scheme

In the time stepping FEM, the system equations are solved at specific time t1 , t2 ...

etc. step by step. The computation precision is directly dependent on the step size

δT . If the step size is too small, a lot of computing time is required. In turn, if the

the step is too large, the error will be unacceptable. Hence, the step size should be

adjusted to compromise between CPU time and acceptable errors. Adjusting the

time step based on local truncation error is adapted in this work. Detail procedures

have been described in Section 3.10. Fig. 5.11 shows the comparison of simulated

cogging torque waveforms with and without using the time adjustment scheme.

In the computation of Fig. 5.11(a), the minimum step size is 0.006 ms and the

average step size is 0.0125 ms. If all other conditions are the same and the step

size is fixed at 0.0125 ms, the computed cogging torque waveform is as shown in

Fig. 5.11(b). The same condition is tested to calculate the electromagnetic torques

on load condition of the motor and simulated results are shown in Fig. 5.12. It is

observed that if the number of time step sizes in the integration time is the same,

the error caused by the fixed step size method is significantly greater.

5.5 Transient Analysis of the BLDC Motor

Transient behaviours of the BLDC motor are simulated by using the the time step-

ping FEM model. Using the proposed transient time stepping FEM solver, motor

dynamics in step voltage variation, changing of the mechanical load torque and

locked rotor conditions are analyzed. Experimental results at transient conditions

are described and compared with simulation results for validation of the proposed

model.
 126

Figure 5.11: Simulated cogging torque with and without step size adjustment

scheme

Figure 5.12: Simulated load torque with and without step size adjustment scheme
 127

5.5.1 Step voltage variation

Step voltage variation could arise when the motor speed is controlled through

supply voltage. Fig. 5.13 - Fig. 5.17 show the current, torque and speed transients

when the motor is loaded and an excessive drop in the applied voltage occurs; a

drop from 4.5 V to 2.5 V at 5 ms is being assumed.

Figure 5.13: Step voltage change

Figure 5.14: Speed response during step voltage change

 128

Figure 5.15: Simulated back-emf during step voltage change

Figure 5.16: Stator current transient during step voltage change

 129

Figure 5.17: Developed torque during step voltage change

From Fig. 5.16, it can be seen that the stator current goes up from 0.5 A to

1.45 A in reverse direction at this instant and then it slowly goes down to zero. It

takes 30 ms to reach to zero value. After that it goes up again from 30 ms onwards.

The motor speed also drops exponentially from the rated speed, 7200 rpm to 5800

rpm within 35 ms.

Possible reason for this phenomenon is that although the supply voltage drops

suddenly, the generated back-emf could not be adjusted simultaneously at this in-

stant (Fig. 5.15) because of the magnetic saturation as well as the rotor inertia,

so that according to the stator circuit equation (5.14), the stator current will be

increased from 0.5A to 1.45 A in magnitude. However, the supply voltage is lower

than the back-emf at this instant and therefore the motor current will be negative

flowing in the reverse direction. A few seconds later, the generated back-emf is

decreased corresponding with the speed reduction and at 25 ms the back-emf is

equal to the supply voltage where the current will be zero. When the back-emf is

lower than the supply voltage, the current will be increased to positive direction

again. This can be seen clearly within the time period between 30 ms to 40 ms

in Fig. 5.15. The result in Fig. 5.14 shows that the motor speed drops exponen-
 130

tially, because of the effects of the rotor inertia and stator winding inductance,

which causes distortion at low speed, and the magnetic saturation coupled with

the demagnetizing effect of the phase currents at low torque.

When the motor current flows in reverse direction, the motor will be generated

negative torque as shown in Fig. 5.17). It can be seen clearly that when the motor

current suddenly increased up to 1.45A in reverse direction, the motor torque

also decreased to negative value simultaneously at this instant. It is likely due

to the linear relationship between the motor torque and speed in BLDC motor.

In addition, the reversal of the torque eventually decelerates the motor to a speed

which is needed to match the new level of the supply voltage. The case shown is

for the motor driving a load requiring constant torque. This is why the current

eventually recovers its original value before the change in the supply voltage occurs.

It needs to be mentioned here that, any drop in voltage results in a negative

current. The drop has to be such magnitude that the new voltage level is lower

than the generated back-emf. The transient responses where the supply voltage is

dropped from 4.5 V to 3.5 V are shown in Fig. 5.18. The time instant when the

motor voltage drop from 4.5V to 3.5V, motor back-emf is lower than the supply

voltage as shown in Fig. 5.19, therefore, the motor current drops exponentially

from 1.5 A to 0.5A within 40 ms as in Fig. 5.19 and motor speed also drops from

rated value 7200 rpm to 5000 rpm again in exponential form. Experimental results

for this transient condition are shown in Fig. 5.20. It can be seen that experimental

results are a little difference with the simulating results. Possible reason is that for

the drive system in practical condition, it is difficult to send back the current from

the motor to inverter circuit. Therefore, in adjusting the speed drop from high

speed to low speed, the motor is difficult to work in generator state. In the FEM

simulation, it is assumed that the motor drive operates in ideal condition therefore

speed drops exponentially corresponding with back-emf as shown in Fig. 5.19.

 131

Figure 5.18: Input simulated step voltage waveform and output back-emf waveform

Figure 5.19: Calculated transient speed and stator current waveform

 132

Figure 5.20: Measured transient speed and stator current waveform

5.5.2 Step change variation in mechanical load torque

The transient response for this condition was verified for a step increase of the

mechanical load torque from 0.002Nm to 0.01 Nm when the stator phase voltage

is 4.5 V and the rated speed is 7200 rpm. The simulation results are shown in

Fig. 5.21and Fig. 5.22. Speed versus time graph shows that the motor speed

decreases from rated speed of 7200 rpm to 6000 rpm within 30ms. Stator current

and electromagnetic torque increase two times of the rated conditions respectively.

The results show that the dynamic behaviour of the motor adjusts its operating

conditions to the change of the load by decreasing the rotor speed and increasing

the motor torque to counter-balance the load torque. It is known that the torque

is directly proportional to the stator current if the motor is operated with constant

magnetic field. Hence the stator current increases up to 2.5A in this transient

condition.
 133

Figure 5.21: Speed and back-emf transients due to an increase in load torque

Figure 5.22: Current and torque transient due to an increase in load torque
 134

5.5.3 Locked rotor condition

When a load torque is applied, current is drawn from the supply, resulting a voltage

drop in the motor resistance. This voltage drop is possible only if the back-emf

falls to the value equal to the difference between the supply voltage and the voltage

drop in the motor resistance, and hence the speed falls. If a sufficient load torque

is applied, the speed falls to zero and the motor is then stalled [103]. Then back-

emf is zero and the supply voltage is drop across the motor resistance. Since the

motor resistance is very small, the resulting stall current is extremely large. It is

not normally permissible to allow the full locked rotor current to flow, even for a

short time, because it would either demagnetize the magnets or destroy the power

transistors or burn the winding insulation. Therefore, locked rotor current is one

of the important parameters for motor designers and operators.

In this work, motor stall condition is simulated by using the transient time

stepping model. The motor is loaded from 0.002 Nm to 0.05 Nm suddenly at 2.48

ms where the supply voltage is 4.5 V and the rated speed is 7200rpm. Fig. 5.23

and Fig. 5.24 show the transient response of the motor at stall conditions. It can

be seen that motor speed falls sharply from its rated speed to zero speed within a

very short time interval in about 16.30 ms. The current goes up to 4.1 A, nearly

27 times of the rated current. Attention needs to be drawn to the rapid response

of current to the change in the voltage in this type of motors. This is due to the

very low value of the inductance of the winding.

 135

Figure 5.23: Speed and back-emf transient at locked rotor conditions

Figure 5.24: Current and torque transient at locked rotor conditions

 136

5.6 Conclusion

This chapter attempted to analyze the steady state and dynamic performances of

the BLDC motor using the current fed 2D magneto-static FEM model and time

stepping FEM model. In steady state analysis, a new approach to overcome the

extensive computation time to reach the steady state was implemented. In this ap-

proach, magneto-static FEM model was developed prior to the time domain model

and the solutions (magnetic vector potential A) obtaining from magneto-static

FEM were used for the initial conditions for the time domain analysis. Results

showed that the computation time is greatly reduced to reach to a steady state and

the time domain solver can also be started with the initial conditions quite close

to the steady state solution. Moreover, the proposed model had been successfully

simulated the steady state performance of the motor such as the computation of

the stator current, electromagnetic torque, back-emf, determination of the torque-

speed characteristic, and computation of cogging torque. To validate the proposed

steady state model, experimental results for the steady state conditions were pre-

sented and compared with computation results. It was shown that these two sets

of results agreed closely.

Dynamic conditions of step voltage variation, change of load torque and locked

rotor condition were also investigated. Experiments for these dynamic conditions

were carried out and compared with computational results. Results showed that

the time stepping FEM model can be used to investigate the static and transient

responses of the motor under various time steps.

Chapter 6

Application Characteristics of
BLDC Motors for Hard Disk
Drives

6.1 Introduction

The need for large storage capacity and increased data access performance of mod-

ern hard disk drive technology over the passed decade has meant the continuous

development of small spindle motors. The spindle motor is one of the important el-

ements in a computer hard disk drive (HDD) system and in many ways determines

the drive capacity and performance through its configuration, operating speed,

mechanical and electrical performances like run-out, resonance, starting torque,

run-current etc [118]-[120].

The spindle motor rotates the disk and media while recording heads are sta-

tionary and positioned by an actuator. An accurate assessment of requirement of

starting torque in a spindle motor is vital [121]-[123]. The starting torque can be

20 times as high as the the running torque while the starting current is severely

limited by the computer power supply and other electronics limitations. In addi-

tion, the starting time, otherwise called spin-up time (the amount of time that is

required for the disk platters to get up to full operational speed from a stationary

start) is largely determined by the drive specifications and has to be very fast.

137
 138

Therefore, the starting torque, the starting current and the starting time are the

important quality indexes for the HDD spindle motor. An accurate evaluation of

these quantities is always very challenging.

In addition, some of the requirements of the motor run-up performances are

conflicting and difficult to meet. For example, spin-up time is the important pa-

rameters of the HDDs using power management scheme. Power management is a

set of protocols used to allow HDDs to reduce the amount of power consumption

especially when they lie idle. Spindle motor is allowed to spin down after a couple

of minutes of idle time and then spin up to operational speed whenever one uses

the disk again. Users increasingly want hard disk that will spin up from a station-

ary position to operating speed quickly, which also requires faster spin-up time.

However, to achieve faster spin up time generally required higher the supply volt-

age which can result in higher starting current demand and higher running current

at normal running condition. The consequences are increased in power consump-

tion in both starting and running. Power consumption is an area of concern for

PCs especially for the system of multiple storage drives and certainly for laptop

users. Modern PCs use very limited power packs such as the PC/XT, AT, Baby

AT and LPX form factors power supplies for PCs can provide maximum 12A for

12V output voltage level whereas the ATX/NLX, SFX AND WTX form factors

can provide only 8A for 12V voltage level. Generally, 25% is taken by the actuator

assembly and the rest 75% of the 12V power can be taken by the spindle motor.

Therefore, the starting current and the supply voltage are limited in practical case.

It is needed to calculate accurately the motor run-up performances in the design

stages.

An attempt is made in this work to study the starting process of a HDD

spindle motor and analyze the run-up characteristics of the spindle motor with

different drive constraints such as limited starting current and supply voltage, lower
 139

power consumption for both starting and running states by computational means.

Time stepping FEM coupled with closed loop feed back controller is used.

6.2 Coupling with the Control Loop

Spindle motors are outer rotor type brushless permanent magnet DC motors and

require electronic controllers. The phase windings of the motor are energized in

sequence by switching elements in the inverter. In order to simulate precisely the

performances of the motor, characteristic of the inverter circuit and its control loop

feature is needed to include as a coupled system. Moreover, in the hard disk drive

system, fast and accurate control of disk spinning speed is vital for error free read

and write operations. This requires a sophisticated controller for the purpose of

rapid response. In this model, a simple and cascaded speed and current control

structure, based on the current hysteresis PWM switching strategy, is proposed to

couple the time stepping FEM model. Detailed control system block diagram is

shown in Fig. 6.1.

In the control loop simulation, the required feedback signals, rotor speed and

stator current are determined from the time stepping FEM model. In the outer

speed loop simulation, the reference speed is 7200 rpm where actual motor speed is

calculated from the time stepping FEM model. In the inner current loop, reference

current for the current hysteresis loop is determined from the PI controller of the

speed loop. The feedback current signal is determined from the time stepping FEM

model. The output of the current hysteresis loop is the firing signal for the inverter

circuit and again these are the input voltage for the time stepping FEM model.

Detailed modeling technique for the current hysteresis control is as follows. The
 140

reference current is determined by the PI controller of the speed loop;

Kp (ω ∗ − ω) + Ik∗ − Im 6 Is∗ 6 Im






∗
Is = Im Is∗ > Im (6.1)



Is∗ > −Im

−Im


where  ∗
 Ik(0) = 0

Ik∗ = ω∗ − ω (6.2)
 Ik∗ = Ik−1
 ∗
+ ∆T
Ts
Therefore, the reference current is

i∗s = Is∗ i∗ (6.3)

In equation (6.1), (6.2) and (6.3), ω ∗ is the reference speed, ω is the speed feedback

from the time stepping FEM model, Kp is the proportionality factor, Ts is the

integral factor; ∆T is the time step size; Im is the maximum allowable value of

stator current and i∗ is the expected current waveform with unity value of 0 and 1.

Finally, the stator voltage is controlled by the following current loop when i∗s 6= 0;

is < i∗s − ∆i


 +Vs



Vs(k) = −Vs is < i∗s + ∆i (6.4)



i∗s − ∆i 6 is 6 i∗s + ∆i

Vs(k−1)


By coupling the closed loop speed and current control with the time step-

ping FEM, the stator windings could be fed with the actual output voltage from

the inverter to FEM model. As mentioned before, in the time stepping FEM, the

magnetic field equations, the stator circuit equations and the motion equation are

all solved simultaneously according to the time step. The effects of high order har-

monic, the non-sinusoidal quantities, the eddy current, the saturation and the rotor

rotation can be directly included into the whole modeling system. This model is

not so difficult for implementation, and accurate to a great extent. Starting char-

acteristics of the BLDC motor using the developed cascaded control loop structure

with time stepping FEM model are analyzed in the next sections.
 141

Figure 6.1: Control system block diagram

6.3 Analysis of the Starting Process of a HDD

Spindle Motor

Starting characteristics of a HDD spindle motor are analyzed under the following

different conditions.

1. Starting without drive limits

2. Starting with current limits and

3. Starting with speed limits

6.3.1 Motor starting without drive limits

The starting processes of the motor without drive limits are analyzed using the

time stepping FEM model. This is an open loop simulation. Closed loop control

structure is not included in this process. Both no-load and loaded conditions are

considered.

No-load condition: Firstly, it is assumed that the motor runs freely without

any platters and spacer rings. Fig. 6.2 shows the input voltage waveform for the

FEM block. The calculated starting characteristics of speed, back-emf, current and
 142

torque with motor starting time are shown in Fig. 6.3 - Fig. 6.6.

Figure 6.2: Input voltage waveform against time

Figure 6.3: Motor starting speed without limits

Figure 6.4: Back-emf waveform without limits

 143

Figure 6.5: Starting current without limits

Figure 6.6: Starting torque without limits

 144

From speed vs. time graph (Fig. 6.3), the motor takes 40 ms to reach the

rated speed of 7200 rpm and the motor speed keeps on increasing. There is no

voltage limit in this simulation process so that the motor is operated over-speed

region. The back-emf also keeps on increasing since it is directly proportional to

the motor speed. In the over-speed region, the back-emf is a little higher than the

motor supply voltage as shown in Fig. 6.4. In a practical case, the supply voltage

has to be regulated using voltage control mechanism in this over-speed condition.

Fig.6.5 shows the starting stator current profile of the motor without any

limit. The stator current goes up to 3.5A at starting which is more than nearly 10

times larger than the rated value, but the current goes down to its normal running

current, 0.36 A, within 40ms. The stator current keeps on decreasing nearly to

zero within the next 40 ms. The results show that when the motor speed increases,

the motor current also decreases proportionately. In addition, motor torque is also

very high at starting as shown in Fig 6.6. However, when the motor speed goes

up, torque decreases proportionately.

When motor reaches to a steady state condition, the developed torque is very

small as shown in Fig.6.6. A very small torque is required in normal running con-

dition in this type of drives. At starting, the motor speed goes up, the back-emf

also increases and the voltage overhead (the difference between the supply voltage

and the back-emf) decreases proportionally so that the motor current decreases.

When the motor speed reaches to the rated value of 7200 rpm in 40 ms, the motor

current also hits its rated value, 0.36A, and the motor torque is also at its rated

value, 0.01Nm. The controller needs to maintain this rated condition in order to

keep machine stability and reliability. Voltage control is needed to regulate the

motor speed and current control is required to stabilize the motor current. How-

ever, in this simulation, the motor is running without voltage and current limiters

so that the motor spins up to 7600 rpm whereas the stator current and developed
 145

torque becomes nearly zero.

Loaded with one platter: The next step is to study the motor starting charac-

teristic under loaded conditions. The motor is loaded with a 3.5 inch platter and

a spacer ring as shown in Fig. 6.7. Loaded condition means the inertia of platter

and spacer ring are added to the mechanical equation in the FEM simulation. The

supply voltage is 4V. Fig. 6.8, Fig. 6.9, Fig. 6.10 and Fig. 6.11 show the simulation

results.

Figure 6.7: Motor loaded with one platter

Figure 6.8: Speed against time waveform when motor is loaded with one platter
 146

Figure 6.9: Back-emf waveform when motor is loaded with one platter

Figure 6.10: Stator current waveform when motor is loaded with one platter

Figure 6.11: Torque against time graph where motor is loaded with one platter
 147

Loaded with two platters: For a second experiment, the spindle motor is

loaded with two platters and a spacer ring as shown in Fig. 6.12. The simulation

results are shown in Figs. 6.13, 6.14, 6.15 and 6.16. The speed against spin

up time for loaded and no-load conditions from the simulation results are shown

in Fig. 6.17. Results from measurement are presented in Fig. 6.18. It is found

that as the number of spindle platters increases, the motor speeds are decreased

accordingly. This generally means that the spin-up speed and power consumption

of a drive with four platters is much higher than those figures for the same drive

with one or two platters. In practice, the number and the size of the platters vary

in different hard disk drive designs and have an important impact on performance

in several ways. The trend is towards smaller platter sizes.

Based on these results, we can conclude that precise computation of the start-

ing process of the BLDC motor is one of the important issues for studying the motor

performance and control implementation. Moreover, fast and accurate control for

spinning speed is vital for error free operation in BLDC motor. Evaluation of mo-

tor starting performance using the developed model including speed and current

controls are presented in the next section.

Figure 6.12: Motor loaded with two platters

 148

Figure 6.13: Speed against time waveform when motor is loaded with two platters

Figure 6.14: Back-emf waveform when motor is loaded with two platters

Figure 6.15: Stator current waveform when motor is loaded with two platters
 149

Figure 6.16: Torque against time graph where motor is loaded with two platters

Figure 6.17: Calculated motor speed under no load and loaded conditions

Figure 6.18: Measured motor speed under no load and loaded conditions
 150

6.3.2 Motor starting with current limits

The power supply for hard disk drive for PC computer is +12V for the spindle

motor and actuator; and +5V for drive circuits respectively. Roughly, 75% of

the +12V power is taken by the spindle motor and the rest 25% by the actuator

assembly.

As shown in Fig.6.5, if there is no limit, motor starting current can be as

high as 3.5A. The higher the motor starting current is concerned with drive power

consumption, heat and temperature; and must be given special consideration in

systems with multiple storage drives. The ratings of the power supply, particu-

larly at the +12V level, sometimes may not be sufficient for starting conditions.

Therefore, the motor starting current is limited by the controller in practice.

In this work, the cascaded closed loop control structure which was given in

Section 6.1 is used to control the motor current. In the current control loop, a

current limiter is used to set the current limit and a current hysteresis controller is

used to control the current in keeping within the required limit. Fig. 6.19 shows

the motor voltage from the current hysteresis controller. The motor starting cur-

rent with 1.5 A current limit is shown in Fig. 6.20. It can be seen that the motor

starting current is within the limits of 1.5A. However, the motor speed is slower

to reach to the rated speed compared with the first case- without limits (Fig. 6.3).

It takes 57 ms which is 17ms longer, to spin up to the rated speed of 7200 rpm

(Fig. 6.22). The motor torque is also lower compared to that in the first case as

shown in Fig. 6.23 because torque is directly proportional to the current.

 151

Figure 6.19: Voltage comes from the hysteresis controller

Figure 6.20: Motor starting current with 1.5A current limit

Figure 6.21: Motor back-emf waveform when current is limited at 1.5A

 152

Figure 6.22: Speed profile with 1.5A current limit

Figure 6.23: Starting torque profile with 1.5A current limit

 153

6.3.3 Motor starting with speed limit

When the motor is running with no speed limits as discussed in the previous section,

where motor supply voltage was fixed so that the motor speed keeps on increasing

to over speed region with a higher back-emf value. In order to avoid this condition,

a speed limit is needed. In this work, cascaded closed loop control which has been

proposed in a previous section 6.1, can be used to implement the speed limits of

the motor. In the outer speed loop, we set the motor reference speed to the rated

speed 7200 rpm. If the speed is more than its rated value, the controller adjusts

the speed to its rated value again.

The simulated results are shown in Fig. 6.24- Fig. 6.27. Motor speed profiles

with and without speed limits are also compared and results are shown in Fig. 6.28.

The point where motor speed reaches to its rated value, speed controller maintains

the speed in its rated speed 7200rpm. According to Fig. 6.24, the motor speed

reaches to its rated value at 40 ms and from that point speed is limited to 7200

rpm. When the speed is constant, motor back-emf is constant as shown in Fig. 6.26.

Motor current is also in a steady state in this instant as shown in Fig. 6.27, since

the motor input voltage is constant.

Figure 6.24: Starting speed profile with speed limit

 154

Figure 6.25: Motor supply voltage

Figure 6.26: Motor back-emf profile with speed limit

Figure 6.27: Stator current profile with speed limit

 155

Figure 6.28: Speed with speed limit and without limits

6.4 Computational Analysis of the Run-up

Performances of a HDD Spindle Motor

In this section, the run-up characteristics of the spindle motor with various supply

voltages and limited starting currents, which are the practical constraints in motor

design stages, are analyzed. Possible Conditions are justified to optimize the spin-

up time and power consumptions in both starting and running conditions. Time

stepping FEM coupled with closed loop feedback controller is used.

6.4.1 Case I: Motor runs freely under various

stator phase supply voltages

In this condition, motor is allowed to run freely up to its rated condition. Feedback

control adjusts the speed to rated speed 7200 rpm. Simulated results are shown

in Fig. 6.29 and listed in Table 6.1. The stator phase supply voltages are 6V,

5V, 4.5V and 4V whereas the spin-up times to reach the rated speed 7200rpm

are 12.7ms, 16.8ms, 21.0ms and 32.1ms respectively as shown in Fig. 6.29. The
 156

highest the stator phase supply voltage 6V resulted the fastest the spin-up time

of 12.7ms to reach the rated speed and the lowest the stator phase supply voltage

4V resulted the slowest spin-up time of 32.1ms. However, if the supply voltage is

high, the spin-up time is fast, at the same time the starting current is also high

as shown in Fig. 6.30 and in many cases the current may go beyond the cut-off

point. As a result, the motor power consumption is high in both starting and

running conditions. Simulated data are summarized in Table 6.1.

Figure 6.29: Motor speed vs. spin-up time with different supply stator phase

voltages

6.4.2 Case II: Motor running with current limiter

Starting current is limited at 1.5A using the current limiter in the feedback closed

loop control. Fig. 6.31 shows the spin-up times against the speed with different

supply voltages. Motor power consumptions with current limit under different

supply voltages are listed in Table 6.2. The power consumption for the 6V is

also the highest. Comparison of spin-up time with and without current limits are

shown in Fig. 6.32 and listed in Table 6.3. It is found that the spin-up times for each

voltage level with current limits for Case II are slower than compared with Case
 157

Figure 6.30: Starting current profiles with different supply stator phase voltages

Table 6.1: Power consumptions with different supply phase voltages at no load

condition

Supply Spin− starting steady P ower P ower

voltage time current state consumption consumption at

(V ) (ms) (A) current(A) at starting(W ) steady state(W )

6 12.7 5.19 2.06 31.14 12.36

5 16.8 4.32 1.19 21.60 5.95

4.5 21.0 3.89 0.75 17.50 3.38

4 32.1 3.45 0.32 13.80 1.28

I without current limits. Power consumptions for the operation models with and

without current limit are also compared in two different conditions: (1) starting

(2) running (steady state). Comparisons data are listed in Table 6.4 and shown in

Fig. 6.33. It is found that the power consumption for Case I is lower than that in
 158

case II for all supply voltage levels both at starting and running states. Based on

these results, it can be concluded that, one can use the higher supply voltage to

get the faster spin-up time if the current is in the limited range so that the power

consumption is estimated.

Figure 6.31: Speed vs. spin-up time with different supply voltages where current

is limited at 1.5A

Table 6.2: Motor power consumptions with current limits

Supply Spin− starting steady P ower P ower

voltage time current state consumption consumption at

(V ) (ms) (A) current(A) at starting(W ) steady state(W )

6 34.9 1.5 1.5 9.0 9.0

5 35.7 1.5 1.19 7.50 5.95

4.5 38.5 1.5 0.75 6.75 3.38

4 48.7 1.5 0.32 6.0 1.28

 159

Figure 6.32: Supply voltage vs. spin-up time with and without current limit

Figure 6.33: Comparison of power consumption with and without current limit at

starting
 160

Table 6.3: Spin-time with and without current limits

Supply Spin − time spin − time

voltage(V ) without current limit(ms) with current limit(ms)

6 12.7 34.9

5 16.8 35.7

4.5 21.0 38.5

4 32.1 48.7

Table 6.4: Comparison of power consumption with and without current limits

Supply P ower P ower P ower P ower

voltage consumption at consumption at consumption at consumption at

(V ) starting with starting without steady state steady state

. limits(W ) limits(W ) with limits(W ) without limits(W )

6 9.0 31.14 9 12.36

5 7.5 21.6 5.95 5.95

4.5 6.75 17.5 3.38 3.38

4 6.0 13.8 1.28 1.28

6.4.3 Case III: Motor running with voltage adjusting scheme

Although the spindle motor requires high power at start up, the power consumption

at normal running is very low. Therefore, we can reduce the supply voltage to

optimize the motor power when the motor reaches its rated speed. The closed loop

model which has been developed in Section 6.1 is used to implement the voltage

control. In the simulation program, the flag is set to check whether the speed

reaches to its rated value. When the speed reaches to the rated value, the motor

current starts to reduce to its rated value by adjusting the voltage. Hysteresis

controller is used to adjust the supply voltage. For example, if the rated current
 161

for this prototype motor is 0.2A, set the reference current of the current loop in

the proposed controller to 0.2A. Outer speed loop also adjust the motor speed into

its rated value 7200rpm. Therefore motor supply voltage has been adjusted by

controller to maintain these drive requirements.

Calculated results are shown in Fig. 6.35. Results without voltage limits are

also shown to compare with the results using adjusting supply voltage scheme. The

current profiles are shown in Fig. 6.34 and Fig. 6.35. It can be seen clearly that

the motor current is reduced to its rated value 0.2A starting from the motor speed

reaches to its rated value 7200 rpm by using voltage adjusting scheme. By using

the proposed voltage adjusting scheme, the motor power consumption at motor

running condition can be saved.

 162

Figure 6.34: Motor transient responses without voltage adjusting scheme

 163

Figure 6.35: Motor transient responses with voltage adjusting scheme

 164

6.5 Conclusion

In this chapter, efforts were made in order to investigate the important quality

indexes of the spindle motor such as an accurate assessment of the requirement

of starting torque, spin-time and starting current. Starting performances of the

spindle motor with drive current limits, speed limits and without limits were de-

termined.

In order to analyze the motor with a controller as an actual system, a new

approach to couple the time stepping FEM with closed loop control structure was

implemented. Cascaded speed and current hysteresis control loop structures was

used. By coupling the control loop features with the time stepping FEM, the

stator windings could be fed with the actual input voltages to the time stepping

FEM model. In addition, motor performances under transient conditions can be

controlled instantaneously as an actual motor-controller system. Using this new

scheme, simple and efficient techniques to analyze the spindle motor performance

using time stepping FEM couple with cascaded control loop were developed. Com-

putational analysis of run-up performances of HDD spindle motor under starting

and normal running conditions were investigated.

Chapter 7

Discussions and Conclusions

This thesis attempted to investigate the computation and analysis of steady state

and dynamic performances of BLDC motor. Modelling, simulation and perfor-

mance analysis of the BLDC motor using numerical methods were presented.

An accurate analysis of a real machine would call upon three dimensional

analysis. However, comparing with 2D FEM, the 3D FEM requires much com-

puting power, storage and computer time usage especially in transient analysis

where rotor movement has to be included in field simulation. In order to study

dynamic performances of the motor, external circuit and rotor rotation should be

coupled with the electromagnetic field in time domain and time stepping technique

is a practical solution for this problem. If one uses 3D time stepping FEM, rotor

meshes are required to move according to the rotor rotation at each time steps. It

may take very long computing time and require very large amount of memory. It is

often neither practical nor always possible in real world applications. In addition,

3D effects such as end winding effect can, however, be taken into account using

empirical formulae and the validity of the formulae have been well proven with ex-

periments and analytical techniques. In this work, numerical models for studying

the steady state and dynamic performances are formulated in two dimensional and

3D effects such as end winding effects are approximated and included. The validity

of 2D FEM model is verified by the experiments. Good agreements in experimental

and simulation results show that 2D FEM is validated to study the steady state as

165
 166

well as the transient performance of the BLDC motor.

In addition, spindle motor is used as a working example in this work. The

axial length of the spindle motor is quite short and it is seemed that 3D FEM is

necessary. However, the air gap length of the spindle motor is very small (0.05mm

∼ 0.15mm). Hence, 2D FEM is good enough to use in performance analysis of this

particular case.

Dynamic model of the BLDC motor was developed based on the 2D time

stepping FEM. This model was the numerically strong coupling model whereas the

system equations are solved simultaneously at each time steps in time varying elec-

tromagnetic solver. Due to the direct coupling of transient electromagnetic field,

the electric circuits and the motion in this dynamic model, the solution could be

taken into account the eddy current effect, the saturation effect, the rotor move-

ment, high order harmonic of the electromagnetic fields and the non-sinusoidal

quantities which are very difficult to include in conventional finite element method

coupled with equivalent circuit models. 3D effects i.e end winding effects are taken

into account using empirical formulae. Dynamic model was used in studying the

transient analysis of the BLDC motor at step voltage variations, load torque chang-

ing and locked rotor condition. Results shows that the proposed model can investi-

gate the transient responses of the motor under various motor operating conditions.

It can give a clear representation of the motor performances such as stator current,

torque, motor back-emf, speed with its operating times respectively. In addition,

the model can be used to simulate dynamic characteristics of the motor under

various time steps. These are the important features for the machine designers to

do simple simulation of the dynamic performance of the motor with desired motor

parameters. Proposed dynamic model was also used to investigate the starting

processes of the BLDC motor. It has been observed that the important motor

starting parameters such as starting currents; motor spin up time with various
 167

supply voltages; required starting torques for given motor parameters etc., can be

calculated precisely under time frames.

The analysis of the steady state performance of nonlinear electromagnetic sys-

tems using time stepping FEM requires very long computational times. Normally,

the transient solver begins with a flat start by initializing all state variables to zero.

This results in a large DC components of current and several hundred integral time

steps are needed before reaching its steady state. An improved steady state model

of the BLDC motor was implemented using time stepping FEM coupled with two

dimensional magneto-static FEM. In this steady state model, magneto-static FEM

was used as a pre-computation stage for time stepping FEM. It is found that using

this improved method, the transient solver can be started with initial conditions

quite close to the steady state solutions.The first few cycles are almost free of di-

rect current. It can reduce the time spent in reaching a steady state solution is cut

by roughly a factor of three. The model can take into account the non-sinusoidal

quantities, stator harmonics and rotor movement which are very difficult to sim-

ulate instantaneously in the steady state analysis using traditional finite element

method. Using the proposed model, calculation of the steady state current, cogging

torque, back-emf and determination of torque-speed characteristics of the BLDC

motor were analyzed. Results show that the torque was directly proportional to

the winding currents and speed of the motor can be controlled by measuring the

stator phase voltage applied to the motor. These results were used to implement

the application of simple speed and current control for BLDC motor drive system.

In order to determine the accuracy and validation of the proposed dynamic

and steady state model, DSP based BLDC motor test stand was implemented.

Simple and reliable methods of motor performance measurements were presented.

A new approach for detecting the motor starting sequences for controller was de-

veloped. Based on the experience of hardware set up in this experiment, it is

 168

recommended that very low rotor inertia incremental encoder is necessary in order

to couple with this type of small capacity HDD spindle motor. Since the motor

is designed to run two 3.5 inch disc platters, it is needed to calculate prior the

inertia of the whole coupling hardware components such as shaft inertia, coupler

inertia, encoder inertia and platters inertia in order to avoid the motor to run the

overloaded conditions.

BLDC motors cannot work without the electronic controllers. In order to an-

alyze the motor with controller as an actual system, a new approach to couple the

time stepping FEM with closed loop control structure was implemented. Cascaded

speed and current hysteresis control loop structures were used. By coupling the

control loop features with the time stepping FEM, the stator windings could be

fed with the actual input voltages to the time stepping FEM model. In addition,

motor performances under transient conditions can be controlled instantaneously

as an actual motor-controller system. Using this new scheme, application char-

acteristics of the HDD spindle motors were investigated. Important features of

the spindle motor at starting conditions under no load and loaded conditions were

analyzed. It is found that the proposed model worked satisfactorily when used in

simulation for motor under real transient cases with voltage, current and speed lim-

its. This analysis provides assistance to implement the power management scheme

especially for the notebook PCs and investigations of spin up times under drive

limits such as current limits, voltage limits and speed limits. Results are verified

with experiments. It is found that the simulation results of the motor have lit-

tle discrepancies with experimental results especially in transient conditions. This

is because although the control loop structure is included in the FEM model, the

transient characteristics of the controller cannot be taken into account in this work.

It is another big issue to analyze the dynamic behaviour of control system. Further

work is needed to develop a controller dynamic model such as by using Matlab

 169

simulink in order to couple the fully dynamic behaviour of motor-controller drive

system.

Computational analysis of run-up performances of the HDD spindle motor

performances under starting and normal running conditions were investigated.

These studies are demonstrated that motor performances can be optimized at nor-

mal running conditions by adjusting the supply voltages in order to maintain the

motor with its rated conditions. Current hysteresis control is suitable for fast and

accurate response.
Bibliography

[1] Essam S. Hamdi, “Design of small electrical machines,” John Wiley and Sons,

1996.

[2] Jacek F. Gieras and Michell Wing, “Permanent magnet motor technology-

Design and application,” Marcel Dekker, Inc. 1997.

[3] J. R. Hendershot Jr. and T. J. E Miller, “Design of brushless permanent-

magnet motors,” Magan Physis Publishing and Calerndon Press, Oxford

1994.

[4] Cho, Chahee Peter, “ Analysis and design considerations of a high-power

density, dual air gap, axial-field, brushless, permanent magnet motor,” Ph.D

thesis, University of New Hampshire, May 1995.

[5] N. Ertugrul, P. P. Acarnley, “Indirect rotor position sensing in real-time for

BLDC motor drives,” IEEE Trans. on Power Electronics, vol.13, No.4, pp.

608-616, July 1998.

[6] Remigius C. Okonkwo, “ Design and investigation of linear brushless dc mo-

tors with high energy permanent magnets,” Verlag Dr. Kster, 1999.

[7] Takashi Kenjo, “Permanent magnet and brushless DC motors,” Sogo Elec-

tronic Publishing Company, Tokyo, 1984.

[8] Pei Bai Zhou, “Numerical analysis of electromagnetic fields,” Springer-Verlag,

1993.

170
 [9] K. Hameyer and R. Belmans, “Numerical Modelling and design of electrical

machines and devices,” WIT Press, 1999.

[10] S. J. Salon, “Finite element analysis of electrical machines,” Kluwer, 1995.

[11] N. A. Demerdash and D. H. Gillot, “A new approach for determination

of eddy currents and flux penetration in on-linear ferromangetic material,”

IEEE Trans. on Magn., vol. 74, pp. 682-685, march 1974.

[12] A. G. Jack and B. C. Mecrow, “Methods for magnetically non-linear problems

involving significant hyeteresis and eddy currents,” IEEE Trans. on Magn.,

vol. 26, no. 2, pp. 424-429, march 1990.

[13] David Meeker, “Finite element method magnetics - User Manual(Version

3.4),” http://www/femm.berlios.de, December 2004.

[14] G. R. Slemon, A. V. Gumaste, “Steady-state analysis of a permanent mag-

net synchronous motor drive with current source inverter,” IEEE Trans. on

Industry Applications, vol. IA-19, no. 2, pp. 190-197, 1983.

[15] Thomas M. Jahns, Gerald B. Kliman and Thomas W. Neuman, “Interior

permanent magnet synchronous motors for adjustable speed drives,” IEEE

Trans. on Industry Applications, vol. IA-22, no. 4, July/August 1986.

[16] Pillay, Pragasen, “Modeling Simulation and Analysis of Permanent Magnet

Synchronous and Brushless DC Motor Drives,” Ph.D thesis, Virginia Poly-

technic Institute and State University, 1987.

[17] C. C. Hwang, S. B. Jhon, S. S. Wu, “Reduction of cogging torque in spindle

motors for CD-ROM drive,” IEEE Trans on Magn., vol.34, No.2, pp.468-470,

March 1998.

171
[18] Francis Piriou and Adel Razek, “Finite element analysis in electromagnetic

systems accounting for electric circuits,” IEEE Trans. on magnetic, vol.29,

no.2, March 1993.

[19] Zhou P., Rahman M. A and Jabbar M. A, “Field circuit analysis of permanent

magnet motors,” IEEE Trans. on magnetics, vol.30, pp.1350-1359, July 1994.

[20] M. A. Rahman, Ping Zhou, “Analysis of brushless permanent magnet syn-

chronous motors,” IEEE Trans. on Magnetics, vol. 43, no. 2, pp. 250-267,

April 1996.

[21] Andrew M. Knight, John C. Salmon, “Modeling the dynamic behavior of

single phase line-start permanent magnet motors,” IEEE Industry Society,

54th meeting, session 59, pp. 2582-2588, 1999.

[22] Shailesh Waikar, Tilak Gopalarathnam and Hamid A. Toliyat, “ Evaluation of

multiphase brushless permanent magnet (BPM) motors using finite element

method (FEM) and experiments,” IEEE Trans. on Industry Appl. Vol.2, no.1,

pp.396-401, May 1999.

[23] C. Bi, Z. J. Liu and S. X. Chen, “Estimation of back emf of PM BLDC

motors using derivative of FE solutions,” IEEE Trans. on Magnetics, vol. 36,

no.4, July 2000.

[24] M. V. K. Chari and S. J. Salon, “Numerical methods in electromagnetism,”

Academic Press, A Harcourt Science and Technology Company, 2000.

[25] K. Hameyer, J. Driesen, H. D. Gersem and R. Belmans, “The classification

of coupled field problems,” IEEE Trans on magnetics, vol.35, no.3, pp.1618-

1620, May 1999.

172
[26] S. J. Salon, R. Palma and C. C. Hwang, “Dynamic modeling of an induction

motor connected to an adjustable speed drive,” IEEE Trans on magnetics,

vol.25, no.4, pp. 3061-3063, July 1989.

[27] Z. Q. Zhu, G. W. Jewell and D. Howe, “Finite element analysis in the design

of permanent magnet machiens,” IEE Seminar on Current Trends in the Use

of Finite Elements in electrmechanical Design and Analysis, pp. 1-7, January

2000.

[28] S. L. Ho and W. N. Fu, “A comprehensive approach to the solution of direct

couple multislice model of skewed rotor induction motors using time stepping

eddy current finite element method,” IEEE Trans on Magnetics, vol.33, no.3,

pp. 2265-2273, May 1997.

[29] S. Wang, J. Kang, K. Park, H. Yoon and G. Jang, “2D and 3D FEA of a

BLDC motor,” The international journal of Computation and Mathematics

in Electrical and Electronics Eng., vol.19, no.2, pp. 529-537, 2000.

[30] T. W. Nehl, F. A. Fouad, N. A. Demerdash and E. A. Maslowski, “ Dy-

namic simulation of radially oriented permanent magnet electronically op-

erated synchronous machines with parameters obtained from finite element

field solutions,” IEEE Trans. Vol.IA-18, no.2, pp.172-181, March/April 1982.

[31] T. W. Nehl, F. A. Fouad and N. A. Demerdash, “Digital simulation of

power conditioner-machine interaction for electronically commutated DC per-

manent magnet machines,” IEEE Trans on magnetics, vol.MAG-17, no.6,

pp.3284-3286, November 1981.

[32] N. A. Demerdash and T. W. Nehl, “Dynamic modeling of brushless DC mo-

tors in electric propulsion and electromechanical actuation by digital tech-

niques,” IAS annual meeting, pp.570-578, 1980.

173
[33] N. A. Demerdash and T. W. Nehl, “Dynamic modeling of brushless DC

motors for aerospace actuation,” IEEE Trans., vol.AES-16, no.6, pp. 811-

821, November 1980.

[34] R. H. Miller, T. W. Nehl, N. A. Demerdash, B. P. Overton and C. J. Ford,

“An electronically controlled permanent magnet synchronous machine condi-

tioner system for electric passenger vehicle propulsion,” IAS Annual meeting,

pp.506-510, 1982.

[35] P. F. Muir and C. P. Neumann, “Pulsewidth modulation control of brushless

DC motors for robotic applications,” IEEE Trans., vol.IE-32, no.3, pp.222-

229, August 1985.

[36] D. M. Erdman and H. B. Harms, “Estimating speed-torque profiles of the

brushless DC motor,” Proceedings of the Motorcon Conference, pp.1-8, 1985.

[37] A. K. Wallace and R. F. Spee, “The effect of motor parameters on the per-

formance of brushless DC drives,” IEEE PESC, pp. 591-597, 1987.

[38] P. Pillay and R. Krishnan, “Modelling, simulation and analysis of a per-

manent magnet brushless DC motor drive,” IAS Annual meeting, pp.7-14,

1987.

[39] R. Spee and A. Wallace, “Performance characteristics of brushless DC

drives,” IAS Annual meeting, pp.1-6, 1987.

[40] P. Pillay and R. Krishnan, “Modeling, Simulation, and analysis of permanent

magnet motor drives, Part II: The Brushless DC Motor Drive,” IEEE Trans

on Industry applications, vol. 25, no.2, pp.274-279,March/April 1989.

[41] Y. P. Liu, D. Howe, T. S. Birch and D. M. H. Matthews, “Dynamic Modeling

and Performance Prediction of Brushless Permanent Magnet Drive System,”

174
 Electrical Machines and Drives, Fourth International Conference, pp. 95-99,

1989.

[42] M. M. Elmissiry and S. Chari, “Dynamic Performance of a Permanent Mag-

net, Axial Flux, Toroidal Stator, Brushless D.C.Motor,” 3rd Africon Confer-

ence, pp. 400-403, 1992.

[43] Liuchen Chang, “An improved FE inductance calculation for electrical ma-

chines,” IEEE Trans on magnetics, vol.32, no.4, pp.3232-3245, July 1996.

[44] D. Howe and Z. Q. Zhu, “The influence of finite element discretisation on the

prediction of cogging torque in permanent magnet excited motors,” IEEE

Trans on magnetic, vol.28, no.2, pp.1080-1083, March 1992.

[45] Z. J. Liu, C. Bi, H. C. Tan and T. S. Low, “Modeling and torque analysis

of permanent magnet spindle motor for disk drive system,” IEEE Trans on

magnetics, vol.30, no.6, pp.4317-4319, November 1994.

[46] Z. Q. Zhu, P. J. Hor, D. Howe and J. Rees-Jones, “Calculation of cogging

force in a slotted linear tubular brushless permanent magnet motor,” IEEE

Trans on magnetic, vol.33, no. 7, pp.4098-4100, September 1997.

[47] S. X. Chen, T. S. Low, B. Brhsul, “The robust design approach for reducing

cogging torque in permanent magnet motor,” IEEE Trans on magnetics,

vol.34, no.4, pp. 2135-2137, July 1998.

[48] C. Bretn, J. Bartolom, J. A. Benito, G. Tassinario, I. Flotats, C. W. Lu,

and B. J. Chalmers, “Influence of machine symmetry on reduction of cogging

torque in permanent magnet brushless motors,” IEEE Trans on magnetics,

vol.36, No. 5, pp.3819-3823, September 2000.

175
[49] M. A. Jabbar, Win Lai Aye, Z. J. Liu and M. A. Rahman, “Computation of

forces and torques in electric machines,” Canadian Conference on Electrical

and Computer Engineering, vol.1, pp.370-374, May 2000.

[50] C. Bi, Z. J. Liu and S. X. Chen, “Estimation of back-emf of PM BLDC

Motors using derivative of FE solutions,” IEEE Trans on magnetics, vol.36,

no.4, pp.697-699, July 2000.

[51] K. Atallah, Z. Q. Zhu and D. Howe, “An improved method for prediction iron

losses in brushless permanent magnet DC drives,” IEEE Trans on magnetics,

vol.28, no.5, pp. 2997-2999, September 1992.

[52] Z. J. Liu, C. Bi, and T. S. Low, “Analysis of iron loss in hard disk drive

spindle motor,” IEEE Trans on magnetics, vol.33, no.5, September 1997.

[53] Z. Q. Zhu, K. Ng and D. Howe, “Design and analysis of high speed brushless

permanent magnet motors,” EMD 97, conference publication no. 444, pp.

381-385, September 1997.

[54] M. S. N. AI-Din, A. F. Kader and J. AI-Samarai, “A new method to compute

eddy current losses by the finite elements method,” IEEE Industry Applica-

tions Society Annual Meeting, New Orleans, Louisiana, pp.3-9, October5-9,

1997.

[55] G. W. Jewell and D. Howe, “Computer-aided design of magnetizing fixtures

for the post-assembly magnetization of rare-earth permanent magnet brush-

less DC motors,” IEEE Trans on magnetics, vol.28, no.5, pp. 3036-3038,

September 1992.

[56] C. C. Chan, K. T. Chau, J. Z. Jiang, “Novel permanent magnet drive for

electric vehicles,” IEEE Trans on Industrial Electronics, vol.43, no.2, April

1996, pp.331-339.

176
[57] Gyu-Hong Kang, Jung-Pyo Hong and Gyu-Tak Kim, “Design and analysis of

slotless-type permanent magnet linear brushless motor by using equivalent

magnetizing current,” IEEE Trans on Industry Applications, vol.37, no.5,

September/October 2001, pp. 1241-1247.

[58] S. Williamson, “Induction motor modeling using finite element,” Int. Conf.

Electrical Machines, Paris, France, pp. 1-8, September 5-8 1994.

[59] A. Arkkio, “Analysis of induction motors based on the numerical solution of

the magnetic field and circuit equations,” Helsinki Acta Polytechnica Scan-

dinavica, Electrical Engineering Series, no. 59, 1987.

[60] T. H. Pham, P. F. Wendling, S. J. Salon and H. Acikgoz, “Transient finite

element analysis of an induction motor with external circuit connections and

electromechanical coupling,” IEEE Trans. Energy conversion, vol. 14, no.4,

pp.1407-1414, December 1999.

[61] M. J. DeBortoli, S. J. Salon, D. W. Burow and C. J. Slavik, “Effects of rotor

eccentricity and parallel windings on induction machine behaviour: a study

using finite element analysis,” IEEE Trans. Magn. Vol. 29, No. 2, pp.1676-

1682, March 1993.

[62] S. L. Ho, W. N. Fu, and H. C. Wong, “Estimation of stray losses of skewed

induction motors using coupled 2-D and 3-D time stepping finite element

models,” IEEE Trans. on Magnetics, vol. 34, pp. 3102-3105, Sept. 1998.

[63] S. L. Ho, H. Li, and W. N. Fu, “Inclusion of interbar currents in a network-

field coupled time-stepping finite element model of skewed-rotor induction

motors,” IEEE Trans. on Magnetics, vol. 35, no. 5, pp. 4218-4225, Sept.

1999.

177
[64] R. De Weerdt, E. Tuinman and K. Hameyer, “Finite element analysis of

steady state behaviour of squirrel cage induction motors compared with mea-

surements,” IEEE Trans. on Magnetics, vol.33, no.2, pp2093-2096, March

1997.

[65] S. L. Ho, H. L. Li, “Dynamic modeling of permanent magnet synchronous

machines using direct-coupled time stepping finite element method,” Electric

machines and drives Conference, Seattle, WA, USA, pp.113-115, May 9-12,

1999.

[66] Jinyun Gan, K. T. Chau, Yong Wang, C. C. Chan and J. Z. Jiang, “Design

and analysis of a new permanent magnet brushless DC machine,” IEEE Trans

on magnetics, vol.36, no.5, September 2000.

[67] Seung-Chan Park, Byung I1 Kwon, Hee-Soo Yoon, Sung-Hong Won and

Young-Gyu Kang, “Analysis of exterior-rotor BLDC motor considering the

eddy current effect in the rotor steel shell,” IEEE Trans on magnetics, vol.35,

no.3, May 1999.

[68] Yong Wang, K. T. Chau, C. C. Chan and J. Z. Jiang, “Transient analysis

of a new outer-rotor permanent magnet brushless DC drive using circuit-

field-torque coupled time stepping finite element method,”IEEE Trans. on

magnetics, vol.38, no.2, pp.1297-1330, March 2002.

[69] C. W. Chan and H. L. Li, “Simulation of rotor rotation in electromagnetic

field problem using time stepping finite element technique,” IEEE Trans. on

Magnetics, vol.40, no.6, pp 1204-1209, September, 1997.

[70] B. I. Kwon, K. I. Woo, D. J. Kim, and S. C. Park, “Finite element analysis

for dynamic characteristics of an inverter-fed PMLSM by a new moving mesh

technique,” IEEE Trans. on Magnetics, vol.36, no.4, pp 1574-1577, July 2000.

178
[71] S. J. Salon, M. J. Lee, “Coupling of finite element and analytical solution

in the airgap of electric machines,” IEEE Trans. on Magnetics, vol.27, no.5,

September 1991.

[72] Chunt Mi, Gordon R. Slemon and Richard Bonert, “Modeling of iron loss of

surface-mounted permanent magnet synchronous motors,” 36th IAS Annual

meeting, vol.4, pp.2585-2591, Oct 2001.

[73] S. L. Ho and W. N. Fu, “A comprehensive approach to the solution of direct

couple multislice model of skewed rotor induction motors using time stepping

eddy current finite element method,” IEEE Trans on Magnetics, vol.33, no.3,

May 1997.

[74] Hui Tan, W. N. Fu and Jianzhong Jiang, “A multi-pole brushless DC motor

and its analysis using time stepping finite element method,” Proceeding of the

3rd. International Power Electronic and Motion Control Conference (PIEMC

2000), vol.3,pp.1308-1311, August 15-18, 2000.

[75] W. N. Fu, Z. J. Liu and C. Bi,“A dynamic model of disk drive motor and

its applications,” IEEE Trans on Magnetics, vol.38, no.2, pp.973-976, March

2002.

[76] G. H. Jang, J. H. Chang, D. P. Hong and K. S. Kim, “ Finite element analysis

of an electromechanical field of a BLDC motor considering speed control and

mechanical flexibility,” IEEE Trans on Magnetics, vol.38, no.2, pp.945-948,

March 2002.

[77] P. P. Silvester, R. L. Ferrari, “ Finite elements for electrical engineers (2nd.

edition),” Cambridge University Press, 1990.

[78] M. V. K. Chari, P. P. Silvester, “ Finite elements in electrical and magnetic

field problems,” Jhon Wiley, 1980.

179
[79] H. R. Schwarz, “ Finite element methods,” Academic Press, 1988.

[80] S. S. Rao, “ The Finite element method in engineering (2nd. edition),” Perg-

amon Press, 1989.

[81] K. H. Huebner, E. A. Thronton, “ The Finite element method for engineers,”

Jhon Wiley and Sons, 1982.

[82] Ping Zhou, “Field and circuit combined analysis of permanent magnet

synchronous motors,” Ph.D Thesis, Memorial University of Newfoundland,

Canada, May 1994.

[83] S. R. Hoole, “Computer-aided analysis and design of electromagnetic de-

vices,” Elsevier Science Publishing Co., Inc, 1989.

[84] D. A. Lowther, “Computer Aided Design in Magnetics,” 1986.

[85] Nay Lin Htun Aung, “Computation and analysis of performance of small PM

motors,” Master Thesis, National University of Singapore, 2002.

[86] Matthew N. O. Sadiku, “Numerical techniques in electromagnetics,” CRC

press, 1992.

[87] M. V. K. Chari, S. J. Salon, “Numerical methods in electromagnetism,” Im-

print San Diego : Academic Press, 2000.

[88] R. L. Stoll, “The analysis of eddy currents,” Oxford University Press,1974.

[89] D. C. Jites and D. L. Atherton, “Model of ferromagnetic hysteresis,” J. Apply

Physics, vol.75, No.5, pp.2594-2597, March 1994.

[90] T. F. Pirious, A. Razek, “Finite element analysis in electromagnetic systems

accounting for electric circuits,” IEEE Trans on Magnetics, vol.29, No.2,

pp.1669-1675, March 1993.

180
[91] A. Demenko, “Time-stepping FE analysis of electric motor drives with semi-

conductor converters,” IEEE Trans on Magnetics, vol.30, No.5, pp.3264-3267,

September,1994.

[92] T. H. Kim, J. H. Choi, K. C. Ko and J. Lee, “Finite element analysis of brush-

less DC motor considering freewheeling diodes and DC link voltage ripple,”

IEEE Trans on Magnetics, vol.39, No.5, pp.3274-3276, September,2003.

[93] F. Pirious, A. Razek, “Coupling of saturated electromagnetic systems to

nonlinear power electronic devices,” IEEE Trans on Magnetics, vol.24, No.1,

pp.274-277, January 1988.

[94] T. J. E. Miller, M. I. McGilp, D. A. Staton and J. J. Bremner, “Coupling

of saturated electromagnetic systems to nonlinear power electronic devices,”

IEE Proc. Electr. Power Appl., vol.146, No.2, pp.129-137, March 1999.

[95] frederick W. Grover, “Inductance calculation: Working formulas and tables,”

Dover Publications Inc.,1946.

[96] J. Engstrom, “Inductance of slotless machines,” IEE Proc. Electr. Power

Appl., vol.146, No.2, pp.129-137, March 1999.

[97] P. Salminen,“Fractional slot permanent magnet synchronous motors for low

speed applications,”Thesis for the degree of Doctor of Science (Technology)

Lappeenranta University of Technology, December 2004.

[98] S. L. Ho and W. N. Fu, “A fully automatic mesh generation method for the

movement field modeling,”The second Chinese Int. Conf. Electrical Machines,

Hangzhou, China, pp. 624-630, 31st August- 2nd September 1995.

[99] Richard M. Crowder,“Electric drives and their controls,” Oxford Science Pub-

lications, Claredon Press, 1995.

181
[100] P. L. George, “Automatic mesh generation: Application to finite element

methods,” John Wiley and Sons, England, 1991.

[101] D. T. Lee and B. J. Schachter, “Two algorithms for constructing a Delaunay

Triangulation,” International Journal of Computer and Information Sciences,

vol. 9, No.3, pp. 219-242, 1980.

[102] D. A. Lindholm, “Automatic triangular mesh generation on surfaces of poly-

hedra,” IEEE Trans. Magn. Vol.19, no.6, pp. 2539-2542, November 1983.

[103] P. P. Silvester, “Advances in electrical engineering software,” Computational

mechanics publications, Springer-Verlag, 1990.

[104] Y. Lauze, “Interactive generation of structure/unstructured surface meshes

using adaptivity,” The Third Inter. Conf. on Numerical Grid in Computa-

tional Fluid Dynamics and Related Fields, North -Holland, pp. 491-502, 1991.

[105] S. L. Ho, W. N. Fu and H. C. Wong “Application of automatic choice of

step size for time stepping finite element method to induction motor,” IEEE

Trans. Magn., vol.33, no.2, pp. 1370-1373, March 1997.

[106] C. William Gear, “Numerical initial vluse problems in ordinary differential

equations,” Prentice-Hall, Inc., Englewood Cliffs, 1971.

[107] S. L. Ho and W. N. Fu, “Computation of harmonics stray losses of induction

motors using adaptive time stepping finite element method coupled with

circuits,” Seventh International Electric Machines and Drives Conference,pp.

93-97, September 11-13, 1995.

[108] S. L. Ho, H. L. Li “Adjustment of step size for time stepping finite element

analysis to skewed rotor induction motors fed by SPWM source,” IEEE In-

ternational Electric Machines and Drives Conference, Seattle, Washington,

pp. 679-681, May 9-12, 1999.

182
[109] S. J. Salon, D. W. Burow, R. E. Ashley III, L. Ovacik and M. J. DeBortoli,

“Finite element analysis of induction machines in the frequency domain,”

IEEE Trans. Magn., vol.29, no.2, pp. 1438-1441, March 1993.

[110] R. Palma, S. J. Salon and M. J. DeBortili, “Dynamic electromechanical analy-

sis of induction machines using finite elements,” Proceedings of the Interna-

tional Conference on Electric Machines, pp.952-958, 1990.

[111] W. Muller, “Comparison of different methods of force calculations,” IEEE

Trans., Magn., vol.26, no.2, pp.1058-1061, 1990.

[112] M. Antila, E. Lantto and A. Arkkio, “Determination of forces and linearized

parameters of radial active magnetic bearings by finite element technique,”

IEEE Trans., Magn., vol.34, pp. 684-694, May 1998.

[113] M. Marinescu, N. Mainescu, “Numerical computation of torques in perma-

nent magnet motors by Maxwell’s stresses and energy method,” IEEE Trans.,

Magn., vol.24, no.1, January 1988.

[114] C. C. Hwang, S. B. John and S. S. Wu, “Reduction of cogging torque in

spindle motors for CD-ROM drive,” IEEE Trans., Magn., vol.34, no.2, pp.

468-770, March 1998.

[115] G. H. Jang and J. W. Yoon, “Torque and unbalanced magnetic force in a

rotational unsymmetric brushless DC motors,” IEEE Trans., Magn., vol.32,

no.5, pp. 5157-5159, 1996.

[116] C. S. Koh, H. S. Yoon, K. W. Nam and H. S. Choi, “Magnetic pole shape

optimization of permanent magnet motor for reduction of cogging torque,”

IEEE Trans., Magn., vol.33, no.2, pp.1822-1827, January 1988.

183
[117] T. S. Low, S. X. Chen and X. Gao, “Robust torque optimization for BLDC

spindle motors”, IEEE Trans. on Industrial Electronics, vol.48, No.3, pp.656-

663, June 2001.

[118] M. A. Jabbar, “Some novel ideas for disk drive spindle motors,” Motion

control proceedings, Asia-Pacific workshop on advances in Singapore, pp.171-

176, July 1993. ’

[119] S. X. Chen, T. S. Low, H. Lin and Z. J. Liu, “Design trends of spindle motors

for high performance hard disk drives,” IEEE Trans. Magn, vol.32,No.5, pp.

3848-3850, September 1996.

[120] M. A. Jabbar, “Disk drive spindle motors and their controls,” IEEE Transs

Industrial Elec., vol.43, No.2, pp. 271-350, April 1996.

[121] M. A. Jabbar, T. S. Tan and K. J. Binns, “Recent developments in disk drive

spindle motors,” Int. Conf. on Electrical Machines (ICEM’92), Manchester,

UK, vol.2, pp. 381-385, 1992.

[122] M. A. Jabbar, T. S. Low and M. A. Rahman, “Permanent magnet motors

for brushless operation,” IAS annual meeting, pp. 15-19, 1988.

[123] T. S. Low, C. S. Soh, C. Bi and K. T. Chang, “Identity control for spindle

motors in hard disk drives,” IEEE Trans. Magn. vol.31, No.5, pp.3117-3119,

November 1995.

184
List of Publications

1. H. N. Phyu, M. A. Jabbar and Z. J. Liu, “Numerical Modelling and Analy-

sis of a Brushless Permanent Magnet DC Motor using Circuit-Field Coupled

Time Stepping FEM,”The 15th Conference on the Computation of Electro-

magnetic Fields (COMPUMAG 2005), China, 26th - 30th June, 2005.

2. H. N. Phyu, M. A. Jabbar and Z. J. Liu, “Computational Optimization of

the Run-up Performance of a HDD Spindle Motor,” International Electrical

Machines and Drives Conference (IEMDC’05), San Antonio, TX, USA, 15th-

18th May, 2005.

3. H. N. Phyu, M. A. Jabbar and Z. J. Liu, “Numerical Modelling and Analysis

of a Brushless Permanent Magnet DC Motor using Circuit-Field Coupled

Time Stepping FEM,” IEEE Trans. On Magnetics. (Under Review)

4. M. A. Jabbar, H. N. Phyu, Z. J. Liu and C. Bi, “Modelling and Numerical

Simulation of Brushless Permanent Magnet DC Motors in Dynamic Condi-

tions by Time Stepping Techniques,” IEEE Transaction on Industry Appli-

cations, vol.40, No.3, pp. 763-770, May/ June 2004.

5. M. A. Jabbar, H. N. Phyu and Z. J. Liu, “Analysis of the Starting Process

of a Disk Drive Spindle Motor by Time Stepping Finite Element Method,”

IEEE Transaction on Magnetic, vol.40, No.4, pp. 3204-3206, July 2004.

6. M. A. Jabbar, H. N. Phyu and Z. J. Liu, “Analysis of the Starting Process

185
 of a Disk Drive Spindle Motor by Time Stepping Finite Element Method,”

The 9th MMM-Intermag 2004, Anaheim, California, USA, January 2004.

7. H. N. Phyu, M. A. Jabbar, Z. J. Liu and C. Bi, “Modelling and Simulation

of Brushless Permanent Magnet DC Motors in Dynamic Conditions by Time

Stepping Techniques,” IEEE International Electric Machines and Drives Con-

ference (IEMDC’03), Madison, Wisconsin, USA, Vol. 1, pp. 376-381, June

1- 4 2003.

8. M. A. Jabbar, H. N. Phyu and Z. J. Liu, “Modelling and Simulation of

Steady State Performance of Brushless Permanent Magnet DC Motors,” In-

ternational Conference on Applied Electromagnetics and Mechanics (ISEM

2003), Versailles, France, May 2003.

9. M. A. Jabbar, H. N. Phyu and Z. J. Liu, “Dynamic Modelling of Brush-

less Permanent Magnet DC Motors using Time Stepping Finite Element

Method,” International Conference on Evolutionary Computation (ICEC 2002),

Dhaka, Bangladesh, December 2002.

186
Appendix A

Motor Specification

Table A.1: Motor specifications

voltage 12V

No of poles 8

No of slots 12

Rated speed 7200rpm

Stator outside diameter 28mm

Rotor outside diameter 30mm

Stack length 4.5mm

Air gap 0.2mm

Magnet N dF eB

187
Appendix B

Newton Raphson Algorithm

The Newton Raphson method is based on derivatives of the function or quantity to

be updated, and the procedure yields rapid convergence, with the error in a given

step decreasing as the square of the error in the previous step. Therefore, it is one

of the quadratic methods. Consider, for example, the following non-linear equation

Ax = B (B.1)

Where x is unknown and A is the function of x. B is a constant. Define

f (x) = A(x).x (B.2)

Fig. B.1 shows the relationship between f (x) and B. The solution of equation

(B.1)is the intersection of the curve f (x) and the line B. Expand the equation

(B.2) by using the Taylor series

f (x) = f (x(0) ) + f 0 (x(0) )(x − x(0) ) (B.3)

where x( 0) is the initial value of the solution. We have

f (x) = A(x).x = B (B.4)

And

B = f (x(0) ) + f 0 (x(0) )(x − x(0) ) (B.5)

Therefore

f 0 (x(0) )(x − x(0) ) = B − f (x(0) ) (B.6)

188
Appendix B 189

Figure B.1: Relationship between f (x) and B

From the initial estimate, x( 0), we can find

f 0 (x(1) )(x(0) − x(1) ) = B − f (x(1) ) (B.7)

and so forth.

f 0 (x(2) )(x(1) − x(2) ) = B − f (x(2) ) (B.8)

The similar computation can be done continually to k times iterations as follow

f 0 (x(k) )(x(k+1) − x(k) ) = B − f (x(k) ) (B.9)

The procedure is illustrated graphically in Fig. B.2. After several iterations, the

right hand side of the equation (2.96) will close to zero. That is, the sequence

of points converges to the solution. It can be shown that if the function f (x) is

monotonic, the solution converges quadratically. However, if the function is not

monotonic, the derivative will change sign. This can lead to non-convergence as

illustrated in Fig. B.3. Caution is advised if the electrical machine is operated

in saturation region, the magnetic permeability vs flux density is not a monotonic

function. In order to guarantee convergence the permeability in the low flux density

region is usually approximated as a constant. Since most electrical machines are

Appendix B 190

not designed to operate in this region, therefore the approximation is acceptable.

Figure B.2: Newton Raphson procedure

Figure B.3: Effect of non-monotonic function on Newton’s method

Appendix C

Cubic Spline Interpolation

The development of cubic spline algorithm is as follow. Consider a set of sample

points, , i = 1(1) n on an interval [a, b] of a real line with the corresponding values

of the real function, yi , I = 1(1)n. The arrangement of the sample points is

monotonous, i.e.:

a = x1 < x2 < x3 < ....... < xn = b (C.1)

The function is to be approximated with and interpolating cubic spline function,

S(x). The spline function should fulfill the following conditions:

• S(x) is continuous within the interval [a, b] along with derivative up to the

second derivative.

• On every interval [xi , xi+1 ], S(x) is identical to a cubic polynomial called a

subspline.

• At the sample points,xi , i = 1(1)n, the points of of S(x) are y(xi ).

• The boundary conditions, s00 (a) = s00 (b) = 0 or are valid.

Using the approach,

S(x) = y(x) = ai .yi + bi .yi+1 + ci .yi00 + di .yi+1

00
(C.2)

In which:

hi = xi+1 − xi (C.3)

191
Appendix C 192

xi+1 − x
ai = (C.4)
xi+1 − xi
x − xi
bi = (C.5)
xi+1 − xi
(a3i − ai ) (xi+1 − xi )2
ci = (C.6)
6
(b3i − bi ) (xi+1 − xi )2
di = (C.7)
6

The first derivative with respect to x of the interpolating function is

dS (x) dy yi+1 − yi 3a2 − 1 3b2 − 1

= = − i (xi+1 − xi )yi00 + i 00
(xi+1 − xi )yi+1 (C.8)
dx dx xi+1 − xi 6 6

while the derivative is

d2 S (x) d2 y
= = ai yi00 + bi yi+1
00
(C.9)
dx2 dx2

With this, the requirement of the continuity of the second derivative over the

boundaries of the interval [xi , xi+1 ] and [xi−1 , xi ] is satisfied. Because of a required

continuous derivative of first order of the interpolating spline function, the values

of dS/dx at the point x = xi for x ∈ [xi−1 , xi ] and x ∈ [xi , xi+1 ] must be equal.

Employing for both interval yields

hi−1 00 hi + hi−1 00 hi 00 yi+1 − yi yi − yi−1

.yi−1 + .yi + .yi+1 = − (C.10)
6 3 6 hi hi−1

This equation can be evaluated with i = 2(1)n − 1 for every interval. This results

in n − 2 linear independent equations for the n unknown y100 = yn00 = 0 yields two

additional constraints. This is a symmetric diagonal system of equations and is

easy to solve. With the derivatives known, the coefficients of the interpolating

spline function are now determined.

Appendix D

Demagnetization Curve for

Permanent Magnet

Figure D.1: Demagnetization curve for bonded NdFeB magnet

193
Appendix E

Specifications of Inverter Circuit

Components

Inverter circuit for the experimental set up consists of the following two main

components.

1. High speed switching MOSFETs IRF620 and

2. IR2110 high side and low side drivers.

E.1 MOSFET IRF620

Detailed specifications sheets of MOSFET IRF620 are shown in Fig. E.1, Fig. E.2

and Fig. E.3

Figure E.1: Date sheets of absolute maximum ratings

194
Appendix E 195

Figure E.2: Thermal and electrical characteristics sheet (1)

Appendix E 196

Figure E.3: Thermal and electrical characteristics sheet (2)

E.2 IR2110 high side and low side driver

The IR2110 are high voltage, high speed power MOSFET and IGBT drivers with

independent high and low side referenced output channels. Proprietary HVIC

and latch immune CMOS technologies enable ruggedized monolithic construction.

Logic inputs are compatible with standard CMOS or LSTTL output, down to 3.3V

logic. The output drivers feature a high pulse cur- rent buffer stage designed for

minimum driver cross-conduction. Propagation delays are matched to simplify use

in high frequency applications. The floating channel can be used to drive an N-

channel power MOSFET or IGBT in the high side configuration which operates up

to 500 or 600 volts. Typical connection diagram and functional block diagram are

shown in Fig. E.4 and Fig. E.5. Data sheet of absolute maximum ratings from the

supplier is shown in Fig. E.6.

Appendix E 197

Figure E.4: Typical connection diagram

Figure E.5: Functional block diagram

Appendix E 198

Figure E.6: Absolute maximum ratings

Appendix F

Specifications of Incremental
Encoder

Figure F.1: Photograph of Scancon incremental encoder

Basic characteristics are listed below.

• Micro hollow shaft encoder

• Strong compact electronics

• Std. IP 54 (with IDC; IP 50)

• To be connected directly to PLC’S and counters

• Thermal shut down at 155C

199
Appendix F 200

• 5V ± 10%

• Based on 2 precision ball bearings for industrial environment

Electrical and mechanical characteristics are shown in Fig. F.2 and Fig. F.3.

Figure F.2: Electrical specifications

Appendix F 201

Figure F.3: Mechanical specifications

Appendix G

Program Structure for Steady

State Analysis

There are four main parts in the program.

1. Mesh Generation

2. Pre-computation by frequency domain analysis

3. Time stepping FEM computation

4. Post processing

202
Appendix G 203

1. Mesh Generation
Main input file : motor.dat
Start

Motor7.for Produce input data files

For the whole program

Size7.for
Prepare geometric data
For the mesh2d7.for

Mesh2d7.for Mesh2d7.for
Mesh generation using
Eating nodes algorithm

Adapt2d7.for Adapt2d7.for Refine FEM mesh by

using Delaunay method
Stator.dat Rotator.dat

Main input data files for Pre-computation part

2. Pre-computation
Main input data files : motor.dat ,stator.dat ,rotor.dat

Main control program for

Pre2d7.c
The pre-computation

Motor7.dat Produce data files for other programs

Size7.for Prepare geometric data

Con2d7.for Connect the stator mesh

and the rotor mesh

Pre-processor for generating

C2d-nb.for
address matrixes

C2d7.for FEM computation using complex model

Appendix G 204

3. Time stepping FEM computation

Main control program for
Move2d7.c
time stepping FEM

Produce data files for the

Motor7.for
whole program

Ro2d7.for
Rotate FEM mesh

Time stepping FEM

T2d7.for
computation

Get2d7.for Post computation

4. Post Processing
Compute desired parameters by using the output data come from the time stepping
FEM computation.
Prepare data for V2d7.C
Post2d7.for
(Compute magnetic flux density)

V2d7.C Show 2D FEM results such flux plot, flux

density.

Wave2d7.for Show waveforms against time

(stator current, torque, speed, etc.)