Dynamic Control of the

Permanent Magnet Assisted Reluctance Synchronous Machine

with Constant Current Angle


Hugo Werner de Kock


Thesis presented in partial fulfillment of the requirements for the degree of
Master of Science in Electrical and Electronic Engineering with Computer Science
at the University of Stellenbosch


Promoter:
Prof. Maarten J. Kamper, University of Stellenbosch


March 2006
ii
Declaration:

I, the undersigned, hereby declare that the work contained in this thesis is my own original work and
that I have not previously in its entirety or in part submitted it at any university for a degree.


H.W. de Kock


Signature

March 2006
Stellenbosch, South Africa



iii
Abstract

This thesis is about the dynamic control of a permanent magnet assisted reluctance synchronous
machine (PMA RSM). The PMA RSM in this thesis is a 110 kW traction machine and is ideal for the
use in electrical rail vehicles. An application of the dynamic control of the PMA RSM in electrical rail
vehicles is to reduce wheel slip.

The mathematical model of the PMA RSM is derived and explained in physical terms. Two methods
of current control for the PMA RSM are investigated, namely constant field current control (CFCC)
and constant current angle control (CCAC). It is shown that CCAC is more appropriate for the PMA
RSM.

A current controller for the PMA RSM that guarantees stability is derived and given as an analytic
formula. This current controller can be used for any method of current control, i.e. CFCC or CCAC.
An accurate simulation model for the PMA RSM is obtained using results from finite element analysis
(FEA). The accurate model is used in a simulation to verify CCAC. A normal proportional integral
speed controller for the PMA RSM is designed and the design is also verified by simulation.

Practical implementation of the current and speed controllers is considered along with a general
description of the entire drive system. The operation of the resolver (for position measurement) is
given in detail. Important safety measures and the design of the electronic circuitry to give protection
are shown. Practical results concerning current and speed control are then shown.

To improve the dynamic performance of the PMA RSM, a load torque observer with compensation
current feedback is investigated. Two observer structures are considered, namely the reduced state
observer and the full state observer. The derivation of the full state observer and the detail designs of
the observer elements are given. The accurate simulation model of the PMA RSM is used to verify the
operation of the observer structures and to evaluate the dynamic performance. Both observer
structures are implemented practically and practical results are shown.

One method of position sensorless control, namely the high frequency voltage injection method, is
discussed in terms of the PMA RSM. This work is additional to the thesis but it is shown, because it
raises some interesting questions regarding the dynamic control of the PMA RSM.

iv
Opsomming

Hierdie tesis gaan oor die dinamiese beheer van n permanente-magneet-ondersteunde reluktansie
sinchroon-masjien (PMO RSM). Die PMO RSM in hierdie tesis is n 110 kW trekkrag masjien en is
ideaal vir die gebruik in elektriese spoor voertuie. n Toepassing van die dinamiese beheer van die
PMO RSM in elektriese spoor voertuie is om wiel glip te verminder.

n Wiskundige model van die PMO RSM word afgelei en verduidelik in terme van fisiese begrippe.
Twee metodes van stroombeheer vir die PMO RSM, naamlik konstante vloedstroom beheer (KVB) en
konstante stroomhoek beheer (KSB), word ondersoek. Daar word aanbeveel dat KSB vir die PMO
RSM gebruik word.

n Stroombeheerder wat stabiliteit verseker word vir die PMO RSM afgelei en gegee as n analitiese
formule. Hierdie stroombeheerder kan gebruik word vir enige tipe stroombeheer, d.i. vir KVB of
KSB. Eindige element analise word gebruik om n akkurate simulasiemodel van die PMO RSM te
verkry. Hierdie simulasiemodel word dan gebruik om die werking van KSB te bevestig. n Normale
proportioneel-integraal spoedbeheerder word ontwerp en die werking van die spoedbeheerder word
bevestig deur simulasie.

Praktiese implementering van die stroom- en spoedbeheerders word beskryf en n geheelbeeld van die
totale beheerstelsel word getoon. Die werking van die resolver (vir posisie meting) word in detail
verduidelik. Belangrike beveiligingstappe en die ontwerp van elektronika wat die beveiliging daarstel,
word gegee. Praktiese stroom- en spoedmetings word ook getoon.

In n poging om die dinamiese gedrag van die PMO RSM te verbeter, word n draaimomentafskatter
met kompensasiestroom-terugvoer ondersoek. Twee tipes afskatters, naamlik die vereenvoudigde-
toestand-afskatter en die vol-toestand-afskatter word bestudeer. Die afleiding van die vol- toetstand-
afskatter en die detail ontwerp van al die afskatter-elemente word gegee. Die akkurate simulasiemodel
van die PMO RSM word gebruik om die werking van beide afskatters te toets. Beide afskatters word
prakties gemplementeer en die praktiese resultate word getoon.

Een metode van posisie-sensorlose-beheer, naamlik die ho-frekwensie spanningsinjeksie metode
word bespreek in terme van die PMO RSM. Hierdie werk is bykomend tot die tesis, maar word
getoon omdat dit interesante vrae aangaande die dinamiese beheer van die PMO RSM uitlok.
v
Acknowledgements

I would like to express my sincere appreciation to:

My promoter, Prof. Maarten Kamper, for his assistance, advice and encouragement.
Spoornet and the University of Stellenbosch for financial assistance.
The International Office at the University of Stellenbosch for assistance regarding my
exchange visit to the University of Wuppertal, Germany.
Mr. Andr Cabral Ferreira at the University of Wuppertal for his assistance regarding
sensorless control.
Mr. Aniel le Roux for designing the digital signal processor and writing firmware for it and
also for his assistance during initial stages of the control program development.
My friend, Mr. Richard Brady, for his support, encouragement and advice.
Mr. Francois Rossouw and Mr. Andr Swart for their technical support regarding the practical
test setup.
Mr. Yongle Ai, Mr. Edward Rakgati, Mr. Arnold Rix and Dr. Rong-Jie Wang for their
support and assistance.
My parents and my sister for their endless love and support.

vi














I know this world is ruled by infinite intelligence.
Everything that surrounds us everything that exists
proves that there are infinite laws behind it. There can be
no denying this fact. It is mathematical in its precision.

Thomas A. Edison
1847 1931


vii
Glossary of symbols and abbreviations

These symbols and abbreviations are in no particular order. In some equations uppercase letters are
used this refers to the steady state.

Symbol Meaning Unit
magnetic flux Weber [Wb]
or current angle degrees or radians
flux linkage angle degrees or radians
i current Ampere [A]
e induced voltage Volt [V]
v voltage Volt [V]
T torque [kg.m
2
]
r or R resistance Ohm []
rotational speed Radians per second [rad/sec]
L self-inductance Henry [H]
M mutual inductance Henry [H]
J inertia Newton second [Ns]
B friction coefficient Newton metre second per radian
magnetic flux linkage Weber turns [Wb turns]
p pole pairs scalar
P power Watt [W]
dt
d
time derivative scalar
t

partial time derivative scalar

rotor position degrees or radians
or power factor angle degrees or radians
inner power factor angle degrees or radians
j square root of -1 scalar
T
K torque coefficient Newton metre per ampere [Nm/A]


viii
Abbreviation Meaning
DC direct current
RSM reluctance synchronous machine
AC alternation current
PMA RSM permanent magnet assisted RSM
FEA finite element analysis
PM permanent magnet
d-axis direct axis
q-axis quadrature axis
Fig. figure
ABC stationary reference frame
QD0 synchronously rotating reference frame
CFCC constant field current control
CCAC constant current angle control
rms root mean squared
ZOH zero order hold
S-plane continuous plane
Z-plane discrete plane
W-plane bilinear transform plane
LHP left hand plane
RHP right hand plane
Z{} the Z-transform
DSP digital signal processor
ADC analogue-to-digital converter
DAC digital-to-analogue converter
LPF low pass filter
LUT lookup table
PI proportional integral
P proportional
DSP digital signal processor
FPGA field programmable gate array
ISR interrupt service routine
IGBT insulated gate bipolar transistor

ix
Table of Contents

Chapter 1 Introduction ........................................................................................................... 1
1.1 Background ....................................................................................................................... 1
1.2 Problem Statement ............................................................................................................ 7
1.3. Approach to the problem................................................................................................... 8
1.4. Thesis layout...................................................................................................................... 9

Chapter 2 Mathematical model for the PMA RSM............................................................. 10
2.1 The mathematical model for the RSM............................................................................ 10
2.2 The mathematical model for the PMA RSM................................................................... 17
2.3. Power factor and inner power factor ............................................................................... 19
2.4. Parameter values.............................................................................................................. 20
2.5. Method of current control for RSM and PMA RSM....................................................... 24

Chapter 3 Design methodology for the current and speed controllers................................. 30
3.1. Different methods to design a digital controller.............................................................. 30
3.2. Decoupling procedure for the electrical model ............................................................... 31
3.3. Open loop current response............................................................................................. 33
3.4. Closed loop current response .......................................................................................... 35
3.5. The discrete nature of the controller................................................................................ 36
3.6. Design in the W-plane using a Bode plot and gain margin specification ....................... 39
3.7. Revised current controllers.............................................................................................. 44
3.8. Implementation of the current controllers in simulation................................................. 45
3.9. The inverter model and the modified electrical machine model ..................................... 49
3.10. Design of the speed controller......................................................................................... 53
3.11. Implementation of the speed controller in simulation..................................................... 56
3.12. Simulation results for the speed controller...................................................................... 60
3.13. Noise in the measured speed signal................................................................................. 63
3.14. An alternative speed controller........................................................................................ 63
3.15. Summary for design methodology .................................................................................. 64

x
Chapter 4 Practical considerations....................................................................................... 65
4.1. Representation of the machine as seen from the load ..................................................... 66
4.2. The rectifier ..................................................................................................................... 67
4.3. The dumping circuit and the warning light circuit .......................................................... 68
4.4. The load system............................................................................................................... 73
4.5. The inverter ..................................................................................................................... 73
4.6. DC bus voltage and three-phase current measurement ................................................... 75
4.7. Rotor angular position measurement using a resolver .................................................... 78
4.8. Zero position.................................................................................................................... 85
4.9. Overview of the DSP....................................................................................................... 88
4.10. The control program........................................................................................................ 89
4.11. Practical measurements ................................................................................................... 92

Chapter 5 Load torque observer........................................................................................... 95
5.1. Why is the load torque observer needed?........................................................................ 95
5.2. Load torque observer for RSM with CFCC .................................................................... 98
5.3. Reduced state observer for RSM (or PMA RSM) with CCAC..................................... 103
5.4. Full state observer for RSM (or PMA RSM) with CCAC............................................ 114
5.5. Summary for load torque observer with compensation current .................................... 129

Chapter 6 Position sensorless control ................................................................................ 130
6.1. High frequency voltage injection .................................................................................. 131
6.2. Analysis of the current response and the rotor angle tracking scheme ......................... 134

Chapter 7 Summary, conclusions and recommendations .................................................. 139
7.1. Conclusions ................................................................................................................... 139
7.2. Recommendations ......................................................................................................... 140
xi
Appendix R References ..................................................................................................... 141
R.1. Articles on CDROM...................................................................................................... 141
R.2. Books............................................................................................................................. 142
R.3. P-CAD 2000 circuit diagram on CDROM.................................................................... 143
R.4. Datasheets / Brochures on CDROM............................................................................. 143
R.6. Matlab 7 analysis on CDROM...................................................................................... 144
R.7. Programs on CDROM................................................................................................... 144
R.8. Matlab 7 simulation files on CDROM.......................................................................... 144
R.9. Videos on CDROM....................................................................................................... 145
R.10. Websites ........................................................................................................................ 145

Appendix A Torque equation............................................................................................. 146

Appendix B Zero order hold and the Z-transform............................................................. 150

Appendix C The rotating magnetic field ........................................................................... 156

Appendix D Park transform in the time domain................................................................ 158

Appendix E Lookup tables ................................................................................................ 162

Appendix F Signal to noise ratio....................................................................................... 174

Appendix G Full state observer ......................................................................................... 175

Appendix H Additional simulation results......................................................................... 179

1
Chapter 1 Introduction

This chapter provides some background information to orientate the reader with regard to the field of
electrical machines and drives. The problem, which is addressed in this thesis, will then be presented.
An overview of the problem solving approach and the thesis layout follows.


1.1 Background

Many types of electrical machines that can act as motors or generators exist today; these include
amongst others the DC machine, induction machine and wound rotor synchronous machine. These
machines have been studied and perfected in their design over many years. One machine that did not
receive much attention is the reluctance synchronous machine (RSM)
1
, because it was known to a have
very poor performance. This is true when the machine is controlled in an open-loop manner.
However, by controlling the machine in a closed loop with feedback, its performance is comparable
with other popular machines. Indeed, the RSM has recently received renewed attention [A8], mainly
due to the modern field orientated control strategies [A9][A10][A21].

In the market of general-purpose drives, the induction motor still represents a standard as an off-the-
shelf motor [A11], but there is now a shift towards synchronous motor based drives due to guaranteed
efficiency. To be a competitive product in terms of general-purpose applications, a drive must be both
low-cost and well suited to sensorless control
2
. Research shows that the RSM is a good candidate
[A11][A12].

A very basic explanation of the electrical machines mentioned above will now be given for the sole
purpose of orienting the reader. It is interesting to note that the separately excited DC motor and the
RSM can be controlled in the same way. This will be pointed out in the text to follow.


1
Many authors use the term synchronous reluctance machine
2
This is control without a physical position (or speed) sensor such as a resolver or encoder

2
The DC machine
There are two types of DC machines namely permanent magnet DC machines, which give a constant
field flux
f
, and field winding DC machines, which give a variable field flux according to a field
current
f f f
i k = [B1, p.99]. The DC machine has the armature windings in the rotor in contrast to
the AC machine which has the armature windings in the stator. The DC machine gives mechanical
rectification of the AC voltage that is induced in its rotor. The field winding of the DC machine may
be connected in series with the armature winding, or the DC machine may be excited separately.

The machine equations for the separately excited DC machine are as follows:

a f t em
i k T = (1.1.1)
m f t a
k e = (1.1.2)
dt
di
L i R e v
a
a a a a t
'
+ + = (1.1.3)
) (t T B
dt
d
J T
wl m
m
em
+ + =

(1.1.4)

In the equations above the symbols have the following meaning:
em
T is the electromagnetic torque produced by the interaction between the armature current
a
i and the
field flux
f
. The back-EMF
a
e induced in the rotor, is proportional to the field flux and the rotor
speed
m
. To establish the armature current, a terminal voltage
t
v is applied. The steady state value
of the armature current is determined according to equation (1.1.3) by the applied terminal voltage,
back-EMF and the stator winding resistance
a
R . The rate of change of the armature current is limited
by the self inductance
'
a
L . The rotor speed is a function of the produced electromagnetic torque, the
load torque
wl
T , the inertia of the rotor J and the friction coefficient B .
t
k is a constant of
proportionality.


3
In a separately excited DC motor, the field flux is produced by a separately controlled field current.
Note that in the equivalent circuit for the separately excited DC motor, shown in Fig. 1.1, the circuits
for the field and armature are separated.

Fig. 1.1. Equivalent circuit of the separately excited DC motor.

In the steady state, with equation (1.1.2) substituted into equation (1.1.3), the separately excited DC
motor can be described by the following equations:

a f t em
I k T = (1.1.5)
a a m f e t
I R k V + = (1.1.6)
f f f
I R V = (1.1.7)

|
|

\
|
=
em
f t
a
t
f e
m
T
k
R
V
k

1
(1.1.8)

From equation (1.1.8) it is clear that the speed can be increased linearly by keeping both field and
torque constant
3
and increasing the terminal voltage linearly. In this range the armature current and
torque are kept constant at their rated values, therefore it is known as the constant torque region.

If the terminal voltage has reached its rated value, the speed can be further increased by reducing the
field
4
in the machine and keeping the terminal voltage constant. In this region constant torque cannot
be maintained because that would require the armature current in equation (1.1.5) to increase above its
rated value. The torque will therefore have to decrease as the speed increases beyond the rated speed.
This region is known as the constant power region, since both current and voltage are kept constant.
Fig. 1.2 shows (in per unit quantities) the two regions mentioned.

3
This is called constant field control
4
This is called field weakening
Ra
+
Vt
+
Ia
Lf
-
flux
-
Ea
La
Vf
+
Rf
If

4
Per Unit
m

Per Unit
f f a em
i i T , , ,
f f em
i T , ,
t
v
a
e
1.0
1.0
a
e
t
v
a
i ,

Fig. 1.2. Constant torque and constant power regions for the DC motor.

The figure above is with reference to the separately excited DC motor, but this graph can be produced
for any kind of electrical motor and generally follows the same patterns. It is usually the aim of
electrical machine designers to make the constant power region large and to minimize the loss in
torque in the constant power region.

The induction machine
By introducing three-phase power, it is possible to create a rotating magnetic field (this is explained
thoroughly in Appendix C). The induction machine [B1, p.263] uses three-phase current to create a
rotating magnetic field in its stator. The rotating magnetic field induces voltage in the rotor and if
current is allowed to flow (for instance by connecting the rotor to external resistors), then there are
current carrying conductors in a magnetic field and thus torque is produced. The produced torque
causes rotation to occur. It is important to note that the rotor does not rotate at the same speed as the
rotating magnetic field (synchronous speed): the rotor lags behind the rotating field in motor operation
and the rotating field lags behind the rotor in generator operation.

Fig. 1.3. Induction machine with wound rotor and external resistors.

The wound rotor synchronous machine
Injecting a current into the rotor, such that the magnetic poles created by this action can lock with the
rotating magnetic field poles due to the three-phase stator currents, causes the rotor to turn at
synchronous speed. This is the basic operation of a wound rotor synchronous machine.

5

Fig. 1.4. Wound rotor synchronous machine.

The reluctance synchronous machine
If the rotor is magnetically symmetrical and no currents can flow in the rotor, then it will not follow
the rotating magnetic field created by the three-phase stator currents (in fact it will not turn at all). If
the rotor is magnetically asymmetrical
5
however, it will follow the rotating magnetic field exactly due
to the reluctance force. The rotor will therefore turn at synchronous speed. This is the basic operation
of a reluctance synchronous machine (RSM).


Fig. 1.5. Two pole RSM.

From the explanations above, it is clear that the rotor design of the RSM is much different from both
the induction machine and wound rotor synchronous machine, while the stator design is very similar.
Comparing the RSM and DC machine, the RSM has the advantage that it has no brushes or rotor
windings, thus requiring much less maintenance. Control of the RSM is simpler because there are no
rotor parameters to be identified [B2, p.1]. The cooling of the RSM is also easier.

Comparative studies carried out on the performance of the RSM and induction machine in the low-
and medium-power range show that the RSM has higher torque density and efficiency [A2][A3][A4].
It was found that the RSM has more advantages than the induction machine, because the absence of
rotor currents leads to a simple vector control scheme and insignificant rotor losses. The rotor
manufacturing cost for the RSM is relatively low compared to that of the induction machine.

5
Also referred to as magnetic anisotropy or magnetic saliency

6
Even though feedback control increases the performance of the RSM dramatically, there are still some
inherent disadvantages: it has a lower power factor and its performance deteriorates fast in the flux-
weakening region compared to the induction machine [A1]. To address this problem, permanent
magnets (PMs) can be placed inside the flux barriers of the reluctance rotor. These drives are defined
as PM-assisted RSM drives [A5][A6][A7].

This thesis is concerned with the control of a permanent magnet assisted reluctance synchronous
machine (PMA RSM). The rated power of the machine is 110 kW. The rotor was first designed to be
a normal RSM [B17] and was then upgraded to be a PMA RSM [B5]. In both cases finite element
analysis (FEA) with an optimization procedure [B2, p.8] was used. A graphical representation of the
optimally designed PMA RSM is shown in Fig. 1.6. From Fig. 1.6 it is clear that the PM sheets are
curved. A picture of the actual PM sheets is shown in Fig. 1.7.

Fig. 1.6. Optimally designed cross section segment of the PMA RSM.


Fig. 1.7. PM sheets that are placed inside the rotor flux barriers of the RSM.

7
A cross-section of the rotor with the PM sheets in the flux barriers is shown in Fig. 1.8. From this
figure it is clear that there is a direction from the centre outward that has no flux barriers in its path:
this is called the direct axis (D-axis). Flux can travel freely in this direction and is only limited by the
saturation of the steel. The other direction has flux barriers in its path, which makes it difficult for the
flux to travel: this is called the quadrature axis (Q-axis). In fact, there are three D-axes and three Q-
axes, because this is a six-pole rotor. This rotor is magnetically asymmetrical, because flux travels
freely in the one direction, but not in the other.


Fig. 1.8. Complete six-pole rotor of PMA RSM.


1.2 Problem Statement

The problem is to get the best dynamic performance from the PMA RSM, but at the same time to
ensure energy efficient operation. Dynamic performance is the ability of the machine to follow the
speed reference as closely as possible under varying load conditions. The load, for example, might fall
away suddenly in which case the motor will accelerate quickly. It is then the aim of the controller to
bring the motor speed back to the reference speed as fast as possible. The torque equation of the RSM
(see Appendix A) is given by:
q d q d em
i i L L p T = ) ( )
2
3
( (1.2.1)
By keeping the D-axis current constant in (1.2.1), the D-axis self inductance is also constant. The Q-
axis inductance is almost constant, since there is almost no saturation of flux in this direction. In this
method, the torque is therefore only a function of the Q-axis current. This means that the machine is
controlled in a constant field manner, just like the separately excited DC motor.

Constant field control can be viewed as a standard method of controlling the RSM. This method is
however not energy efficient since the field current (D-axis current) is always present even under no
Q-axis
D-axis

8
load conditions. A comparative study [B3] shows that the constant current angle method of
controlling the RSM is more energy efficient, giving maximum torque per ampere, but results in
slightly slower dynamic performance. This can be explained briefly: the rate of change of current is
limited by self inductance, i.e. a greater self inductance results in a lower rate of change of current.
The D-axis self inductance is greater than the Q-axis self inductance and therefore the D-axis current
response is slower than the Q-axis current response. With constant field control, it is only the Q-axis
current that needs to change and therefore this method gives a fast current (and torque) response. With
constant current angle control however, both Q- and D-axis currents are being changed, so this means
the total response time is the slowest response time, which is the D-axis current response time.

The question is therefore what method of current control should be used? The author investigates
the possibilities and proposes additional methods to increase the response time of the machine. It was
also stated that in order to be competitive in general-purpose applications, the machine needs to be
controlled without a position sensor. One method of position sensorless control for the PMA RSM is
explained in concept, although this is not the main focus of the thesis.


1.3. Approach to the problem

A theoretical model for the PMA RSM is derived and the viability of constant field current control and
constant current angle control is then explored. An accurate model for the PMA RSM is obtained
using results from finite element analysis (FEA). It is then possible to design stable current controllers
and a speed controller. Simulation is used to verify these controllers. The controllers are implemented
practically and practical results are compared to the simulation results.

The load torque observer with compensation current feedback is then introduced as a way to improve
the dynamic performance of the machine drive. Again, theoretical development is verified with
simulation and compared to practical results. Finally, one method of position sensorless control for the
PMA RSM is presented.



9
1.4. Thesis layout

The layout of the remainder of this thesis is as follows:

Chapter 2: A model for the PMA RSM is derived theoretically and FEA is used to obtain the
actual variables, namely flux linkages, self inductances and mutual inductances. The
advantages of the PMs are discussed. Finally, the viability of constant field control
and constant current angle control is investigated.

Chapter 3: The design methodology for stable current controllers and the speed controller is
presented. The operation of the controllers is verified by simulation.

Chapter 4: Practical considerations are presented together with practically measured results of the
current and speed control.

Chapter 5: Theoretical development of the load torque observer with compensation current
feedback, as a method of increasing dynamic response, is presented. Simulation
results are compared with practically measured results.

Chapter 6: Position sensorless control using high frequency voltage injection is presented with
respect to the PMA RSM. A simulation block diagram is shown, although the detail
designs of all the elements in the block diagram are not shown, nor any simulation
results. Practical results for the method applied to a small RSM in Wuppertal,
Germany is shown to prove that the method works.

Chapter 7: In this chapter a summary with conclusions is given and recommendations are made
for further research.

10
Chapter 2 Mathematical model for the PMA RSM

In this chapter a mathematical model for the RSM will be developed, which is followed by a
mathematical model for the PMA RSM. A critical evaluation of the current control method for both
machines is then performed.

The mathematical machine model is necessary to understand how the machine works and to be able to
construct a simulation model. Simulation results will eventually be compared with practically
measured results. If the simulation model is accurate, the results should be very similar. The
simulation of the controlled PMA RSM is discussed in Chapter 3.

To understand the physical difference between the RSM and the PMA RSM, consider Fig. 2.1. In the
figure, a cross-section of the RSM (left) and a cross-section of the RSM with permanent magnet sheets
incorporated into the reluctance rotor (right), are shown. The permanent magnets have certain
beneficial effects which are described later in this chapter.


Fig. 2.1. Cross section of the RSM (left) and PMA RSM (right).


2.1 The mathematical model for the RSM

Since the rotor of the RSM does not contain any windings, the electrical model is based only on the
stator. The induced stator voltage is given by Faradays Law and the copper loss component also has
to be taken into account in the voltage equation as follows:
abc s
abc
abc
i r
dt
d
v + =

(2.1.1)
Flux barrier PM sheet

11
The equation above is in the stationary ABC reference frame which is fixed to the stator. The stator
windings are represented as three stationary coils displaced by 120 spatially. The model of the
machine can be simplified by using Parks transform (see Appendix A). The voltage equations in the
QD0 reference frame, which is fixed to the rotor, are given by:
q e
d
d s d
dt
d
i r v

+ = (2.1.2)
d e
q
q s q
dt
d
i r v

+ + = (2.1.3)
The zero-component is indeed zero when balanced three-phase voltages are applied, so it is omitted
here. It is important to gain physical insight into equations (2.1.2) and (2.1.3). The rest of this
subsection explains these equations as well as the torque equation in detail.

The Park transform changes the voltage equations from a stationary ABC reference frame to a
synchronously rotating QD0 reference frame. It is a one-to-one mapping that is dependent on the rotor
position. Note that the QD0 equations are still as seen from the stator. The stator is now represented
by two coils displaced by 90 spatially. The two stator coils rotate synchronously with the rotor. This
representation is shown in Fig. 2.1.1.
magnetic
D-axis
magnetic
Q-axis
D
D'
Q Q'
d

Fig. 2.1.1. Park transform representation of stator coils.

Fig. 2.1.1 is a two-dimensional representation of a three-dimensional object. In this representation
there are two coils to consider. Each coil lies flat on its own plane. The two planes are perpendicular
to each other and also perpendicular to the page. Therefore,
d
i flows in a coil that lies on a vertical
plane (the D-plane), perpendicular to the page and
q
i flows in a coil that lies on a horizontal plane (the
Q-plane) that is also perpendicular to the page. A current in a coil produces magnetic flux linkage that
moves through the area spanned by the coil in a direction given by the right hand rule. For example, if
d
i flows in the direction indicated in Fig. 2.1.1 (in at the cross and out at the dot), then flux linkage
moves perpendicularly through the D-plane in the direction of the magnetic D-axis.

12

The currents and flux linkages can be displayed together on a phasor diagram. In phasor diagrams,
peak values are always used to indicate magnitude and not root mean squared (rms) values. The QD0
currents and flux linkages are DC quantities however, and therefore the notion of peak values or rms
values is not applicable (see Appendix D). In this thesis, the unit for the currents in the QD0 reference
frame are therefore simply [A] and not [A rms] or even [A peak].

In the phasor diagram
d
i and
d
are in the same direction, although it is known that the current
actually moves in a plane and the flux linkage moves through that plane. It is convenient to speak of
the D-axis current and the D-axis flux linkage. The phasor diagram Fig. 2.1.2, shows the D-axis
current, D-axis flux linkage, Q-axis current, Q-axis flux linkage, stator current, stator flux linkage,
current angle and flux angle.
Q-axis
D-axis
d

d
i
q
i
q

s
i

Fig. 2.1.2. Vector diagram for currents and flux linkages in the QD0 reference frame.

In Fig. 2.1.2, the stator current,
s
i , is the vector sum of the D-axis current and Q-axis current
components. It should be noted that the magnitude of the stator current gives the peak amplitude of
the currents in the ABC reference frame (see Appendix D). Therefore, if the rated current of the
machine is peak A rms A i
rated
283 200 = = , then A i
rated s
283 = .

Using Fig. 2.1.1 and Fig. 2.1.2, the D-axis voltage equation is given by ( )
q d d s d
j
dt
d
i r v + + = in
the time domain, and by ( )
q d d s d
j s i r v + + = in the Laplace domain. This equation should be
evaluated at the appropriate frequency, which in this case is the electrical rotational speed of the
machine
e
. Therefore ( )
q d e d s d
j j i r v + + = , which can be written as
e q d e d s d
j i r v + = .

13

Transforming back to the Laplace domain:
re q d d s d
s i r v + = and finally to the time
domain:
re q d d s d
dt
d
i r v + = . This is not a physical explanation of equation (2.1.2), but it
gives a mathematical explanation for the negative speed voltage term
re q
. Using Fig. 2.1.1 and
Fig. 2.1.2, the Q-axis voltage equation is given by ( )
d q q s q
j
dt
d
i r v + = . Following the same
reasoning as for the D-axis voltage equation, the Q-axis voltage equation (2.1.3) can be derived.

Appendix A shows that the torque equation for the RSM is given by ) ( )
2
3
(
d q q d em
i i p T = ,
using the Park transform. This equation is motivated further using Fig. 2.1.1 and Fig. 2.1.2. Torque is
generated by the interaction between flux linkage and current, which is given by:
) sin( = = i i i T (2.1.4)

Equation (2.1.4) shows that if the current vector and flux linkage vector are displaced 90 spatially,
then maximum torque can be expected. The direction of the torque is given by the thumb of the right
hand if the other fingers of the right hand are curled from the flux linkage vector to the current vector.
In Fig. 2.1.2, the D-axis flux linkage with the Q-axis current gives a positive torque (out of the page,
towards the reader), while the Q-axis flux linkage with the D-axis current gives a negative torque (into
the page, away from the reader). The sum of these torque components gives the total torque, i.e.
d q q d total
i i T = .

This derivation however, is for one D-axis and one Q-axis and therefore a one pole-pair machine. For
a machine with p pole-pairs, the torque needs to be scaled with p. Furthermore, the power in the ABC
reference frame must be equal to the power in the QD0 reference frame, which leads to the scaling
factor of )
2
3
( in the QD0 reference frame. The torque equation should therefore by scaled by )
2
3
( .
The correct torque equation is thus given by:
( )
d q q d em
i i p T = )
2
3
( (2.1.5)

Equivalently, using magnitudes and angles in Fig. 2.1.2, the total torque is given by
) sin( = =
s s s s total
i i T and it should be scaled to give the correct torque:

14
[ ] ) sin(
2
3
=
s s em
i p T . Expanding this equation using a trigonometry identity:
( ) sin cos cos sin
2
3
=
s s em
i p T . Taking the stator flux and stator current into the bracket:
( ) sin cos cos sin
2
3
s s s s em
i i p T = , which is equivalent to equation (2.1.5).

The speed voltage terms
q e
and
d e
+ in equations (2.1.2) and (2.1.3) can be explained
physically using a torque concept. Three-dimensional views of the coils for the Q-axis and D-axis
currents are shown in Fig. 2.1.3 and Fig. 2.1.4.
q
i
d

T
q-axis
d-axis
T
q
V
+
q
i
q
i
q
i
d

Fig. 2.1.3. Q-axis coil in the QD0 representation.

d
i
d
i
d
i
d
i
d

T
q-axis
d-axis
q

T
d
V
+

Fig. 2.1.4. D-axis coil in the QD0 representation.


15
First consider the Q-axis coil, shown in Fig. 2.1.3. The applied Q-axis voltage causes a Q-axis current
to flow, which is limited by the resistance of the coil, r
s
. This current in the coil causes a magnetic
field that moves perpendicularly through the area spanned by the coil, in a direction given by the right
hand rule. At the same time, the applied D-axis voltage causes a D-axis current to flow, and the
current in turn causes a magnetic field that moves perpendicularly through the area spanned by the coil
(Fig. 2.1.4).

In both Fig. 2.1.3 and Fig. 2.1.4, there is one D-axis and one Q-axis and therefore this represents a
singe pole-pair machine for which the electrical speed and mechanical speed are the same. It is thus
valid to say that the relationship between power and torque is given by
e
T P = . Power can be
expressed in terms of voltage and current, while torque can be expressed in terms of flux linkage and
current. Therefore
e
i i v = , and so
e
v = .

The torque generated by the interaction of the D-axis flux linkage and the Q-axis current (Fig. 2.1.3) is
in an anti-clockwise direction and is therefore positive according the vector diagram convention. The
torque generated by the interaction of the Q-axis flux linkage and the D-axis current (Fig. 2.1.4) is in a
clockwise direction and is therefore negative. This explains the signs of the speed voltage terms in
equations (2.1.2) and (2.1.3).

In a RSM with
q d
i i = , note that
q d
>> due to the flux barriers. Therefore the positive torque
shown in Fig. 2.1.3 is much larger than the negative torque shown in Fig. 2.1.4. By reducing the Q-
axis flux linkage even further, the net positive torque will increase. This can be accomplished by
introducing PMs that generate a constant flux linkage in the negative Q-axis direction. This explains
why the PMs are placed inside the flux barriers, as shown in Fig. 2.1, and why they produce flux
linkage in the negative Q-axis direction.

In equations (2.1.2) and (2.1.3), both D-axis and Q-axis flux linkages are functions of the D-axis
current, Q-axis current and the electrical rotor position. Taking the Q-axis for example, the total
differential
q
d is defined in terms of the differentials
d
di ,
q
di and
e
d as follows [B16, p.947]:

e
e
q
q
q
q
d
d
q
q
d di
i
di
i
d

= (2.1.6)
Using the dash notation for partial derivatives, the partial derivatives of the flux linkage with respect to
the currents can be written as self inductance and mutual inductance:

e
e
q
q q d q q
d di L di M d

+ + =
' '
(2.1.7)

16


From equation (2.1.7), the time derivative of the Q-axis flux linkage is given by:
dt
d
dt
di
L
dt
di
M
dt
d
e
e
q q
q
d
q
q

+ + =
' '
(2.1.8)
If the rotor is skewed, the change in flux linkage due to rotor position is relatively small and can be
omitted. The rotor of the PMA RSM however, is not skewed, because a skewed rotor means that the
PMs have to be skewed as well. It is assumed in further analysis, that the change in flux linkage due to
the change in rotor position is negligible. Equation (2.1.8) is thus approximated by:
dt
di
L
dt
di
M
dt
d
q
q
d
q
q
' '
+ =

(2.1.9)

By analogy, the derivative of the D-axis flux linkage can be expressed as:
dt
di
M
dt
di
L
dt
d
q
d
d
d
d ' '
+ =

(2.1.10)
Substituting equations (2.1.9) and (2.1.10) into equations (2.1.2) and (2.1.3), the electrical model of the
machine is complete and is given by equations (2.1.11) and (2.1.12).
q e
q
d
d
d d s d
dt
di
M
dt
di
L i r v + + =
' '
(2.1.11)
d e
d
q
q
q q s q
dt
di
M
dt
di
L i r v + + + =
' '
(2.1.12)
For a complete model it is necessary to include a mechanical model and then to combine the electrical
and mechanical models. The mechanical equations are given by:
L m eq
m
eq em
T
dt
d
J T + + =

(2.1.13)
p
e
m

= (2.1.14)
Equation (2.1.13) states that the produced electro-mechanical torque is equal to the sum of three terms.
The first term takes into account the equivalent inertia, the second term accounts for the friction, and
the third term is the load torque that is to be driven by the machine. Equation (2.1.14) relates the
mechanical speed of the rotor
m
, to the speed at which the magnetic field rotates
e
, with the
number of pole-pairs, p.

To connect the mechanical and electrical models, the torque equation, which is obtained using the
equivalent circuit of the RSM and the power equation (see Appendix A), is used:

17
( )
d q q d em
i i
p
T =
2
3
(2.1.15)
Equations (2.1.11) through to (2.1.15) represent the complete mathematical model for the RSM.


2.2 The mathematical model for the PMA RSM

The mathematical model for the RSM is given by equations (2.1.11) through to (2.1.15). These
equations need to be modified if PMs are added to the rotor. Simply substituting
pm q
in the
place of
q
is all that is necessary. Note that the flux linkage due to the permanent magnets is constant
and the derivative thereof is zero. The motivation behind the use of permanent magnets is the effect
they have on the applied voltage, produced torque, power factor and inner power factor, as will be
pointed out in the text that follows.

The RSM voltage equations (2.1.11) and (2.1.12) are modified to the PMA RSM voltage equations by
substitution:
( )
pm q e
d
d s d
dt
d
i r v

+ = (2.2.1)
( )
d e
pm q
q s q
dt
d
i r v

+

+ = (2.2.2)
Since the flux linkage due to the PMs is constant, equation (2.2.2) reduces to:

d e
q
q s q
dt
d
i r v

+ + = (2.2.3)
From equations (2.2.1) and (2.2.3), the PMA RSM voltage equations are thus:
( )
pm q e
q
q
d d
d
d
d s d
dt
di
i dt
di
i
i r v

+ =
d e
q
q
q
d
d
q
q s q
dt
di
i dt
di
i
i r v

+

+ = ,
which can be written as:
( )
pm q e
q
d
d
d d s d
dt
di
M
dt
di
L i r v + + =
' '
(2.2.4)
d e
d
q
q
q q s q
dt
di
M
dt
di
L i r v + + + =
' '
(2.2.5)


18
From equation (2.2.4), the PMs reduce the D-axis speed voltage. This results in a reduced D-axis
supply voltage and thus a reduced stator supply voltage, since
2 2
q d s
v v v + = . It is this reduction in
supply voltage that gives the PMA RSM better performance in the field weakening or high speed
region, compared to the RSM. When the RSM reaches its rated speed, there is simply not enough
voltage to force more current into the machine, even though more current might be available. For the
PMA RSM with the same rated speed as the RSM, the reduced supply voltage means that there is now
a surplus amount of voltage that can be used to force more current into the PMA RSM.

The RSM torque equation (2.1.15) is modified to the PMS RSM torque equation as follows:
( ) ( )
) cos(
2
3
2
) 2 sin(
) (
2
3
) (
2
3
2
3
2

s pm s q d
d pm d q q d
d pm q q d em
i p i L L p
i i i p
i i p T
+ =
+ =
=
(2.2.6)

From the equation above it can be seen that the produced torque is increased by the added PMs. It is
important to distinguish between the partial derivatives, which are called instantaneous inductances or
transient inductances, and linearized inductances. For example the partial derivative
'
d
L in equation
(2.2.5) is an instantaneous self inductance, while
d
L in equation (2.2.6) is a linearized self inductance.
Fig. 2.2.1 shows an example of the D-axis flux linkage, instantaneous self inductance
'
d
L and
linearized self inductance
d
L :
) (
d d
i
d
i
'
d
d
d
L
i
=

d
d
d
L
i
=

Fig. 2.2.1. D-axis flux linkage, instantaneous inductance and linearized inductance.

19

To summarize, the equations that describe the PMA RSM are given as follows:
( )
pm q e
q
d
d
d d s d
dt
di
M
dt
di
L i r v + + =
' '
(2.2.1)
d e
d
q
q
q q s q
dt
di
M
dt
di
L i r v + + + =
' '
(2.2.2)
( ) ( )
d pm q q d em
i i p T =
2
3
(2.2.6)
L m eq
m
eq em
T
dt
d
J T + + =

(2.1.13)
p
e
m

= (2.1.14)


2.3. Power factor and inner power factor

The steady state vector diagram below shows the vectors for the RSM (solid lines) and PMA RSM
(red dotted lines).
D-axis
Q-axis
d
I
q
I
s
I
d

'
s

'
q

pm

'
d
V
d
V
s
j
'
s
j
s s
r I
s s
r I
s
V '
s
V

'

'
q
V

Fig. 2.6. Comparative steady state vector diagram.

20
The power factor (PF) is defined as the cosine of the angle between the supply current and voltage
vectors. Recall that ) cos( | || |
s s
i v P = , where is the PF angle and is between 0 and 90 degrees. It
is shown in the steady state vector diagram of Fig. 2.6, that the PMs reduce the PF angle, which means
an increased PF for the PMA RSM. If the angle is zero (PF = 1), then the reactive power associated
with the machine is zero and therefore all the power form the source can be used to do real work, i.e.
more output power. Given the same input power to the RSM and PMA RSM described here, the PMA
RSM will give a larger power output and in that sense is more efficient than the RSM.

The inner power factor (IPF) is defined as the sine of the angle between the current vector and the flux
vector. This is an indication of the magnitude of the produced torque: recall that
) sin( | || | i i T = = ,where is the IPF angle. The IPF angle is between 0 and 90 degrees. It is
shown in the steady state vector diagram Fig. 2.6, that the PMs increase the IPF angle, which means an
increased IPF.


2.4. Parameter values

For the models of the RSM and PMA RSM, many parameters have to be identified. These include the
D-axis and Q-axis flux linkages, self inductances and mutual inductances. To calculate the flux
linkages and inductances, FEA will be used. The stator winding resistance, number of pole pairs,
equivalent inertia and equivalent friction coefficient will be considered next.

The number of pole pairs is chosen in the design of the machine and is known to be (p = 3). The
friction coefficient is ideally zero, but is taken to be (
sec /
1 . 0
rad
Nm
B = ). The equivalent inertia can
be calculated by noting that the shaft and the rotor have the same mass-coefficient [B5, p.76]:

2 2
2 2
2
122 . 2 ] ) 15 . 0 ( ) 66 . 188 [(
2
1
66 . 188 34 . 0 ) 15 . 0 ( 7850
] ) ( [
2
1
m kg J
kg l r p pv M
r M J
length stack rotor
rotor
= =
= = = =
=
(2.4.1)

The inertia is not a very important parameter in the design of the current and speed controllers,
therefore the mass reduction due to flux barriers can be ignored.


21
The stator winding resistance (per phase) is measured practically. The resistance between two phases,
e.g. phase A and phase C is measured and then divided by two to give the per phase stator winding
resistance. It was found to be:
r
s
= 15m (2.4.2)
The stator winding resistance can be calculated using an analytic formula given by [B2, p.91]. This
analytic formula is used in the FEA program [P2] to calculate the per phase stator winding resistance
for a temperature of 30C and the result is given by:
r
s
= 15m (2.4.3)
The FEA program generates files that give information about the machine, e.g. number of stator slots,
number of conductors per slot etc. The FEA program and examples of the generated files can be found
on the CDROM [P2].

The flux linkages can be obtained accurately using the FEA program. This is a software based
solution where the physical dimensions and other information about the PMA RSM are entered into
the FEA program. The program creates a mesh to partition the object into finite elements. For each
element, the vector potential is determined and from all the vector potentials other attributes, like the
flux linkages or torque, can be calculated. From the flux linkages it is possible to calculate self
inductances and mutual inductances.

The FEA program described in [B2] was used to optimally design the RSM [A4][B17]. Then PMs
were included in the design [B5]. The results shown in Fig. 2.4.1 (the next 2 pages) are obtained by
using the same FEA program as in [B5]. From Fig. 2.4.1 it is possible to compare the parameters of
the RSM and PMA RSM. The Matlab code for generating the graphs of Fig. 2.4.1 is given on the
CDROM [F1].

On these graphs, note the demagnetizing effect of the mutual coupling: as the Q-axis current
increases, the D-axis flux linkage decreases for the same value of D-axis current. This means the
mutual inductances must be negative. In fact, where the deviation in the flux linkage is the greatest,
the mutual inductance is at its negative maximum. Comparing the graphs for the RSM and PMA
RSM, it seems that the demagnetizing effect is reduced by adding the permanent magnets. The
demagnetizing effect is however very small.

The most obvious difference between the graphs for the RSM and PMA RSM is the Q-axis flux
linkage. Note that 0 <
q
for A i
q
0 = . Graphs for the self inductances are obtained by numerically
differentiating the data for the flux linkages, and that is why a result at zero current is not shown. The

22
graphs for the mutual inductances are obtained by taking the average change in flux linkage, so that
'
d
M is a function of
d
i , and
'
q
M is a function of
q
i . An alternative method is shown in Appendix E.







Fig. 2.4.1a. FEA results for D-axis flux linkages, self inductances and mutual inductances.
0 100 200 300
-2
-1
0
1
x 10
-4
D-axis mutual inductance for RSM
D
-
a
x
i
s

m
u
t
u
a
l

i
n
d
u
c
t
a
n
c
e

[
H
]
D-axis current [A]
0 100 200 300
-2
-1
0
1
x 10
-4
D-axis mutual inductance for PMA RSM
D
-
a
x
i
s

m
u
t
u
a
l

i
n
d
u
c
t
a
n
c
e

[
H
]
D-axis current [A]
0 100 200 300
0
2
4
6
8
x 10
-3
D-axis self inductance for RSM
D
-
a
x
i
s

s
e
l
f

i
n
d
u
c
t
a
n
c
e

[
H
]
D-axis current [A]
Iq = 0 A
Iq = 240 A
0 100 200 300
0
2
4
6
8
x 10
-3
D-axis self inductance for PMA RSM
D
-
a
x
i
s

s
e
l
f

i
n
d
u
c
t
a
n
c
e

[
H
]
D-axis current [A]
Iq = 0 A
Iq = 240 A
0 100 200 300
0
0.2
0.4
0.6
0.8
1
D-axis flux linkage for RSM
D
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]
D-axis current [A]
Iq = 280 A
Iq = 0 A
0 100 200 300
0
0.2
0.4
0.6
0.8
1
D-axis flux linkage for PMA RSM
D
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]
D-axis current [A]
Iq = 280 A
Iq = 0 A

23






Fig. 2.4.1b. FEA results for Q-axis flux linkages, self inductances and mutual inductances.
0 100 200 300
-2
-1
0
1
x 10
-4
Q-axis mutual inductance for RSM
Q
-
a
x
i
s

m
u
t
u
a
l

i
n
d
u
c
t
a
n
c
e

[
H
]
Q-axis current [A]
0 100 200 300
-2
-1
0
1
x 10
-4
Q-axis mutual inductance for PMA RSM
Q
-
a
x
i
s

m
u
t
u
a
l

i
n
d
u
c
t
a
n
c
e

[
H
]
Q-axis current [A]
0 100 200 300
0
1
2
3
4
x 10
-3
Q-axis self inductance for RSM
Q
-
a
x
i
s

s
e
l
f

i
n
d
u
c
t
a
n
c
e

[
H
]
Q-axis current [A]
Id = 0 A
Id = 240 A
0 100 200 300
0
1
2
3
4
x 10
-3
Q-axis self inductance for PMA RSM
Q
-
a
x
i
s

s
e
l
f

i
n
d
u
c
t
a
n
c
e

[
H
]
Q-axis current [A]
Id = 0 A
Id = 240 A
0 100 200 300
-0.2
-0.1
0
0.1
0.2
0.3
Q-axis flux linkage for RSM
Q
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]
Q-axis current [A]
Id = 0 A
Id = 280 A
0 100 200 300
-0.2
-0.1
0
0.1
0.2
0.3
Q-axis flux linkage for PMA RSM
Q
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]
Q-axis current [A]
Id = 0 A
Id = 280 A

24
2.5. Method of current control for RSM and PMA RSM

In this subsection the author examines two different methods of current control for the RSM and PMA
RSM, namely constant field current control (CFCC) and constant current angle control (CCAC). It is
shown that for the RSM, both CFCC and CCAC are suitable, but for the PMA RSM, only CCAC is
suitable.

The torque equation for the RSM is given by:
( )
d q q d em
i i
p
T =
2
3
(2.5.1)
Equation (2.5.1) can also be written in terms of linearized inductances:

q d q d
d q q q d d
i i L L p
i i L i i L p T
=
=
) ( )
2
3
(
) ( )
2
3
(
(2.5.2)
If the D-axis current is kept constant, then equation (2.5.2) can be expressed as:
q T
i K T = (2.5.3)
From equation (2.5.3), the torque of the RSM can be controlled by increasing or decreasing the Q-axis
current. This method, by which the D-axis current is kept constant and the torque is controlled by the
Q-axis current, is known as CFCC, because the D-axis current is the field current. The D-axis
current is reduced in the flux weakening or high speed region. Note that the right hand side of the
QD0-plane is used, as illustrated in Fig. 2.5.1.
Q-axis
D-axis
q
i s
i
d
i

Fig. 2.5.1. Constant field current control: the area used in the QD0-plane.


25
To explain CCAC, equation (2.5.2) can be expressed as:
2
) 2 sin(
) (
2
3
2

s q d em
i L L p T = (2.5.4)
By keeping the current angle constant in equation (2.5.4), the torque is controlled by increasing and
decreasing the current magnitude. Note that now both the D-axis and Q-axis currents change with
changing current magnitude. For CCAC, the top half of the QD-plane is used. Positive torque is
generated by the current vector in the first quadrant and negative torque is generated by the current
vector in the second quadrant, as illustrated in Fig. 2.5.2.
Q-axis
D-axis
+ d
i
q
i
+ s
i
+

d
i
s
i

Fig. 2.5.2. Constant current angle control: two possible current vectors.

CFCC was compared to CCAC in terms of response time of the currents and also in terms of the
energy efficiency in Chapter 1. It was stated that CFCC gives a faster current response compared to
CCAC, but CFCC is less energy efficient compared to CCAC. It is shown next that both CFCC and
CCAC are suitable for the RSM, but only CCAC is suitable for the PMA RSM.

Using FEA [F2], it is possible to calculate the torque as a function of current angle, by keeping the
current magnitude constant. Here the stator current is kept at the rated value: A i
s
283 = . In Fig.
2.5.3 below, the torque is shown as a function of current angle in two different ways. The graph on the
left is a Cartesian plot, while the graph on the right is a polar plot. On the polar plot the torque
magnitude is given by the radius, and the current angle is the angle taken from the positive horizontal
axis in the anti-clockwise direction. The polar plot is on a complex plane.


26
0 100 200 300 400
-800
-600
-400
-200
0
200
400
600
800
RSM torque versus current angle
T
o
r
q
u
e

[
N
m
]
Current angle [degrees]
500
1000
30
210
60
240
90
270
120
300
150
330
180 0
Q-axis
D-axis

Fig. 2.5.3. Torque versus current angle for RSM.

On the Cartesian plot, it can be seen that the torque is positive for current angles from 0 to 90. This
region corresponds to the first quadrant of the polar plot. The torque locus corresponding to current
angles from 0 to 90 is drawn in the first quadrant, since the torque is positive. The torque is negative
for current angles from 90 to 180. This region corresponds to the second quadrant of the polar plot.
In the complex plane, ( ) + = 180 T T , where 0 > T . Therefore, the torque locus for the
current angles from 90 to 180, finds itself in the forth quadrant, not the second quadrant.

The torque is positive for current angles from 180 to 270. This corresponds to the third quadrant on
the polar plot, and since the torque is positive, the torque locus finds itself in the third quadrant. The
torque is negative for current angles from 270 to 360. This corresponds to the forth quadrant on the
polar plot. The torque locus on the polar plot for current angles from 270 to 360 finds itself in the
second quadrant, because the torque is negative.

From Fig. 2.5.3, note that the maximum positive torque is produced with a current angle of 54 or
(180+54) and maximum negative torque is produced for a current angle of (180-54) or (360-54).
Other current angles therefore produce less torque; that is why CFCC does not always adhere to the
maximum torque per ampere criteria.


27
Since the torque locus of the RSM is symmetrical in the right hand half of the complex plane, CFCC is
a suitable method of current control. Also, since the torque locus of the RSM is symmetrical in the top
half of the complex plane, CCAC is a suitable method of current control.

The torque as a function of current angle for the PMA RSM is shown in Fig. 2.5.4. Again, the current
magnitude is kept constant: A i
s
283 = .
0 100 200 300 400
-1000
-800
-600
-400
-200
0
200
400
600
800
1000
PMA RSM torque versus current angle
T
o
r
q
u
e

[
N
m
]
Current angle [degrees]
500
1000
30
210
60
240
90
270
120
300
150
330
180 0
Q-axis
D-axis

Fig. 2.5.4. Torque versus current angle for PMA RSM.

When mapping the Cartesian plot on the left to the polar plot on the right, it should be noted that a
negative torque is represented on the polar plot by a positive torque with an additional current angle of
180, as explained previously. It can be seen from Fig. 2.5.4 that there is a large positive torque in the
first quadrant and a large negative torque in the second quadrant, followed by a small positive torque
in the third quadrant and a small negative torque in the forth quadrant. After the polarity of the
different parts of the torque locus have been considered, the absolute value of the torque versus the
current angle can be shown as in Fig. 2.5.5.


28

Fig. 2.5.5. Absolute value of torque vs. current angle for PMA RSM.

Since the torque locus of the PMA RSM is not symmetrical in the right hand half of the complex
plane, CFCC is not a suitable method for current control. If CFCC is used for the PMA RSM, the
machine will function well as a motor, but very poorly as a generator. The torque locus of the PMA
RSM is symmetrical in the top half of the complex plane and therefore CCAC is a suitable method of
current control. If CCAC is used for the PMA RSM, the machine functions well as a motor and a
generator.

The torque equation of the PMA RSM also shows that CFCC would not be a suitable method to
control the machine. The torque equation of the PMA RSM is given by:
) (
2
3
d pm d q q d em
i i i p T + = (2.5.5)
Equation (2.5.5) can also be written as:

d pm q d q d em
i p i i L L p T
2
3
) (
2
3
+ = (2.5.6)
If the D-axis current is kept constant, equation (2.5.6) can be expressed as:

pm q T em
K i K T + = (2.5.7)

From equation (2.5.7), there is always a constant positive torque generated by the constant positive D-
axis current and constant PM flux linkage and therefore the negative torque is smaller for CFCC.

By using CCAC, with a current angle of = 54 , the machine produces maximum torque per ampere
and is therefore more efficient. CCAC uses the top half of the QD0-plane and is suitable for both
RSM and PMA RSM. In other words the Q-axis current is always kept positive, while the D-axis
200
400
600
800
1000
30
210
60
240
90
270
120
300
150
330
180 0
Q-axis
D-axis

29
current traverses the entire D-axis. Since the D-axis current has a slower response time compared to
the Q-axis current, CCAC is slower than CFCC. The tradeoff is therefore energy efficiency vs.
response time. It can also be seen from the graphs that the PMA RSM has a maximum torque of 812
Nm, while the RSM only reaches 700 Nm. This verifies the earlier theoretical development.

From this discussion it is evident that only CCAC is an option for current control of the PMA RSM. It
is the objective to have fast dynamic performance from the PMA RSM and this can be achieved using
a load torque observer with compensation current feedback (Chapter 5). First, a machine model is
built in simulation followed by current controllers and a basic speed controller. Only then can a load
torque observer be investigated.




30
Chapter 3 Design methodology for the current and speed controllers

In this chapter, the design methodology for the current and speed controllers for the PMA RSM is
presented. Background theory about digital control systems is given and the theory is applied to
design the controllers. Throughout the chapter various Matlab commands and tools are used to
support the theory.


3.1. Different methods to design a digital controller

Many methods exist to design a controller. A serious issue is that the controller is digital. The
implication is that the output of the digital controller has to be transformed from a discrete signal to a
continuous signal. This is usually accomplished by a sample and hold element at the output of the
digital controller. The hold means that the sampled value of the controller output is held constant
for one sampling period (assuming a zero order hold). The zero order hold (ZOH) introduces a delay
of half the sampling period (see Appendix B).

Design by emulation requires that the sampling time is sufficiently small (the sampling frequency
should be at least ten times the chosen closed loop bandwidth to avoid inaccuracies). The design for
the controller takes place in the continuous-time domain with an S-plane representation, and it is then
transformed to a discrete-time representation in the Z-plane using one of many transformations
(examples are Backward Euler, Tustins method or Matched pole-zero [B9, p.658][B7, p.330]). Note
that there is an approximation when this method is used, because it is assumed that the delay
introduced by the ZOH is negligible.

Another method is to design the controller completely in the Z-plane [B9, p.668] by translating both
the plant and the ZOH to their discrete equivalents (see Appendix B). There is no approximation when
using this method, because the time delay is accounted for.

The discrete equivalent of the ZOH and plant can be transformed to the W-plane using a bilinear
transformation,
w
T
w
T
z
2
1
2
1

+
[B7, p.393] . This transformation will be called the W-transform. The
design of the controller can then take place in the W-plane [B7, p.400]. The W-plane is very much
like the S-plane, except that the primary strip for frequency in the S-plane, which is

31
) ( ,
2 2

j s
j
j
j
s s
= < < is mapped to ) ( , jv w jv = < < in the W-plane. This means
that the frequency scale in the W-plane is distorted. The whole idea of using the W-plane instead of
the S-plane is that the effect of the ZOH is included, and frequency response techniques (Bode plot)
can be used. When the design of the controller is completed in the W-plane, it is transformed back to
the Z-plane using the inverse W-transform
1
1 2
+

z
z
T
w .

A combination of the methods discussed in this subsection will be used to design the current and speed
controllers for the PMA RSM. First the current controllers are discussed and for that the electrical
model of the PMA RSM should be considered.


3.2. Decoupling procedure for the electrical model

The equations for the electrical model of the PMA RSM are coupled: there is a Q-axis term in the D-
axis equation and vice versa. A decoupling procedure is suggested to simplify the design of the
current controllers. The electrical model for the PMA RSM is given in subsection 2.2 and is shown
here again for convenience:
( )
pm q re
q
d
d
d d s d
dt
di
M
dt
di
L i r v + + =
' '
(3.2.1)
d re
d
q
q
q q s q
dt
di
M
dt
di
L i r v + + + =
' '
(3.2.2)

The block diagram that represents equations (3.2.1) and (3.2.2) is shown in Fig. 3.2.1. Note that this
block diagram is drawn in Matlabs Simulink. Inputs and output are indicated by numbered capsules.
Unfortunately the Greek alphabet is not available in Simulink and therefore the electrical rotor speed is
written as Wre, the D-axis mutual inductance as Md, the Q-axis flux linkage as Q_flux etc. Also note
that the instantaneous inductance
'
d
L is written as Ld in the simulation diagram and Ld
d
L , where
d
L is the linearized inductance. Note that Fig. 3.2.1 is just a representation of the equations (3.2.1)
and (3.2.2); the block diagram is not used for simulation.

32

Fig. 3.2.1. Coupled electrical model of the PMA RSM.

From the block diagram in Fig. 3.2.1 it is evident that to control the D-axis and Q-axis currents would
be difficult, since they are coupled. A decoupling procedure is suggested in Fig. 3.2.2. The speed
voltages and mutual inductance terms are subtracted by the controller where they are added in the
machine and vice versa.

Fig. 3.2.2. Decoupling procedure for the electrical model.

This procedure results in decoupled differential equations represented by the block diagram shown in
Fig. 3.2.3.

Fig. 3.2.3. Decoupled electrical model.


33
It is important to make the distinction between the real machine voltages,
d
V and
q
V , and the applied
control voltages, '
d
V and '
q
V , in the current controllers. The relationship is given below:
) ( '
pm q re
q
d d d
dt
di
M V V + = (3.2.3)
d re
d
q q q
dt
di
M V V + + = ' (3.2.4)
It will become clear where this procedure fits in when the design of the current controllers are
complete. The open loop response is investigated next to show why it is necessary to use closed loop
control.


3.3. Open loop current response

Equipped with mathematically decoupled models for the D-axis plant and the Q-axis plant (shown in
Fig. 3.2.3), it is now possible to evaluate the open loop current response of the machine. Generally the
open loop response of any plant is very slow this means that high frequency inputs to the plant are
heavily attenuated and thus cannot serve as significant control signals. It is equivalent to state that the
bandwidth of the plant is small. The bandwidth can easily be determined by looking at a Bode diagram
or a root locus for the plant.

For example, the decoupled D-axis plant is given by:
s d d
d
r s L v
i
+
=
'
1
(3.3.1)
Suitable parameters for the D-axis plant (see subsection 2.4) are = m r
s
15 and mH L
d
2
'
= . If
these values are substituted into equation (3.3.1), the open loop D-axis plant has a single pole at
5 . 7
'
=

d
s
L
r
. The open loop Bode plot and step response for this plant is shown in Fig. 3.3.1.

34

Fig. 3.1.1. Bode plot (top) and step response (bottom) for open loop D-axis plant.

From Fig. 3.1.1 it can be seen that the bandwidth of the open loop D-axis plant is 7 rad/sec (recall that
the pole position is at -7.5). A small bandwidth means a slow response to a step input: 522 . 0 =
s

sec in this case. A slow current response cannot be tolerated in the control of the PMA RSM. It is
therefore necessary to use closed loop current control. The general idea is to design a controller to
generate a control input for the plant. The controller and plant are in a closed loop with feedback from
the plant to the controller.


-10
0
10
20
30
40
M
a
g
n
i
t
u
d
e

(
d
B
)
System: G
Frequency (rad/sec): 7.05
Magnitude (dB): 33.7
10
-1
10
0
10
1
10
2
10
3
-90
-45
0
P
h
a
s
e

(
d
e
g
)
Bode Diagram
Frequency (rad/sec)
Step Response
Time (sec)
A
m
p
l
i
t
u
d
e
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0
10
20
30
40
50
60
70
System: G
Settling Time (sec): 0.522

35
3.4. Closed loop current response

Consider the block diagram in Fig. 3.4.1 which describes a closed-loop control system for the D-axis
part of the decoupled electrical machine equations.

Fig. 3.4.1. D-axis current control system.

For example, let the controller be a constant gain: 10 ) ( = s D . The closed loop transfer function is
then
10
10
) (
+ +
=
s d
cl
r s L
s H . With reference to the chosen parameters for the open loop plant in
subsection 3.3, the closed loop Bode plot and step response is shown in Fig. 3.4.2.

Fig. 3.4.2. Bode plot (top) and step response (bottom) for closed loop D-axis plant and controller.

-30
-25
-20
-15
-10
-5
0
M
a
g
n
it
u
d
e

(
d
B
)
System: Hcl
Frequency (rad/sec): 5.03e+003
Magnitude (dB): -3.08
10
0
10
1
10
2
10
3
10
4
10
5
-90
-45
0
P
h
a
s
e

(
d
e
g
)
Bode Diagram
Frequency (rad/sec)
Step Response
Time (sec)
A
m
p
lit
u
d
e
0 0.2 0.4 0.6 0.8 1 1.2
x 10
-3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
System: Hcl
Settling Time (sec): 0.000781

36
From the closed loop Bode plot it can be seen that the bandwidth of the closed loop system is 5000
rad/sec. Furthermore, the settling time for a step input to the closed loop is ms
s
8 . 0 = . Therefore
the controller and plant in a closed loop can give the required performance needed for current control
of the PMA RSM.

The above analysis was performed in the continuous-time domain. The next subsection addresses
issues relating to the discrete nature of the controller and to stability.


3.5. The discrete nature of the controller

In reality, the controller will not be analogue and will not produce a continuous output signal. The
reference current will also be a discrete signal, which will be generated within a digital signal
processor (DSP). There are many advantages when using digital controllers instead of analogue
controllers. A control system that contains both discrete and continuous parts is called a sampled data
system [B15, p.1]. The sampled data system for the D-axis current controller (discrete) and D-axis
plant (continuous) can be represented by the block diagram in Fig. 3.5.1.
Vd'
Id
(dig)
1
Id
(actual)
1
Ld.s+rs
ADC
DAC D(z)
2
clock
1
Idref
(dig)

Fig. 3.5.1. D-axis sampled data current control system.

In Fig. 3.5.1 there are clearly an analogue section and a digital section. The digital controller is given
by D(z) and the continuous plant model by
s d
r s L
s G
+
=
'
1
) ( . The DSP produces a reference current.
This current is just a number representing the real reference current. The number representing the
reference current is compared to a number representing the measured machine current. The difference
between these numbers is the control input for the digital current controller. The output of the digital
current controller is a number that represents the control voltage for the machine.

The function of the analogue-to-digital converter (ADC) and the digital-to-analogue converter (DAC)
is to translate between the numbers representing the currents or voltages and the actual currents and
voltages.

37

The ADC has a continuous input from the current sensor. This input is sampled and the sampled value
is stored in a buffer with a finite number of bits. The ADC is thus a sampler followed by a quantizer
and a coder [B6, p.22]. The sampled value is not equal to the real value and the error depends on the
number of bits used to store the number. Assuming that this error is very small and that the time it
takes for the ADC to read the value into the buffer is much smaller that the sampling period, the ADC
can be modeled as in Fig. 3.5.2.

Fig. 3.5.2. ADC model.

In the model for the ADC above, Ks is the conversion constant between the actual and digital values.
For a floating point DSP 1 =
s
K , because the DSP has the ability to store real or floating point
numbers. Generally for a fixed point DSP 1
s
K , because a fixed point DSP only has the ability to
store integer numbers. The model in Fig. 3.5.2 should be interpreted as follows: at every sampling
instant (which occurs after a sampling period has passed) the switch moves to the actual current just
long enough for the digital current to take on a value that is related to the actual current by the
conversion constant. This value is held in memory until the next sampling instant. The switch then
returns to the ground state and waits for the next sampling instant.

The DAC converts digital numbers to a continuous signal. A zero order hold (ZOH) function is
necessary to accomplish this. The DAC can be modeled as in Fig. 3.5.3. Again, the switch is at a
ground state between sampling instances and waits for the sampling period to pass. At the sampling
instant, the switch moves to the other state just long enough for the zero-order hold to take on a scaled
version of the digital voltage. The ZOH keeps this voltage for the rest of the sampling period until the
next value is obtained.

Fig. 3.5.3. DAC model.

38

The ADC and DAC conversion factors need not be the same, but they are assumed to be equal in the
development that follows. Omitting the switches and clock, the sampled data system is shown in Fig.
3.5.4.

Fig. 3.5.4. D-axis sampled data current control system with modeled ADC and DAC.

In Fig. 3.5.4, the reference current and feedback current are being multiplied by the same number and
then compared. Equivalently, they could be compared and then multiplied by the constant. In this
manner, the constant cancels with the reciprocal term following D(z). The control system is thus
simplified and is shown in Fig. 3.5.5.

Fig. 3.5.5. Simplified D-axis current control system.

The ZOH and the machine model can be combined and transformed to the Z-plane using equation
(3.5.1) [B7, p.333, p.400]. This concept is explained in Appendix B. In equation (3.5.1), Z[ ] is the Z-
transform. A control system that is completely in the Z-plane follows, as shown in Fig. 3.5.6.
(

=

s
s G
z z G
) (
) 1 ( ) (
1
(3.5.1)

Fig. 3.5.6. D-axis current control system in the Z-plane.

The derivation above leads to a control system that is completely in the Z-plane. Therefore the D-axis
current controller can be designed in the Z-plane. This is to be preferred above the design in the
continuous domain, since the delay introduced by the ZOH is included in the design. Recall that the
entire left-half-plane (LHP) of the S-plane is mapped into the unit circle on the Z-plane [B15, p.158].

39
Stable poles from the LHP of the S-plane are all crowded near 1 + = z on the real axis in the Z-plane if
the sampling time is sufficiently small.

It is the aim to design a stable current controller. It would be possible to design a current controller
using root locus in the Z-plane. Stability is then indicated by closed loop poles being inside the unit
circle. This however just shows that the poles are stable or unstable but it does not show how stable
they are. If the design of the current controller takes place in the W-plane, then the effect of the ZOH
is accounted for and frequency response techniques can be used. Design in the W-plane using a Bode
plot enables the designer to specify a gain margin, which is an indication of how stable the closed loop
poles are.


3.6. Design in the W-plane using a Bode plot and gain margin specification

To be able to design a stable current controller, it is first necessary to understand what instability
means. With a sampling frequency of
s
, input signals with a frequency of less than
2
s

(the Nyquist
frequency) can be sampled. Since the plant introduces a 90 phase delay (due to the single pole) and
the ZOH introduces another 90 phase delay at
s
/2 (see Appendix B) there will be a 180 phase delay
at
s
/2. If the loop gain (controller gain multiplied by plant gain) is 1, or 0 dB, at
s
/2 then instability
will occur, because the closed loop response is then given by
0
1
180 1 1
180 1
=
+

= infinity.

To ensure a stable controller, the loop gain is chosen to be 10 dB at
s
/2. That gives a gain margin of
10 dB. This restriction limits the bandwidth of the closed loop system and therefore gives a lower
limit for settling time to a step input.

The design of the current controller in the W-plane will proceed as follows:
Start with the transfer function for the open loop plant in the continuous domain.
Transform the transfer function to a pulse transfer function (discrete domain transfer function)
and include the effect of the ZOH.
Use the W-transform to transform the pulse transfer function back to the continuous domain.
Now the effect of the ZOH is accounted for, but the frequency is distorted so that
2
s

=
corresponds to = w .

40
The controller is chosen to be only a proportional (P) controller and not a proportional integral
(PI) controller. The controller gain is computed so that the loop gain is -10 dB at
= w rad/sec.
The designed controller is transformed back to the discrete domain using the inverse W-
transform, but since it is only a constant, no transformation is needed.

The continuous plant model is given by:
s d
ds
r s L
s G
+
=
1
) ( (3.6.1)
To transform the plant from the S-plane representation to the Z-plane representation, the Z-transform
is used. The Z-transform however, requires the sampling time. A sampling frequency of 5000 =
s
f
Hz (choice of frequency is subject to the maximum switching frequency of the inverter and the
operational speed of the DSP), means a sampling time of T = 200 s. With this information we can
get the equivalent pulse (discrete) transfer function for the plant and include the effect of the ZOH
using equation (3.5.1):
) (
1
1 1
) 1 (
) (
) 1 ( ) (
'
'
'
1
1
d
s
d
s
L
T r
s
L
T r
d s
dz
e z r
e
sL r s
z
s
s Gds
z z G

=
)
`

+
=
)
`

=
(3.6.2)
The W-transform is now used to translate equation (3.6.2) back to the continuous domain. The W-
transform is a bilinear transformation, and in fact a trapezium approximation to integration (this is
shown in subsection 3.11). It is therefore an accurate mapping between the discrete and continuous
domains. The W-transform is given by:

w
T
w
T
z
2
1
2
1

+
= (3.6.3)


41
Substituting equation (3.6.3) into equation (3.6.2), the transfer function for the open loop plant,
including the effect of the ZOH, in the continuous domain is obtained:

|
|
|
|

\
|

'
'
2
1
2
1
1
) (
d
s
d
s
L
T r
s
L
T r
dw
e
w
T
w
T
r
e
w G (3.6.4)

Equation (3.6.4) might as well be expressed in terms of the continuous variable s instead of using
w. One should keep in mind though that the frequency axis is distorted.
To evaluate equation (3.6.4) at the Nyquist frequency
2
s

= , substitute = w :
) 1 (
1
2
1
2
1
1
) (
'
'
'
'
d
s
d
s
d
s
d
s
L
T r
s
L
T r
L
T r
s
L
T r
dw
e r
e
e
T
T
r
e
G

=
|
|
|
|

\
|

= (3.6.5)

The controller gain K
d
, should be chosen so that loop gain at the Nyquist frequency is -10 dB:
10
) 1 (
1
log 20
'
'
10
=
|
|
|
|

\
|

d
s
d
s
L
T r
s
L
T r
d
e r
e
K
'
'
'
'
1
1
10
1
) 1 (
10
20
10
20
10
d
s
d
s
d
s
d
s
L
T r
L
T r
s
L
T r
L
T r
s
d
e
e
r
e
e r
K

+
=

= (3.6.6)

|
|

\
|

=
|
|

\
|
=

'
20
10
'
20
10
2
tanh
10
2
coth 10
d
s
s
d
s
s d
L
T r
r
L
T r
r K (3.6.7)

Evaluating equations (3.6.1) through to (3.6.7) for r
s
= 15 m,
'
d
L = 4 mH and T = 200 s gives:
015 . 0 004 . 0
1 1
) (
'
+
=
+
=
s r s L
s G
s d
ds


42
99925 . 0
04998 . 0
) (
1
) (
'
'

z
e z r
e
z G
d
s
d
s
L
T r
s
L
T r
dz

w
w
w
w
e
w
T
w
T
r
e
w G
d
s
d
s
L
T r
s
L
T r
dw
+

+
=
|
|
|
|

\
|

75 . 3
) 10000 (
40
1
99925 . 0
0001 . 0 1
0001 . 0 1
04998 . 0
2
1
2
1
1
) (
'
'

dB G G G
dB
dw dw dw
32
40
1
log 20 ) ( , 180 ) ( ,
40
1
) ( =

= =

=
65 . 12
1
40
10
20
10
=

=

d
K

These results can be verified using Matlab. The Matlab code to verify the answers above and the
relevant Bode diagrams follow:
Create the continuous transfer function:
Gs = tf([1],[4e-3 15e-3])
Transfer function:
1
---------------
0.004 s + 0.015
Discretize the continuous function and include the effect of the ZOH:
Gz = c2d(Gs,1/5000,zoh)
Transfer function:
0.04998
----------
z - 0.9993
Sampling time: 0.0002
Transform back to the continuous domain using the bilinear (also called Tustins method [B9, p.659])
approximation:
Gw = d2c(Gz,'tustin')
Transfer function:
-0.025 s + 250
--------------
s + 3.75

43
The Bode plot for the continuous plant in the W-plane is shown next. Notice that the gain at the
Nyquist frequency (this is an arbitrary frequency, but it should approach infinity) is -32 dB and the
phase is 180.
bode(Gw)
-40
-20
0
20
40
M
a
g
n
i
t
u
d
e

(
d
B
)
System: Gw
Frequency (rad/sec): 6.7e+005
Magnitude (dB): -32
10
-1
10
0
10
1
10
2
10
3
10
4
10
5
10
6
180
225
270
315
360
P
h
a
s
e

(
d
e
g
)
Bode Diagram
Frequency (rad/sec)
Fig. 3.6.1. Bode plot of continuous D-axis plant in the W-plane.

The plant is now scaled by the controller constant 65 . 12 =
d
K and the Bode plot is shown again.
Notice that the gain at the Nyquist frequency is now -10 dB.
loop = 12.65*Gw; bode(loop)
-20
0
20
40
60
M
a
g
n
i
t
u
d
e

(
d
B
)
System: loop
Frequency (rad/sec): 5.53e+005
Magnitude (dB): -10
10
-1
10
0
10
1
10
2
10
3
10
4
10
5
10
6
180
225
270
315
360
P
h
a
s
e

(
d
e
g
)
Bode Diagram
Frequency (rad/sec)

Fig. 3.6.2. Bode plot of continuous D-axis plant and controller in the W-plane.


44
Using Matlab, it was possible to verify the equations and specifically the formula for the controller
gain. To summarize, the controller gain that guarantees stability with a gain margin of 10dB for the D-
axis current loop is:
|
|

\
|

=

'
20
10
2
tanh
10
d
s
s
d
L
T r
r
K (3.6.7)
By analogy, the controller gain that guarantees stability with a gain margin of 10dB for the Q-axis
current loop is:
|
|

\
|

=

'
20
10
2
tanh
10
q
s
s
q
L
T r
r
K (3.6.8)
Now, it is known that the D-axis self inductance changes with changing D-axis current. The Q-axis
self inductance however, is nearly constant (see subsection 2.4). The gain of the Q-axis current
controller can be calculated once and implemented. For the implementation of the D-axis current
controller however, a lookup table (LUT) with D-axis current as input and D-axis self inductance as
output is necessary. At every sampling instant, the new value for the D-axis self inductance is
obtained from the LUT and the new value for the controller gain is computed. In this manner, stability
is always ensured.

The formula for the D-axis current controller is computationally intensive and is not suitable for
implementation. A mathematically less intensive formula is derived next.


3.7. Revised current controllers

The current controllers are proportional controllers and the formulas for the proportional constants are
given by equations (3.6.7) and (3.6.8) for the D-axis and Q-axis current loops respectively. It can be
assumed that the Q-axis self inductance is constant and the controller gain for the Q-axis current loop
need only be calculated once. The D-axis self inductance can have values ranging from 1 mH to 8 mH
(see subsection 2.4) and it is therefore necessary to calculate the D-axis current controller gain at every
sampling instant.

The computational intensity of the formula given by equation (3.6.7) can be reduced: regressing to
equation (3.6.6) from which equation (3.6.7) is derived and using a series expansion for the
exponential terms, an approximation for equation (3.6.6) can be derived:

45
|
|

\
|
=
|
|

\
|

|
|

\
|
+
|
|

\
|
+
|
|

\
|
+
|
|

\
|
+ +
=

+
=

'
'
20
10
'
'
20
10
2
' '
2
' '
20
10
20
10
2 10
2
10
... 1 1
... 1 1
10
1
1
10
'
'
d
s d
d
s
d
s
s d
d
s
d
s
d
s
d
s
s
L
T r
L
T r
s d
L
T r
T
L
L
T r
L
T r
r K
L
T r
L
T r
L
T r
L
T r
r
e
e
r K
d
s
d
s

2 10
'
20
10


T
L
K
d
d
(3.7.1)

The derivation of equation (3.7.1) assumes that the sampling time is sufficiently small so that any
terms with
2
T or greater powers are negligible. If the sampling frequency is 5000 Hz, then equation
(3.7.1) can be evaluated:

' '
20
10
278 . 3162 2 5000 10
d d d
L L K = =

(3.7.2)

The gain for the Q-axis current controller is now evaluated. From Fig. 2.4.1, the Q-axis self
inductance is typically 0.75 mH. Evaluating equation (3.6.8) for mH L
q
75 . 0
'
= , = m r
s
15 and
5000
1
= T sec, the Q-axis current controller gain is given by:
372 . 2 =
q
K (3.7.3)

With the stable current controllers in place, the complete current control system for the PMA RSM is
explained next.


3.8. Implementation of the current controllers in simulation

The complete block diagram for the current controllers is shown in Fig. 3.8.1. Note that although the
decoupling procedure in subsection 3.2 suggests that the terms with the mutual inductances be
included, it is neglected here, because the values of mutual inductances are very small. The lookup
tables that are used to obtain the D-axis flux linkage, Q-axis flux linkage and D-axis self inductance
are shown in Fig. 3.8.2. Lookup tables are explained thoroughly in Appendix E.

46

2
Vq control
1
Vd control
range check
Qflux
2.372
Q conrtoller
Dind
Dflux
3162.3
D conrtoller
5
Iq ref
4
Id ref
3
we measured
2
id measured
1
iq measured

Fig. 3.8.1. Current controllers with decoupling.


Fig. 3.8.2. Graphs representing lookup tables for current controllers.
-300 -200 -100 0 100 200 300
0
2
4
6
8
x 10
-3
D-axis self inductance
D
-
a
x
i
s

s
e
l
f

i
n
d
u
c
t
a
n
c
e

[
H
]
D-axis current [A peak]
-400 -200 0 200 400
-2
-1
0
1
2
D-axis flux linkage for Iq = 226 A peak
D
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]
D-axis current [A peak]
-400 -200 0 200 400
-1.5
-1
-0.5
0
0.5
1
1.5
Q-axis flux linkage for Id = 170 A peak
Q
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]
Q-axis current [A] peak

47
In Fig. 3.8.1, the lookup table for the D-axis self inductance takes the D-axis current as input. The
current is prevented from going out of range 300 300 < <
index
I A. This is also important in the
practical implementation. The block diagram in Fig. 3.8.1 is now encapsulated into a subsystem called
Current Control. An accurate electrical model for the PMA RSM is needed to test the current
controllers. This model uses the lookup tables represented in Fig. 3.8.2 with additional lookup tables
for the mutual inductances and Q-axis self inductance. The machine model is shown in Fig. 3.8.3 and
the additional lookup tables in Fig. 3.8.4.

2
Iq
1
Id
Iq Qind
Id Dind
Id
Iq
Dmind
Qf lux
Df lux
Qmind
LUT
1/s
1/s
-K-
Rs
du/dt
du/dt
3
Vq control
2
Vd control
1
we measured

Fig. 3.8.3. Accurate electrical model of the PMA RSM.


48

Fig. 3.8.4. Additional lookup tables for the accurate electrical model.

The electrical machine model is now encapsulated into a subsystem called Electrical model. The
Current control and Electrical model subsystems, with current references and output blocks are shown
in Fig. 3.8.5 [S1]. The measured speed is set to zero: this can be viewed as a locked rotor test. Note
that the current output of the Electrical model is followed by a ZOH element. This represents the
analogue-to-digital conversion.
ZOH
ZOH
Iq Ref
Iq
Id Ref
Id
we measured
Vd control
Vq control
Id
Iq
Electrical Model
iq measured
id measured
we measured
Id ref
Iq ref
Vd control
Vq control
Current Control
0
Constant

Fig. 3.8.5. Current controller test setup.

-300 -200 -100 0 100 200 300
0
2
4
6
8
x 10
-3
Q-axis self inductance
Q
-
a
x
i
s

s
e
l
f

i
n
d
u
c
t
a
n
c
e

[
H
]
Q-axis current [A] peak
-300 -200 -100 0 100 200 300
-2
0
2
4
x 10
-4
Q-axis mutual inductance
Q
-
a
x
i
s

m
u
t
u
a
l

i
n
d
u
c
t
a
n
c
e

[
H
]
D-axis current [A peak]
-300 -200 -100 0 100 200 300
-2
0
2
4
x 10
-4
D-axis mutual inductance
D
-
a
x
i
s

m
u
t
u
a
l

i
n
d
u
c
t
a
n
c
e

[
H
]
Q-axis current [A] peak

49
As a first test, the Q-axis current reference is set to zero and the D-axis current reference to 100 A at t
= 1 sec. The second test is to set the D-axis reference current to zero and the Q-axis current reference
to 100 A at t = 1 sec. The simulation results are shown in Fig. 3.8.6.


Fig. 3.8.6. D-axis current response (left) and Q-axis current response(right).

From Fig. 3.8.6 it can be seen that both currents respond to their reference commands within 2 ms.
This simulation however is not very realistic, because the inverter is not included. In reality, the D-
axis and Q-axis voltages are transformed to the ABC reference frame using the Park transformation
and given as voltage references to the inverter. The inverter synthesizes the three-phase voltages and
applies it to the PMA RSM. The model for the PMA RSM needs to be modified to accept three-phase
voltages and deliver three-phase currents. The inverter model and the necessary modifications to the
Electrical model are discussed next.


3.9. The inverter model and the modified electrical machine model

First, the electrical model is modified to accept three-phase voltages and deliver three phase currents.
This is easily accomplished using the Park and inverse Park transformations. The Park transformation
block, inverse Park transformation block and the modified electrical machine model are shown in Fig.
3.9.1, Fig. 3.9.2 and Fig. 3.9.3 respectively. Note that in these figures, theta is the electrical rotor
position and it is obtained by integrating the measured electrical speed.

50
2
Q
1
D
sin
sin
sin
cos
cos
cos
2/3
2/3
2*pi/3
4
theta
3
C
2
B
1
A

Fig. 3.9.1. Park transformation simulation block.


3
C
2
B
1
A
sin
cos
sin
cos
sin
cos
2*pi/3
2*pi/3
3
theta
2
Q
1
D

Fig. 3.9.2. Inverse Park transformation simulation block.

51
3
Ic
2
Ib
1
Ia
Iq Qind
Id Dind
A
B
C
theta
D
Q
Park
Id
Iq
Dmind
Qflux
Dflux
Qmind
LUT
D
Q
theta
A
B
C
Inv Park
1/s
1/s
1/s
-K-
Rs
du/dt
du/dt
4
we measured
3
Vc control
2
Vb control
1
Va control

Fig. 3.9.3. Modified electrical model for PMA RSM.

The model for the inverter is shown in Fig. 3.9.4. The ABC voltage references are compared to a
triangular voltage. For example, if Va is greater than Vtri, then VAo is set to +Vdc/2 and if Va is less
than Vtri, then VAo is set to Vdc/2, where Vdc is the DC bus voltage. The Dead time block is
shown in Fig. 3.9.5.
VAo
VBo
VCo
3
VCo
2
VBo
1
VAo
Vtri scope
Vtri
Relay2
Relay1
Relay
In Out
Dead time2
In Out
Dead time1
In Out
Dead time
3
Vc
2
Vb
1
Va

Fig. 3.9.4. Model for the inverter.

In the dead time block, the integrator is reset for a rising edge or a falling edge. The output of the
integrator is time, and it is also the input to a switch. If the middle input to the switch is greater or
equal to a limit (the dead time), then the output is equal to the top input to the switch, otherwise the
output is equal to the bottom input to the switch.

52
1
Out
Switch
1
s
Integrator
0
1
1
In

Fig. 3.9.5. Model for the dead time

The dead time is set to s T
dead
10 = (see subsection 4.5) and the DC bus voltage is set to 900 V.
Here it is assumed that the DC bus voltage is constant, but in reality the DC bus voltage varies with
load: if the PMA RSM is heavily loaded, the DC bus voltage is equal to 890 V, and when it is lightly
loaded it is equal to 930 V.

The current controller test setup of Fig. 3.8.5 is modified to include the model for the inverter, the
modified electrical model of the machine and the Park transformations where needed [S2]. This is
shown in Fig. 3.9.6. Note that the ABC reference voltages at the output of the current controller are
limited to a magnitude of 90% of the DC bus voltage. That means that the modulation index is limited
to 0.9. The locked rotor tests are performed again: first the D-axis current reference is set to 100 A
and the Q-axis current reference to zero; secondly the Q-axis current reference is set to 100 A and the
D-axis current reference to zero. The simulation results are shown in Fig. 3.9.7.

D/A
mod
limit
A/D
A
B
C
theta
D
Q
Park
Iq Ref
Iq
Va
Vb
Vc
VAo
VBo
VCo
Inverter
D
Q
theta
A
B
C
Inv Park
Id Ref
Id
Va control
Vb control
Vc control
we measured
Ia
Ib
Ic
Electrical Model
iq measured
id measured
we measured
Id ref
Iq ref
Vd control
Vq control
Current Control
0 Constant

Fig. 3.9.6. Current control test setup with inverter and Park transformations.

53

Fig. 3.9.7. D-axis current response(left) and Q-axis current response(right).

From Fig. 3.9.7, the current response times are still ms
s
2 < even with inverter dead time included.
These results are confirmed practically and are given in Chapter 4. With the current controllers in
place, the speed controller will be considered next.


3.10. Design of the speed controller

The speed controller should ensure that the rotational speed of the machine can follow a constant
speed reference with zero steady state error. This is easily accomplished by choosing a PI controller.
To design the speed controller, it is first necessary to consider the mechanical part of the PMA RSM.
The differential equation that relates the produced electro-mechanical torque to the mechanical
rotational speed and the load torque is given by:
L rm eq
rm
eq em
T
dt
d
J T + + =

(3.10.1)
The mechanical speed is related to the electrical speed by the number of pole pairs:
p
re
rm

= (3.10.2)
The produced electro-mechanical torque for the PMA RSM is given by:
) (
2
3
d pm d q q d em
i i i p T + = (3.10.3)
The output of the speed controller is the stator current reference. For constant current angle control
(CCAC), the current angle is kept constant at an angle which gives the maximum torque per current.
From Fig. 2.5.4, the value for this angle is given as = 54
maxtorque
. The magnitude of the stator
current changes as the speed controller requires it to: when the speed reference is set to a new value or
the load torque changes, there is a speed error. The speed error is the input to the speed PI controller

54
and the output is the stator reference current. Therefore, a large speed error will result in a large
reference current.

It is necessary to limit the reference current, and the limit should be the rated current of the machine.
For the PMA RSM with 110 =
rated
P kW, the rated stator current is A I
rated s
283 = . The current limit
cannot be less than the rated current for the machine, because then the machine cannot produce the
rated torque, Nm T
rated em
812 = in this case (see Fig. 2.5.4 or Fig. 2.5.5). From equation (3.10.1), the
current limit also determines the acceleration of the machine:

eq
L rm eq rated em
rm
J
T T
dt
d

=

(3.10.4)

Neglecting the friction coefficient
eq
in equation (3.10.4), taking the load torque
L
T as zero and the
inertia
2
2122 . 2 m kg J
eq
= from subsection 2.4, the maximum acceleration for a constant rated
torque output of Nm T
rated em
812 = is given by:

2
sec
65 . 382
122 . 2
812 rad
dt
d
rm
= =

(3.10.5)

This means, if the rated speed reference of
sec
66 . 125
sec 60
min 2
min
1200
rad
rev
rad rev
m
= =

is given (as
a step) and the speed controller maintains the maximum current demand, the machine will reach this
speed within sec 33 . 0
65 . 382
66 . 125
= . The PI controller will of course decrease the current demand as the
speed reaches the reference speed, but the point is that even though the current is limited to the rated
value, very high acceleration is possible. It is suggested that an acceleration limiter (slew rate limiter)
be placed after the speed reference.

Due to the large inertia of the machine, it does not make sense to design the PI controller to give a very
small settling time (the time for the speed to reach the reference value), ms
s
10 = for example. In
that case, the speed controller will demand an unreasonably high stator current, which will be limited
anyway. To simplify the design of the PI controller, the response time of the speed loop should be
chosen to be much larger than the response time of the current loop, so that it can be assumed that the
current reaches its reference value instantaneously. In that case, the produced torque can be described
as follows:

s s T em
i i K T = ) ( (3.10.6)

55
In equation (3.10.6), the torque coefficient
T
K , is a function of the stator current
s
i . Using equation
(3.10.1) and (3.10.6), the block diagram for the speed loop is shown in Fig. 3.10.1.

Is ref
Tem
limit to
rated current
We ref
We
Torque load
Kt
p
Jeq.s+Beq
G(s)
PI
D(s) Accel
limiter

Fig. 3.10.1. Speed loop with PI controller.

A result from finite element analysis (FEA) that relates the stator current to the produced electro-
mechanical torque for a constant current angle of 54 is shown in Fig. 3.10.2.

Fig. 3.10.2. Torque vs. stator current (left) and torque coefficient vs. stator current (right) both for
constant current angle of 54.

From Fig. 3.10.2 it can be seen that the torque coefficient varies in a range
A
Nm
K
T
5 . 1 0 < < , but
since the settling time for the speed loop is not crucial, the torque coefficient can be taken as
A
Nm
K
T
1 = and settling time can be chosen as sec 2 =
s
. Design by emulation can be used to
design the PI controller, since the delay introduced by the ZOH is negligible.

Create the continuous plant:
p = 3;
Jeq = 2.122;
Beq = 0.1;
G = tf([p],[Jeq Beq])
Transfer function:
3
-------------
2.122 s + 0.1
0 50 100 150 200 250 300
100
200
300
400
500
600
700
800
900
1000
Torque vs. Stator current
Stator current [A peak]
T
o
r
q
u
e

[
N
m
]
0 50 100 150 200 250 300
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Kt vs. Stator current
Stator current [A peak]
T
o
r
q
u
e

c
o
n
s
t
a
n
t

56
Import this plant into the Matlabs root locus tool and design the speed controller to give a settling
time of sec 2 =
s
to a step input. The root locus tool and the step response are shown in Fig. 3.10.3.

Fig. 3.10.3. Root locus tool (left) and step response (right) for speed loop.

Export the controller to the workspace. This is given by:
C
Zero/pole/gain:
3.0668 (s+1.522)
----------------
s

The designed PI controller therefore has a proportional constant of 3 =
p
K and a integral constant of
5 . 4 =
i
K . The last step of design by emulation requires that the continuous controller be discretized.
The bilinear transformation is preferred, since it gives an accurate approximation to integration. There
is however one obstacle: due to the saturation of the controller output (current limit), integrator
windup can result. This is discussed in the next subsection.


3.11. Implementation of the speed controller in simulation

For a large speed error, the integrator branch of the speed PI controller integrates the error, producing a
signal which contributes to the reference current. The reference current is limited however and it may
take a long time for the speed to reach the reference speed (much longer than the settling time
designed for). If the integrator is allowed to integrate the error during saturation of the current, it will
reach a very large value. The PI controller therefore demands a greater and greater reference current
and it cannot be given. When the speed finally reaches the reference speed, the output of the PI
controller is still very high. The speed then surpasses the reference speed (assuming a positive speed

57
reference), making the speed error negative. It then takes a long time for the integrator to reduce its
output value. This is known as integrator windup.

A simple solution is to limit the output of the integrator. The limit should be large enough to cater for
the steady state speed error. This solution results in a trial-and-error implementation. Another
solution is to stop the integration whenever the current saturates and to allow integration when the
current is not saturated. The second approach is followed and implementation for the simulation is
shown in Fig. 3.11.5.

The bilinear approximation to integration is used to materialize the discrete integrator. The bilinear
transformation gives a trapezium approximation to integration: this is explained briefly. The speed
error is the term that is being integrated, as shown in Fig. 3.11.1.
time
err
) 1 int( k
s
T
) 1 ( k err
) (k err

Fig. 3.11.1. Bilinear approximation to integration.

From the figure above, which is in the discrete time domain:
( )
( ) ) ( ) 1 (
2
) 1 int(
) ( ) 1 (
2
) ( ) 1 int( ) int(
k err k err
T
k
k err k err
T
T k err k k
s
s
s
+ + =
+ + =
(3.11.1)
Equation (3.11.1) can be easily transformed into the Z-plane representation:
( )
1
1
2 1
1
2 ) (
) int(
) ( ) (
2
) int( ) int(
1
1
1 1

+
=

+
=
+ + =

z
z T
z
z T
z err
z
z err z z err
T
z z z
s s
s
(3.11.2)


58
The S-plane representation of the transfer function for the integrator is:

s s err
s 1
) (
) int(
= (3.11.3)
From equation (3.11.3) and (3.11.4), the transformation is given by:
1
1 2
+

z
z
T
s
s
(3.11.4)

To implement the discrete integrator of equation (3.11.2), two possibilities exist namely Direct Form I
and Direct Form II . The Direct Form I implementation is shown in Fig. 3.11.2.
1
int
z
1
z
1
Ts/2 1
err

Fig. 3.11.2. Direct Form I implementation of the discrete integrator.

From the block diagram in Fig. 3.11.2, there are two actions performed in series and they may be
executed in reversed order, as shown in Fig. 3.11.3.
1
int
z
1
z
1
Ts/2 1
err

Fig. 3.11.3. Direct Form I in reversed order.

Since there are two delay elements that give the same result, they may be combined. This leads to the
Direct Form II implementation, shown in Fig. 3.11.4.
temp
1
int
z
1
Ts/2 1
err

Fig. 3.11.4. Direct Form II implementation of the bilinear integrator.

The block diagram in Fig. 3.11.4 is encapsulated into a subsystem called Disc Int and is used in the
block diagram for the speed PI controller, shown in Fig. 3.11.5.

59
Is ref
2
Iq ref
1
Id ref
-K-
sin(54)
-K-
cos(54)
Switch
3
Kp
4.5
Ki
Is peak limit
err int
Disc Int
0
|u|
|u|
Abs
2
We ref
1
We measured

Fig. 3.11.5. Speed PI controller with anti-integrator-windup and constant current angle reference
output.

In Fig. 3.11.5, the switch monitors the PI controllers output: if the value is less than the rated current,
the input to the integrator is the speed error, otherwise the input to the integrator is zero. The D-axis
and Q-axis current references are obtained by keeping the current angle constant at 54. The Q-axis
current reference is kept positive so that the top half of the QD-plane is used, as depicted in Fig. 2.5.2.

A simulation block for the mechanical plant is still needed and for this equation (3.10.3) together with
the lookup tables for the flux linkages from subsection 3.8 are used. Note that the flux linkage due to
the permanent magnets is already included in the lookup table for the Q-axis flux linkage. The
mechanical model is shown in Fig. 3.11.6.

1
We
p
Jeq.s+Beq
Iq
Id
Df lux
Qf lux
3*p/2
3
Iq
2
Id
1
T load

Fig. 3.11.6. Mechanical part of the PMA RSM

The speed controller is now implemented in simulation to test the performance. The block diagram of
Fig. 3.11.5 is encapsulated into a subsystem called speed PI and the block diagram of Fig. 3.11.6 is
encapsulated into a subsystem called Mechanical Model. The complete control system is shown in
Fig. 3.11.7. The inverter and Park transformations are omitted first, but are included in subsection
3.13.

60
ZOH
We ref
We
Torque load
We ref
We measured
Id ref
Iq ref
Speed PI
T load
Id
Iq
We
Mechanical Model
Iq
Id
Vd control
Vq control
we measured
Id
Iq
Electrical Model
iq measured
id measured
Id ref
Iq ref
we measured
Vd control
Vq control
Current Control

Fig. 3.11.7. Simulation block diagram of speed and current controllers omitting the Park
transformations and inverter.


3.12. Simulation results for the speed controller

For the first simulation using the system in Fig. 3.11.7 [S3], the acceleration limiter is omitted, an
electrical speed reference of 125 rad/sec is given at t = 1 sec and the load torque is zero. The resulting
speed, D-axis current and Q-axis current responses are shown in Fig. 3.12.1.


Fig. 3.12.1. Electrical speed (left), Q-axis current (middle) and D-axis current (right).

From Fig. 3.12.1, the steady state value for the speed is 125 rad/sec and the settling time is 3 seconds
(not the 2 seconds for which it was designed). The steady state value of the Q-axis current is 5.1 A
and for the D-axis current it is 3.85 A, which means a steady state value for the current angle of 53.


61
For the next test, the acceleration limiter is included and set to a value such that the step reference
speed is reached after 5 seconds. A load torque of 700 Nm is then applied at t = 6 sec. The resulting
speed, D-axis current and Q-axis current responses are shown in Fig. 3.12.2.


Fig. 3.12.2. Electrical speed (left), Q-axis current (middle) and D-axis current (right).

In Fig. 3.12.2, from the speed response it can be seen that the load torque is compensated for in about
3 seconds (the same amount of time that the speed reaches the reference value with no acceleration
limiter). The steady state value of the Q-axis current is 200 A and for the D-axis current it is 146 A,
which means a steady state value for the current angle of 54.

To be absolutely sure that the system is stable the inverter and the Park transformations can be
included in the simulation. The Electrical Model subsystem and Mechanical subsystem are combined
into one subsystem called PMA RSM. A simulation block that converts the electrical speed into
position that is between and is shown in Fig. 3.12.3 and the subsystem into which it is
encapsulated is called Speed / Pos. The complete control system is shown in Fig. 3.12.4. The
experiment for Fig. 3.12.2 is now repeated [S4] with the complete control system. The results are
shown in Fig. 3.12.5.

1
Theta
sin
acos cos
Sign
1
s
1
We

Fig. 3.12.3. Simulation block diagram for converting electrical speed into electrical position.


62
D/A
mod
limit
A/D
theta
We ref
Torque load
We ref
We measured
Id ref
Iq ref
Speed PI
We Theta
Speed / Pos
A
B
C
theta
D
Q
Park
T load
Va control
Vb control
Vc control
we measured
Ia
Ib
Ic
We
PMA RSM
Iq
Va
Vb
Vc
VAo
VBo
VCo
Inverter
D
Q
theta
A
B
C
Inv Park
Id
Ia sampled Ia
iq measured
id measured
we measured
Id ref
Iq ref
Vd control
Vq control
Current Control
Accel
Limiter

Fig. 3.12.4. Complete control system.

Position Description y-axis range x-axis range
Top, left Electrical rotor position - to rad 0 to 2.5sec
Top, middle Q-axis current -50 to 250 A 0 to 10 sec
Top, right D-axis current 0 to 200 A 0 to 10 sec
Bottom, left
a
i before sampling
-300 to 300 A 0 to 10 sec
Bottom, middle
a
i before sampling
30.3 to 30.7 A 0.953 to 0.956 sec
Bottom, right
a
i after sampling
30.5 to 30.7 A 0.953 to 0.956 sec


Fig. 3.12.5. Simulation results for complete control system.

63
In the top row of Fig. 3.12.5, the electrical rotor position is shown on the left, the Q-axis current in the
middle and the D-axis current on the right. Notice that the currents have a small noise component:
this is due to the PWM switching. The bottom row shows the A-axis current, before, sampling on the
left. A small section of the A-axis current before sampling is shown in the middle and the same time
interval for the A-axis current after sampling is shown on the right. These results show that sampling
at the right instant gives the average current. The same sampling method is used in practice.


3.13. Noise in the measured speed signal

In practice, the speed signal contains noise. The noise can be attenuated to an acceptable level using a
low pass filter (LPF). In general it is not good practice to add a filter in the feedback path, but if the
design procedure in subsection 3.10 through to 3.12 were followed, it can be considered. In the case
of a slow speed controller (compared to the current controller), a first order LPF adds a single real pole
to the system that is separated by a great distance with respect to the closed loop poles. The LPF
therefore attenuates the noise, but it does not affect the closed loop poles or stability. The speed
measurement LPF is considered again in subsection 4.11.


3.14. An alternative speed controller

The design for the speed controller in subsections 3.10 through to 3.12 is one approach to the problem.
The speed controller is designed for the sole purpose of giving zero steady state error to a constant
speed reference and it is not designed to compensate specifically for the load torque. In fact, as Fig.
3.12.2 shows, a sudden change in load torque has a big influence on the speed.

It is possible to design the speed controller to be much faster, but at a stage, the delay due to the
current loop has to be taken into consideration. The noise in the measured speed signal and the effect
of adding a low pass filter for the speed measurement also has to be taken into consideration. Varying
parameters, such as the torque coefficient
T
K , have to be accounted for. If this approach is followed,
a suitable speed controller might be something other than a PI controller.

The approach followed in this thesis is to have a speed controller that only caters for the speed
reference. The load torque is compensated for using a load torque observer with compensation current
feedback (see chapter 5). The speed controller works in parallel with the load torque observer and
both give a current reference. In effect, the load torque observer with current compensation is a load

64
decoupling for the speed loop, similar to the speed voltage decoupling for the current loop shown in
subsection 3.2.


3.15. Summary for design methodology

In this chapter the design of stable current controllers and a suitable speed controller for the PMA
RSM is demonstrated. Some key issues that are addressed include the discrete nature of the controller,
stability, decoupling of the electrical machine equations and integrator windup.

Proportional controllers are used for the current loops. A formula is derived for the controller gain that
guarantees stability with a margin of 10 dB. This formula can easily be adjusted to give any gain
margin of stability. The formula is generic and can be used for the D-axis and Q-axis current loops. A
computationally less intensive approximation to the formula is derived for implementation.

A proportional integral (PI) controller is used for the speed loop. The response time of the speed loop
is chosen to be much greater than the response time of the current loop, so that it can be assumed that
the current reaches its reference value instantaneously. This assumption leads to a simple design. Care
is taken to prevent integrator windup.

An accurate simulation model is achieved by using results from finite element analysis. A model for
the inverter with dead time is given. Simulation results reveal that the current and speed controllers
give stable results that correspond well to the designed criteria.

It was shown that the load torque is compensated for with the same time constant for which the speed
controller was designed. This is a problem and is addressed in Chapter 5. Practical considerations and
the practical implementation of the current and speed controllers are discussed in Chapter 4.


65
Chapter 4 Practical considerations

In Chapter 3, current and speed control of the PMA RSM is derived theoretically and it is shown by
simulation that satisfactory results are obtained. These results however, need to be confirmed
practically and there are many practical considerations, which will be dealt with in this chapter.
Practically measured results are given at the end of this chapter, which can be compared to some of the
simulation results of Chapter 3.

Conventions play an important part. The first thing to do is to make sure that all the components of the
drive system adhere to these conventions. Fig. 4.1 illustrates the system as a whole (the details of the
digital signal processor (DSP) are omitted here, but are given in subsection 4.9). This chapter gives an
explanation of the different components shown in Fig. 4.1 and the conventions that are followed.


Fig. 4.1. Complete drive system.

66
4.1. Representation of the machine as seen from the load

The rectifier and inverter are connected so that the colours red, white and blue correspond to phases A,
B and C respectively. The connection to the machine (PMA RSM) is arbitrary, although it is preferred
that the machines rotor rotates anticlockwise, as seen from the load, for the ABC sequence of voltages
or currents. To test this, a 25 Hz ABC sequence of voltages is applied to the machine (no current or
position feedback) so that the machine behaves like an induction motor. It is then verified that the
ABC voltage sequence is indeed applied to the machine (not the ACB sequence) and that the rotor
rotates anticlockwise as seen from the load. Anticlockwise is chosen as the positive direction, because
that is the convention in phasor diagrams.

The representation of the machine as seen from the load is shown in Fig. 4.1.1. This representation is
correct, because the ABC sequence of currents gives a magnetic flux wave that rotates anticlockwise.
The rotor will therefore also turn anticlockwise as it follows the flux wave like an induction motor.
Note that the representation in Fig. 4.1.1 is for a 2 pole machine, although the actual PMA RSM has 6
poles as shown in Fig. 4.1.2.

C axis
A axis
B axis
A
A'
B
B'
C
C'
A C' B
MMF direction:
ABC sequence peaks: A' C B'

Fig. 4.1.1. Machine representation. Fig. 4.1.2. Cross section of the 6-pole rotor.



67
4.2. The rectifier

The rectifier is fed by a rms AC V V
source
660 = (line-to-line) source. The rectifier is a three-phase
full bridge rectifier and yields a DC voltage of
V V V
LL DC
891 2
3
= =

(4.2.1)
The measurement of the DC voltage however, shows that it varies with load: under no load conditions
the voltage is typically V V
DC
920 = and under full load conditions it is typically V V
DC
890 = . An
explanation of this type of rectifier and equation (4.2.1) can be found in [B10, p.103].

Note that in Fig. 4.1 there are three sets of switches to the left of the rectifier. E is the emergency stop
button that disconnects the power supply from the rectifier, and is normally in the closed position. S1
is the contactor that connects the supply to the rectifier. Considerable over-voltages and large inrush
currents can result at turn-on if the AC voltage is suddenly applied to the circuit by means of the
contactor. The solution is to use a current limiting resistor that is shorted out by S2 after a few cycles
subsequent to turn-on, in order to avoid power dissipation and substantial loss of efficiency due to the
presence of the resistor [B10, p.112].

The rectifier keeps the capacitor bank charged at a voltage of approximately V V
DC
900 = . The
inverter uses this DC voltage to create a variable frequency three-phase voltage supply for the PMA
RSM. DC current is drawn out of the capacitors and flows through the inverter switches to the PMA
RSM if the PMA RSM is operating as a motor. Under certain circumstances, the PMA RSM can
operate as a generator, in which case the current flows out of the PMA RSM, through the inverter
switches and back into the capacitor bank. In this case, the capacitor bank is also being charged via
the inverter by the current from the PMA RSM, and so the capacitor voltage rises above
V V
DC
900 = . Special care has to be taken for this situation: two analogue electronic circuits (one
for the dumping circuit and one for the warning light circuit), one power electronics switch (IGBT)
and a large resistor provide a method to deal with this situation, as described in subsection 4.3.


68
4.3. The dumping circuit and the warning light circuit

The dumping circuit and the warning light circuit can be found on the right hand side of the rectifier in
Fig. 4.1. The dumping circuit is needed for protection of the capacitor bank and inverter switches
when the PMA RSM is operating in generator mode. In the generator mode of operation, the PMA
RSM will deliver current instead of absorbing current. This will cause the DC bus voltage to rise
above the value given by equation (4.2.1). The dumping circuit senses the DC bus voltage and works
on a hysteresis loop basis: an IGBT is switched on if the DC bus voltage exceeds a preset value
( V V
DC
1050 > ), so that a resistor can discharge the capacitor bank and the IGBT is switched off
when the voltage reaches a suitable value again ( V V
DC
1000 < ). This concept is illustrated in Fig.
4.3.1.


Fig. 4.3.1. Dumping circuit behavior.

The machine goes into generator mode for example when it is running at a high speed and then
suddenly stops. The kinetic energy due to rotation is transformed into electrical energy in the form of
current flowing out of the machine. Current now flows into the capacitor bank from the supply side
and the inverter side, and so the capacitor bank voltage rises. The voltage can rise to dangerous levels
damaging the inverters IGBTs and the capacitor bank.

The warning light circuit is a visual aid to the operator of the machine: the light shines under normal
circumstances, but the light is switched off when dumping occurs [V2]. The dumping circuit and
warning light circuit have the same AC V V 220
sup
= supply. The shining light therefore indicates
that the dumping circuit has power and is operational. The operator should always make sure that the
light shines before switching on S1 and S2 in Fig. 4.1. If at any time, the light does not shine, it means
that there is a large current flowing through the dumping resistor. The resistor becomes extremely hot
and may cause damage to the surrounding equipment. If this situation persists for too long
( sec 10 >
dump
T ), the operator should cut off the rms AC V V
source
660 = power supply to the
rectifier using the emergency stop button (E in Fig. 4.1).

69

The circuit diagrams for the dumping circuit and warning light circuit are shown in Fig. 4.3.2 [C1].
The circuits are explained briefly as follows. Both circuits need a DC power supply and for this a
transformer [D11], bridge rectifier [D10] and regulators [D4] and [D5] are used. The transformer has
a single primary coil for the rms AC V V 220
sup
= supply and two secondary coils with a winding
ratio such that the voltage in each secondary coil is rms AC V V 12
sec
= . That means a peak value of
peak AC V V 17 12 2
sec
= = in the secondary coil. The diode bridge rectifies the secondary
voltage and the regulator (with external capacitors) keeps the voltage at V V
cc
15 + =
+
or V V
cc
15 =

depending on the type of regulator. The dumping circuit requires only a V V
cc
15 + = supply, while
the warning light circuit requires a V V
cc
15 = supply.

For the dumping circuit, the high DC bus voltage is divided to a suitable low voltage. The low voltage
passes through a buffer to the inverting input of an operational amplifier (op-amp). This op-amp [D7]
is used as a comparator: if the inverting input
n
V , is greater than the non-inverting input
p
V , then the
output is equal to the negative supply voltage to the op-amp
GND
V , otherwise if
p
V is greater than
n
V ,
the output is equal to the positive supply voltage to the op-amp
cc
V .

If the DC bus voltage becomes too large,
n
V becomes greater than
p
V and so the output of the
comparator is equal to
GND
V . The output of the comparator is the input to the IGBT driver chip [D2].
The IGBT driver chip inverts its input signal and gives an output that is suitable to drive the IGBT. If
the input to the driver chip is
GND
V (DC bus voltage is too large), the output of the chip is high, i.e.
cc
V , and therefore the IGBT is switched on.


70


Fig. 4.3.2. The dumping circuit, warning light circuit and a representation of the capacitor bank.

71
It can be seen from Fig. 4.3.2 that there are two IGBTs in this module [D8], but only one is used. To
make sure that the right hand side IGBT does not conduct, the base and emitter terminals are
connected. If the left hand side IGBT conducts, the dumping resistor and capacitor bank are connected
in parallel, as shown in Fig. 4.1. Therefore, when the IGBT conducts, the capacitor bank can
discharge through the dumping resistor.

Two variable resistors, R9 and R10, set the hysteresis width and position along the horizontal axis, as
shown in Fig. 4.3.1. The equations that describe the operation of the circuit are shown next and the
values for R9 and R10 are calculated to give a switch-off voltage of V V
dc
1050 = and switch-on
voltage of V V
dc
1000 = . From Fig. 4.3.2:

1 2
2
R R
R
V V
dc n
+
= (4.3.1)


5 10 7 9
10 7 9
10 7 5 9
1 1 1
0
0
R R R R
R R
V
R
V
V
R R
V V
R
V
R
V V
out cc
p
out p p cc p
+
+
+
+
+
=
=
+

(4.3.2)

With V V
dc
1050 < ,
cc out
V V = so that the output of the IGBT driver chip is low and the IGBT does
not conduct. To have
cc out
V V = , we should have
n p
V V > . At the point where V V
dc
1050 = , there
is a change from
cc out
V V = to V V V
GND out
0 = = and at the moment of changeover
n p
V V = .
Substituting V V
dc
1050 = ,
cc out
V V = and other known values from Fig. 4.3.2 into equations (4.3.1)
and (4.3.2), and then setting equation (4.3.1) equal to equation (4.3.2), will give:


7 . 10685
1
150
1 1
6800
1
150
1 1
150
15 15
10 56
56
1050
10 9
10 9
10 9
=
+
+
+
+
+
+
+
=
+
R k R
R k R
R k R
M k
k
(4.3.D1)


72
After the changeover from the nonconductive state of the IGBT to the conductive state,
V V V
GND out
0 = = . During this time, the capacitor bank will discharge through the dumping resistor
and when the DC bus voltage reaches V V
dc
1000 = , the output voltage changes from V V
out
0 = to
cc out
V V = . Substituting V V
dc
1000 = and V V
out
0 = and other known values from Fig. 4.2.2 into
equations (4.3.1) and (4.3.2), and then setting equation (4.3.1) equal to equation (4.3.2) leads to:

|
|

\
|
+
+
=
+
+
+
=
+
6800
1
150
1
7 . 1
1 1
6800
1
150
1 1
15
10 56
56
1000
10 9
10 9
9
R k R
R k R
R
M k
k
(4.3.D2)

Therefore, with two unknowns
9
R and
10
R , the two equations (4.3.D1) and (4.3.D2) can be solved
simultaneously. The result is:

=
=
74396
11220
10
9
R
R
(4.3.D3)

Setting the variable resistors equal to the designed values in equation (4.3.D3), the dumping circuit
should switch on the IGBT if V V
dc
1050 > and then switch it off again when V V
dc
1000 < . To test
the circuit before implementation, it is advisable to use parallel resistors across
1
R and
2
R in Fig. 4.3.2
and to test the circuit at a lower bus voltage.

The warning light circuit is explained as follows. When current flows through the dumping resistor, a
current sensor [D3] transforms the large dumping current to a small current (from the data sheet, the
ratio is given by 1:1000) at the sensing output of the transformer. Note that the current transformer
needs a voltage supply of V V 15
sup
= . With a DC bus voltage of V V
dc
1050 = and a dumping
resistance of = 27
dump
R , the primary current of the transformer is given by A I
dump
39 = . The
secondary current of the transformer is therefore given by mA I 39
sec
= .

The secondary current is too small to energize the coil of the relay [D9]. For this relay, at least
mA I
energize
89 = is needed to energize the coil (causing the switch in the relay to move from one
position to the other) and therefore a current amplifier is used [D1]. The switch inside the relay is
normally in such a position that there is a positive voltage across the lamp (warning light) and

73
therefore the lamp shines. When the coil is energized (dumping occurs), the switch inside the relay
moves to the other position such that the voltage across the lamp is zero.

In Fig. 4.3.2 a representation of the capacitor bank is also shown. In the practical setup used, the
capacitor bank can be found inside the inverter casing [D15, p.123]. The capacitor bank has 12 sets of
3 capacitors in series and the sets are in parallel. Each capacitor is F C
cap
6800 = and rated for
400V. The 3 capacitors in series give a rated voltage of 1200 V. The total capacitance is given by:
( ) mF C
total
27 6800 6800 6800 12
1
1 1 1
= + + =


(4.3.3)


4.4. The load system

The load system that is coupled to the PMA RSM consists of two hydraulic dynamometers [W5].
Load is increased by increasing water flow to the dynamometers. The water flow is controlled by an
analogue electronic control unit. Additional switches were installed by the author to open and close
the valves completely, thereby synthesizing a load step.

The response time of this load system is not very fast. The applied torque magnitude is a function of
the rotational speed of the machine: the torque increases as the speed increases. Therefore a positive
torque step is approximated with a gentle slope. A negative load torque step is not possible while the
machine is producing positive torque. Positive load torque can be reduced to zero fairly quickly, but it
is still not a very good approximation of a step.

The load system used is a limitation to the study, because it is not dynamic enough to test the dynamic
response of the PMA RSM. Another motor (e.g. DC motor) or electrical type of dynamometer would
be a better dynamic load for the PMA RSM.


4.5. The inverter

The inverter [D15, p.123] used in the practical setup is a half-bridge inverter with three phase arms
(see Fig. 4.1). For each phase arm, the two switches (IGBTs) can never conduct simultaneously, since
this would cause the DC bus to be shorted. To make sure this never happens, a small amount of dead
time ( s
dt
4 . 6 = in this case) is built into the switching signals generated by the DSP. Dead time,
or blanking time, is needed because the switches are not ideal.


74
It is suspected that the inverter has hardware that adds additional dead time. To measure the total dead
time, the switching signals on the IGBTs would have to be measured. It is extremely difficult to reach
the IGBTs inside the inverter casing and therefore this measurement could not be done. The dead time
added in the control program is s T
time dead
4 . 6 = based on the choice by [B5, p.173], although it is
not motivated by [B5]. For all simulation purposes, the dead time will be assumed to be
s T
time dead
10 = .

The symmetrical pulse width modulation (PWM) scheme is used to decide when the top switch or the
bottom switch should conduct. This scheme is the same for each arm and is explained briefly. The
switching scheme is based on a triangular waveform with a period that is the inverse of the switching
frequency and a peak-to-peak magnitude that is the DC bus voltage. Since the switching frequency is
much faster than the reference signals frequency, the reference signal can be assumed to be constant
for a few cycles of the triangular waveform. The various signals are shown in Fig. 4.5.1.

Fig. 4.5.1. Switching scheme for half-bridge inverter.

The formula for Ton and Toff (for the top switch) in Fig. 4.5.1 is given as follows:
|
|

\
|
+ =
|
|

\
|
=
DC
ref
s
off
DC
ref
s
on
V
V
T
T
V
V
T
T
2
3
2
2
1
2
(4.5.1)

Equation (4.5.1) applies to all phases, for example
AN ref
V V = for phase A. The switching signal for
the bottom switch is just the inverse of the switching signal for the top switch, as shown in Fig. 4.5.1.

75
By using equation (4.5.1), the fundamental component of the inverter output for each phase will be
equal to the reference signal for each phase respectively. In other words the gain of the inverter
should be equal to unity. Practically it is found however, that for reference values that are small
compared to the DC bus voltage, the gain is less than unity. This is due to the dead time as indicated in
Fig. 4.5.1. For example, if the reference voltage is only 10% of the DC bus voltage, the gain of the
inverter is 85 . 0 =
ref
out
V
V
. The effect of this is that small current references will not be followed exactly,
i.e. there will be a large error.

The formulas for the switching times in equation (4.5.1) require the DC bus voltage to be measured.
The measurement of the DC bus voltage and the three-phase currents are similar and are explained in
the subsection 4.6.


4.6. DC bus voltage and three-phase current measurement

The DC bus voltage is measured with a LEM voltage transducer [D6]. The DC bus voltage is applied
to an external measurement resistor. The current that flows through the resistor is the primary current
of a transformer in the voltage transducer. The secondary current of the transformer can be used in a
current measurement circuit and represents the DC bus voltage. The secondary (low voltage) circuit
has galvanic isolation from the primary (high voltage) circuit.

From the datasheet [D6], the primary nominal RMS current is mA I
PN
10 = . Since the DC bus
voltage is approximately V V
DC
1000 = , the measuring resistance is chosen as
= = k
m
R
M
100
10
1000
. According to the winding ratio 10000:2000, the secondary nominal current
is given by mA I
SN
50 = . This secondary current represents the DC bus voltage of V V
DC
1000 = .

A current measurement circuit, which is simply a resistor and an Op-amp with unity gain, is used to
measure the current. The value for the resistor should be chosen so that the voltage is
V V V
measure
5 . 2 5 . 2 < < . The reason is that the analogue-to-digital converter (ADC) needs an input
voltage that is V V V
ADC
5 0 < < . The voltage for the ADC is obtained by taking
V V V
measure ADC
5 . 2 + = .


76
The maximum DC bus voltage that will be measured is V V
DC
1200 = . Therefore the maximum
secondary current is mA
k
I
SN
60
2000
10000
100
1200
= = . If the resistance on the current measurement card
is chosen to be = 41
meausure
R , the voltage measurement is V m V
measure
46 . 2 60 41 = = . The
chosen resistor value gives a voltage that adheres to the constraint V V V
measure
5 . 2 5 . 2 < < and is
therefore suitable.

The equation for the digitally represented DC bus voltage is therefore:
5 . 2
488
5 . 2 41
2000
10000
100
1
) (
+ = + =
DC
DC dig DC
V
k
V V (4.6.1)

The three-phase current measurement is very similar to the DC bus voltage measurement. Three LEM
current transducers [D12] are used. Each phase current has its own current transducer. Phase A
current, for example, is the primary current for a transformer inside the current transducer. The
secondary current represents the primary current, but has galvanic isolation from the primary circuit.
From the datasheet [D12], the winding ratio is given by 1:5000. The maximum rms current that will
flow is chosen to be ( ) ( ) rms A I I
rated PN
300 200 5 . 1 5 . 1
max
= = = . The maximum secondary
current is therefore given by peak mA rms mA I
SN
85 60
5000
1
300
max
= = = .

The same type of current measuring circuit is used for both DC bus voltage measurement and three-
phase current measurement. The resistor for the current measurement circuit for the three-phase
current measurement is chosen to be = 27
meausure
R so that the maximum measured voltage is
V m V
measure
295 . 2 85 27 = = . The minimum measured voltage is of course
( ) V m V
measure
295 . 2 85 27 = = , and so the voltage measurement adheres to the constraint
V V V
measure
5 . 2 5 . 2 < < and is therefore suitable for the ADC.

The equation for the digitally represented current is therefore:
5 . 2
185
5 . 2 27
5000
1
) (
+ = + =
peak
peak dig peak
I
I I (4.6.2)
The circuit diagram for the current measurement is shown in Fig. 4.6.1. In this figure, two measuring
resistances
1
R and
2
R , can be chosen and the appropriate resistance can be selected with the jumpers
1
J and
2
J . Note that the circuit has a power supply of V V 15
sup
= . The supply is obtained from

77
the mother board on which the DSP, FPGA, ADCs, DACs etc. are mounted. The supply is fed through
to the voltage or current transducers, since they also need a power supply.

The voltage and current transducers form the interface between high power and low power. The low
power signal is however still analogue and needs to be converted to a digital signal. This is performed
inside the system unit labeled DSP in Fig. 4.1. Subsection 4.9 explores the contents of this system
unit. The next subsection however, shows how rotor position is measured using a resolver.


Fig. 4.6.1. The current measurement card designed by Mr. A.D. le Roux.

78
4.7. Rotor angular position measurement using a resolver

A resolver [W1] is a position sensor or transducer which measures the instantaneous mechanical
angular position of the rotating shaft to which it is attached. Resolvers are typically built like small
motors with a rotor (attached to the shaft whose position is to be measured), and a stator (stationary
part) which produces the output signals.

Electrically, the resolver is a transformer in which the coupling between the primary coil and the
secondary coils vary as the sine and cosine of the mechanical rotor angle. Whereas a traditional
resolver has its primary coil on the rotor and its secondary coils in the stator (necessitating brushes and
slip rings or a rotating transformer to couple signals into the primary), modern resolvers have both
primary and secondary coils in the stator and use a unique solid rotor to directly vary the coupling
between the primary coil and the secondary coils.

Like all transformers, the resolver requires an AC carrier or reference signal (sometimes also called the
excitation signal) to be applied to its primary coil. The amplitude of this reference signal is then
modulated by the sine and cosine of the rotor angle to produce the output signals on the two secondary
coils respectively. Figure 4.7.1 is an illustration of a resolver and shows the convention for the wire-
colours.

Fig. 4.7.1. A conventional resolver illustration.

In any transformer, there is a value which relates the output voltage produced by the secondary to that
fed into the primary. For resolvers, this quantity is called the transformation ratio (TR) and is specified
at the point of maximum coupling between primary and secondary. For industrial resolvers, the
standard transformation ratio is 0.5, which means that the maximum voltage produced by either
secondary is half the amplitude of the reference signal.

Let the excitation voltage be:
) cos( t V
P
= (4.7.1)



79
Let the transformation ratio be:

2
1
= TR (4.7.2)
The voltages on the secondary coils are given by:
) cos(
2
1
) cos( ) cos(

t TR V V
p
= = (4.7.3)
) sin(
2
1
) cos( ) sin(

t TR V V
p
= = (4.7.4)
At this point there are some issues that need to be addressed: the frequency of the excitation voltage
should be chosen, a method for applying the excitation voltage to the resolver is needed and a
demodulation method, using the measured voltages of the secondary coils, needs to be identified.

The excitation frequency is chosen to be:
) ( 5 . 0
s
= , (4.7.5)
where sec / 5000 2 rad
s
= is the switching frequency of the inverter. Samples of the excitation
voltage are stored in the DSPs [D13] memory. These digital samples should be written out of
memory and converted to analogue signals to be applied to the resolvers primary coil. The length of
one switching period is s T 200 = . If 40 = N samples of the excitation voltage are stored in
memory then 20
2
=
N
samples need to be written out of memory in one switching period to give the
excitation frequency of ) ( 5 . 0
s
= . That means one sample has to be written out every
s T
DMA
10 = . The samples are written out using a direct memory access (DMA) scheme, which will
be explained shortly. First it is necessary to introduce the FPGA.

Many devices (ADCs, DACs, buttons, turning knobs, PWM card etc.) need to be connected to the
DSP, but the DSP only has a limited number of pins and it is therefore necessary to have a routing
device. The routing device is a field programmable gate array (FPGA) that contains logical hardware
and memory [D14]. The FPGA is discussed further in subsection 4.9.

The samples of the excitation voltage are stored in a record structure in the DSPs memory that has a
field for the destination analogue output port, a field for the timer (a timer that is set up to cause an
interrupt every s T
DMA
10 = ), a field for the value of the voltage for the first channel of the output
port and a field for the value of the voltage for the second channel of the output port. The record
structure is illustrated in Fig. 4.7.2. Note that only one channel (channel 0) is used for the excitation
voltage that is applied to the resolver.

80

Destination port Timer trigger V0 (channel 0) V1 (channel 1)
Fig. 4.7.2. Record structure in the DSPs memory.

The 40 = N samples of the excitation voltage are thus stored in the memory of the DSP as a stack of
40 = N records, each record with the structure given in Fig. 4.7.2. Some low level instructions cause
one record to be transferred to from the DSPs memory to the FPGAs memory. Once in the FPGAs
memory, the record has to wait in a first-in-first-out (FIFO) queue. This FIFO queue is used
exclusively for analogue output. For the record in front of the FIFO queue, the FPGA waits for the
timer (specified in the timer field) to cause an interrupt and then converts the voltages (specified in the
voltage fields) from digital-to-analogue and applies them to the destination port (specified in the
destination port field).

The FPGA has the ability to accept an entire block of records from the DSPs memory and process it
without any further participation from the DSP. This is known as direct memory access (DMA).
Using DMA, the FPGA takes some processing load off the DSP concerning analogue output. In one
switching period of s T 200 = (period one), 20 = N samples of the excitation voltage that are
stored in the memory of the DSP, are transferred to the memory of the FPGA, so that they can be
applied sequentially (with s T
DMA
10 = between samples) during the next switching period (period
two). During period two, the other 20 = N samples of the excitation voltage that are stored in the
memory of the DSP are transferred to the memory of the FPGA, so that they can be applied
sequentially during period three. In period three, the first set of 20 = N samples is transferred again
and so forth.

The voltages of the secondary coils of the resolver need to be converted from analogue-to-digital, i.e.
sampled and stored in the DSPs memory. The information about the rotor position lies in the
amplitudes of the secondary voltages, as shown in equations (4.7.3) and (4.7.4). The secondary
voltages need to be sampled with a frequency of at least twice their highest frequency component
(Nyquist theorem). Since the frequency of the secondary voltages is chosen to be ) ( 5 . 0
s
= , the
switching frequency
s
is sufficient to sample the secondary voltages.

Fig. 4.7.3 is an illustration that depicts the excitation voltage samples in memory, the applied
excitation voltage to the primary coil of the resolver and the resultant secondary coil voltages, for a
rotor angle of = 45 . The figure indicates the samples that are written out to the resolver and the
samples that are taken from the secondary coils.

81

Fig. 4.7.3. Resolver voltages for = 45 and
2
1
= TR .

The demodulation procedure is explained next using equations (4.7.1) through to (4.7.5) and Fig.
4.7.3. From Fig. 4.7.3, note that the secondary voltages of the resolver are only sampled at the
switching instants. The difference in value from one sample to the next therefore gives the peak-to-
peak value of the voltage. The peak-to-peak values of the voltages can be used to find the
instantaneous value of the rotor position. The rotor position changes slowly with respect to the carrier
frequency ) ( 5 . 0
s
= and therefore it is accurate to assume that the peak-to-peak value of the
voltage is the difference between one sample value and the next.

The first step towards demodulation is therefore the sampling of the secondary voltages and obtaining
the peak-to-peak values. Consider always subtracting the previous value from the present value: the
result is the peak-to-peak voltage, but the sign alternates. Now consider always subtracting samples
marked with (1) from samples marked with (0): the result is the peak-to-peak magnitude of the
voltage and it is always positive.

For example, assume that the moment in time is (t = T2), as indicated in Fig. 4.7.3. At this moment,
the present values
) 1 (

V and
) 1 (

V are subtracted from the previous values

) 0 (

V and
) 0 (

V to obtain the
peak-to-peak magnitudes
) 1 ( ) 0 (

V V V = and
) 1 ( ) 0 (

V V V = . At the next sampling instant (t

82
= T3), the previous values
) 1 (

V and
) 1 (

V are subtracted from the present values

) 0 (

V and
) 0 (

V to
obtain the peak-to-peak magnitudes
) 1 ( ) 0 (

V V V = and
) 1 ( ) 0 (

V V V = .

From the discussion of the last three paragraphs and equations (4.7.2) through to (4.7.4), the peak-to-
peak magnitudes are given by:
[ ]
) cos( 2
) cos( ) cos(
) 1 ( ) 0 (

=
=
=
TR
TR TR
V V V
(4.7.6)
[ ]
) sin( 2
) sin( ) sin(
) 1 ( ) 0 (

=
=
=
TR
TR TR
V V V
(4.7.7)
It is important to note that equations (4.7.6) and (4.7.7) assume that change in rotor position from one
sampling instant to the next is negligible. Using equations (4.7.6) and (4.7.7), it is possible to calculate
the instantaneous rotor position:

|
|

\
|

V
V
arctan (4.7.8)
The rotor angle in equation (4.7.8) does not necessarily correspond to the rotor angle
p
e
m

= ,
where
e
is defined as the angle between the A-axis and Q-axis of the PMA RSM. To obtain the
correct rotor angle:

|
|

\
|

= + =

V
V
pos zero m
arctan
_
(4.7.9)

In equation (4.7.9),
pos zero _
is adjusted in the trial-and-error procedure discussed in subsection (4.8).
Equation (4.7.9) provides one method of demodulation, but there are many other methods [D16] that
increase the robustness of the rotor position measurement against noise. These methods might use
techniques such as correlation or structures such as the Kalman filter or a phase locked loop (PLL).
One such method is explained briefly and it is noted that the ideas used in this method resemble the
ideas used for position sensorless control, which are discussed in Chapter 6.

The secondary voltages that are measured are in the stationary reference frame. These voltages
can be transformed to the synchronously rotating 0 QD reference frame. Note that these reference

83
frames belong to the resolver and not to the PMA RSM: the difference in angular position between the
resolvers 0 QD reference frame and the PMA RSMs 0 QD reference frame is given by
pos zero _
.

Transformation from the reference frame to the QD0 reference frame requires the angular rotor
position , but this is exactly what we want to determine. Consider then the transformation from the
reference frame to an observed (estimated) QD0 reference frame, using an observed rotor position

. The difference between the observed QD0 reference frame and the actual QD0 reference frame is
calculated and used to update the observed rotor position. Fig. 4.7.4 shows a phasor diagram from
which the transformation from the reference frame to the observed QD0 reference frame can be
derived.

90
( ) cos
( ) sin

90
D
V

Q
V

Fig. 4.7.4. Transformation between reference frame and 0 QD reference frame.

From Fig. 4.7.4, the resolvers secondary voltages in the observed QD0 reference frame are given by:

)

sin( )

cos(
)

90 cos( )

cos(

V V
V V V
D
+ =
+ =
(4.7.10)

)

cos( )

sin(
)

cos( )

90 cos(

V V
V V V
Q
+ =
+ =
(4.7.11)
Substituting equations (4.7.6) and (4.7.7) into equations (4.7.10) and (4.7.11):

)

cos(
)

sin( ) sin( )

cos( ) cos(

=
+ =
D
V
(4.7.12)

)

sin(
)

cos( ) sin( )

sin( ) cos(

=
+ =
Q
V
(4.7.13)

84
Using equations (4.7.12) and (4.7.13), the error between the observed and actual rotor position is then
given by:

|
|

\
|

=
D
Q
V
V

arctan

(4.7.14)
Therefore, at each switching instant (T1, T2, T3 in Fig. 4.7.3), the secondary voltages of the resolver
are sampled and the peak-to-peak magnitudes are calculated using the previous sampled values of the
voltages. The peak-to-peak magnitudes are then transformed to an observed QD0 reference frame and
the transformed magnitudes are used to calculate the error between the observed and actual rotor
positions. After the error has been calculated, the final step is to update the observed rotor position:
=
|
|

\
|

+
D
Q
V
V

arctan

(4.7.15)
The second demodulation method shown is not necessarily more robust against noise, but it is a step
towards more sophisticated demodulation methods that might improve the robustness against noise.

In this subsection, a general description of a resolver and its characteristic equations are given. A
DMA scheme for applying the excitation voltage to the primary coil is shown and the excitation
frequency is chosen. Two possible demodulation methods are then shown. The second demodulation
scheme is used in the control program for the PMA RSM.

With position measurement in place, it is now possible to set the zero position of the machine (align
the QD0 reference frame of the resolver with the QD0 reference frame of the PMA RSM). The next
subsection reminds us why control of the PMA RSM is performed in the QD0 reference frame and
gives examples of the mapping between the ABC reference frame and the QD0 reference frame. The
definition of zero position in terms of the ABC and QD0 reference frames is given followed by a
procedure to set the zero position.


85
4.8. Zero position

The rotor of the PMA RSM is magnetically asymmetrical giving rise to 2 axes: one with high magnetic
permeance, which is called the Direct-axis (D-axis), and the other with low magnetic permeance,
which is called the Quadrature-axis (Q-axis). A cross section of the actual 6-pole rotor is shown in
Fig. 4.1.2. The D-axis is along the path from the centre outward where there are no flux barriers in its
path. The Q-axis is along the path from the centre outward where there are flux barriers. Fig. 4.1.2
also shows that the permanent magnets are positioned along the Q-axis.

For the control of the PMA RSM, the angular position of the rotor needs to be known. This is because
the control is performed in the QD0 reference frame, which rotates synchronously with the rotor. The
Park transform [B8] is used to transform the ABC reference frame to the QD0 reference frame. A
short example illustrates how this works:

A command is given that requests positive Q-axis and D-axis currents. These positive currents give a
flux linkage wave that is the vector sum of the positive Q-axis flux linkage and positive D-axis flux
linkage components. The rotor experiences a reluctance force to align itself with the flux wave and
rotates in that direction. To produce the positive Q-axis and D-axis currents, a certain combination of
ABC currents is needed. This combination is unique for each electrical rotor position. Therefore,
when rotation occurs, the ABC currents adjust continually to fulfill the requirement of constant QD0
currents.

Fig. 4.8.1 is a phasor diagram that establishes the relationship between the following quantities: D-axis
current
d
i , Q-axis current
q
i , stator current
s
i , current angle , D-axis flux linkage
d
, Q-axis flux
linkage
q
, stator flux linkage
s
, flux angle and electrical angular position
e
. Fig. 4.8.2
illustrates how constant QD0 currents give rise to different ABC currents for different rotor positions.
D-axis
Q-axis
d
i
q
i
q

s
i

Fig. 4.8.1. Phasor diagram for currents and flux linkages.


86

Fig. 4.8.2. pu I I pu I
c b a
5 . 0 , 1 , 20 = = = = (left) pu I I pu I
b a c
5 . 0 , 1 , 80 = = = = (right).

The rotor position, defined as the angle from the A-axis to the Q-axis, together with the QD0 current
references translate into ABC current references using the Park transform. The zero position is
where is Q-axis is parallel to and in the same direction as the A-axis as shown in Fig. 4.8.3, or where
= 0 as shown in Fig. 4.8.2. For a RSM (no permanent magnets in the rotor) the zero position can
be obtained by letting Ia = 0 pu, Ib = -1 pu and Ic = 1pu.

The PMA RSM has the constraint that the permanent magnets need to produce flux linkage that is in
the negative Q-axis direction. The procedure used to find the zero position for the RSM cannot be
used for the PMA RSM, because the constraint mentioned might not be met (the permanent magnets
might end up producing flux linkage in the positive Q-axis direction). An alternative method, where
the permanent magnets are used to find the zero position for the PMA RSM is suggested and this
procedure is explained next.
C axis
A axis
D axis
Q axis
B axis
A
A'
B
B'
C
C'
D
D'
Q'
Q
N
S
pm

Fig. 4.8.3. Definition of zero position for PMA RSM
C axis
A axis
D axis
Q axis
B axis
A
A'
B
B'
C
C'
D
D'
Q'
Q

C axis
A axis
D axis
Q axis
B axis
A
A'
B
B'
C
C'
D
D'
Q'
Q

87
Use the inverse Park transform to transform the rotor flux linkages caused by the permanent magnets,
to the stator flux linkages:
( ) ( ) ( ) ( )
(
(
(

(
(
(
(
(
(
(

\
|
+ |

\
|
+
|

\
|
|

\
|
=
(
(
(

(
(
(
(
(
(
(

\
|
+ |

\
|
+
|

\
|
|

\
|
=
(
(
(

0
0
1
3
2
sin
3
2
cos
1
3
2
sin
3
2
cos
1 sin cos
1
3
2
sin
3
2
cos
1
3
2
sin
3
2
cos
1 sin cos
0
pm
d
q
c
b
a

( )
(
(
(
(
(
(
(

\
|
+
|

\
|

=
(
(
(

3
2
cos
3
2
cos
cos

pm
pm
pm
c
b
a
(4.8.1)
Faradays law gives the induced voltage due to the change in flux linkage:
( )
(
(
(
(
(
(
(

\
|
+
|

\
|
=
(
(
(

=
(
(
(

3
2
sin
3
2
sin
sin

pm
pm
pm
c
b
a
c
b
a
dt
d
e
e
e
(4.8.2)
From equation (4.8.2), the induced voltage in phase A is given by ) sin(
pm
, where 0 >
pm
.
Rotating the rotor by hand and measuring the induced voltage in phase A, the electrical rotor position
should be in phase with the induced voltage as indicated in Fig. 4.8.5 (this is a practical
measurement). The zero position constant in the control program is adjusted until the electrical rotor
position matches that of Fig. 4.8.5. This is a trial-and-error procedure.

Fig. 4.8.5. Rotor position and induced voltage in Phase A due to permanent magnets.

88

Note that the zero position for the RSM could be at either zero point in the induced voltage waveform,
but the PMA RSM requires it to be at the positive slope of the sine wave, according to equation
(4.8.2). For the practical measurement shown in Fig. 4.8.5, a star-point adapter is used to create a
neutral point for measurement of phase A voltage (
AN
V ).


4.9. Overview of the DSP

In Fig.4.1 the DSP is shown as a black box. In subsection 4.7 items such as the FPGA, ADC, DAC,
output port, etc are mentioned. This subsection aims to put the low power devices into perspective.
Fig. 4.9.1 shows the interior of the black box labeled DSP in Fig. 4.1:

Fig. 4.9.1. Low power devices.

In Fig. 4.9.1, the analogue inputs from the CTs, VT, resolver and turning knobs are always available to
the ADCs. When an instruction is executed by the DSP to obtain any of these values (currents, DC
bus voltage, secondary voltages of the resolver, position of a turning knob), it is almost immediately
available to the DSP. The analogue output (for display on an oscilloscope or the excitation voltage to
the resolver) is handled with direct memory access (DMA), as explained in subsection 4.7.

At each sampling instant, the three-phase currents, DC bus voltage, rotor position and position of the
turning knobs are determined. The turning knobs give the speed reference and the acceleration
reference. The Park transform, digital speed controller, digital current controllers and inverse Park
transform give ABC voltage references, which are used in equation (4.5.1) to determine the switching
times for the inverter switches. The switching times are converted to electrical switching signals by
the FPGA. The electrical switching signals are then transformed to optical signals, and travel along

89
optical fibers to the inverter. At the inverter, the optical signals are transformed back to electrical
switching signals. The switching signals are applied to IGBT drivers, which give output signals
suitable for switching the IGBTs.

As a final remark, several buttons and LEDs are also connected to the FPGA. These components can
be used for debugging the control program, switching between different analogue outputs to be viewed
on an oscilloscope and other basic input/output requirements.

The next subsection shows a state diagram of the control program. Reference is made to the
components shown in Fig. 4.9.1.


4.10. The control program

The control program [P1] is structured according to a main program section and interrupt service
routines (ISR). The main program section has initialization code followed by an explicit endless loop.
Two timers are used to control the program flow: timer0 is set to s T 10
0
= and is used to time the
DMA (analogue output), while timer1 is set to s T 200
1
= to time the execution of the ISRs.

Each time timer1 causes an interrupt, an ISR will be executed. Different ISRs correspond to different
states of the program. There are three states, namely the Waiting state, Run state and Error
state. A scheduling procedure is used to determine which ISR will be executed. At first, the Waiting
state ISR is scheduled to be executed. In the Waiting state, the PWM output signals are disabled.
The Waiting state schedules itself to be executed until a button is pressed, at which time the Run state
is scheduled. In the Run state the PWM output signals are enabled and the PMA RSM can be
controlled.

If an error occurs when the program is in the Run state, the Error state is scheduled. Certain types of
errors e.g. current reference out of range, will lead to the state change from the Run state to the Error
state. Over-current and over-voltage (DC bus voltage) errors are checked in hardware by setting limits
for the digitally represented currents and voltage. Hardware error checking is explained next.


90
Recall that the ADC requires the voltage from the current measurement cards to be
V V V
ADC
5 0 < < . From subsection 4.6 the maximum DC bus voltage that can be measured is:
( ) ( ) V V V
dig DC DC
1220 488 5 . 2 5 488 5 . 2
max ) ( (max)
= = = (4.10.1)
The maximum peak current that can be measured is:
( ) ( ) A I I
dig peak peak
5 . 462 185 5 . 2 5 185 5 . 2
max ) ( (max)
= = = (4.10.2)
The minimum DC bus voltage that can be measured is:
( ) ( ) V V V
dig DC DC
1220 488 5 . 2 0 488 5 . 2
min ) ( (min)
= = = (4.10.3)
The minimum peak current that can be measured is:
( ) ( ) A I I
dig peak peak
5 . 462 185 5 . 2 0 185 5 . 2
min ) ( (min)
= = = (4.10.4)

It is possible to set upper and lower limits for the digital representations, i.e. set
max ) (dig DC
V ,
max ) (dig peak
I ,
min ) (dig DC
V and
min ) (dig peak
I , and have hardware checking that the digital signals are
within the limits, i.e.
max ) ( ) ( min ) ( dig DC dig DC dig DC
V V V < < and
max ) ( ) ( min ) ( dig peak dig peak dig peak
I I I < < .

In the case of hardware detecting an over-current or over-voltage, the PWM outputs are disabled, but
the program remains in the Run state. The error is indicated by a red LED on the motherboard. There
is also hardware (logic in the FPGA) checking that the PWM switching signals are updated every
cycle. This is known as a watchdog timer [B13, p.202]. Every time the PWM switching signals are
updated, the watchdog timer is reset. If the watchdog timer is not reset, the PWM outputs are disabled.

To summarize, there are three states which the control program can be in, but there are four states the
control hardware can be in due to hardware error checking. The control program is illustrated next
using a finite state automaton [W4], shown in Fig. 4.10.1.


91

Fig. 4.10.1. Finite State Automaton for control program.
Main
Schedule Wait state
Setup lookup tables
Setup ADC
Setup DMA buffer for resolver
Setup timer0 for s 10 (DMA)
Setup timer1 for s 200
Enable interrupts

Endless loop

Timer1 interrupt
Scheduler

Wait
Schedule Wait state

IF (calibrate complete) THEN
LED blinking
IF (push button) THEN
Schedule Run state
END IF
ELSE
Calibrate ADC
END IF

Setup ADC hardware protection
Set PWM duty cycles
Disable PWM output
Wait scheduled
Error
Schedule Error state
Indicate error with LED
Set PWM duty cycles
Disable PWM output

IF (push button) THEN
Schedule Wait state
END IF

Error scheduled
Run
Schedule Run state
Read
ABC
I and
DC
V
Read speed and acceleration reference (knobs)
Slew rate limiter for speed reference
Read resolver voltages and calculate
e

Calculate rotor speed
e e

&
=
LPF rotor speed
Speed controller
Current controllers

IF (reference error or over speed) THEN
Schedule Error state
ELSE IF (push button) THEN
Schedule Wait state
END IF

Set PWM duty cycles
Enable PWM output

Run scheduled

92
4.11. Practical measurements

This subsection shows practical measurements for currents in the QD0 reference frame, rotor angular
position measurement and speed measurement. For current measurements, the control program
described in Fig. 4.10.1 is used, but the current references are set directly (not by the speed controller).

In Fig. 4.11.1 there are two measurements shown. For the measurement on the left, the Q-axis current
reference is set to zero and the D-axis current reference is set to a square wave with small amplitude.
Mathematically the reference signals can be described by:

= < + <
= < + <
=
=
...} 2 , 1 , 0 { 20 20 10 20
...} 2 , 1 , 0 { 10 20 0 20
0
n ms ms n t ms A
n ms ms n t A
I
I
ref D
ref Q
(4.11.1)
For the measurement on the right, the D-axis current reference is set to zero and the Q-axis current
reference is set to a square wave with small amplitude. Mathematically the reference signals can be
described by:

= < + <
= < + <
=
=
...} 2 , 1 , 0 { 20 20 10 20
...} 2 , 1 , 0 { 10 20 0 20
0
n ms ms n t ms A
n ms ms n t A
I
I
ref Q
ref D
(4.11.2)


Fig. 4.11.1. D-axis current reference (left, top), D-axis current (left, bottom), Q-axis current reference
(right, top), Q-axis current (right, bottom).

Fig. 4.11.1 shows that the D-axis current has a small steady state error (the reference and the response
are both two divisions in magnitude) and the settling time is given by ms
d s
4 = . The Q-axis current
has a very large steady state error (the reference magnitude is 2 divisions, while the response

93
magnitude is only 1 division) and the settling time is given by ms
q s
36 . 1 = . It is believed that the
very large steady state error is caused by the dead time of the inverter.

The experiment for Fig. 4.11.1 is repeated, but this time with a large current reference. In Fig. 4.11.2
the equations for the current references for the measurement on the left are given by:

= < + <
= < + <
=
=
...} 2 , 1 , 0 { 20 20 10 100
...} 2 , 1 , 0 { 10 20 0 100
0
n ms ms n t ms A
n ms ms n t A
I
I
ref D
ref Q
(4.11.3)
The equations for the current references for the measurement on the right are given by:

= < + <
= < + <
=
=
...} 2 , 1 , 0 { 20 20 10 100
...} 2 , 1 , 0 { 10 20 0 100
0
n ms ms n t ms A
n ms ms n t A
I
I
ref Q
ref D
(4.11.4)


Fig. 4.11.2. D-axis current reference (left, top), D-axis current (left, bottom), Q-axis current reference
(right, top), Q-axis current (right, bottom).

From Fig. 4.11.2, it can be seen that the D-axis current response has a very small steady state error and
the settling time is given by ms
d s
4 . 2 = . The Q-axis current response also has a very small steady
state error and the settling time is given by ms
d s
36 . 1 = .

Practical measurements for the rotor position and speed are shown next. The speed is calculated by
taking the derivative of the rotor position. The position signal has a small amount of noise and
therefore the speed signal also has noise. The amount of noise in the speed measurement can be
reduced dramatically by a simple 1
st
order Butterworth low pass filter (LPF) with a cutoff frequency at
sec / 100 rad
c
= . The choice of this frequency is arbitrary, but a low cutoff frequency results in

94
less noise output. In Fig. 4.11.3, the speed measurement before (top) and after (bottom) the LPF is
shown on the left. The electrical (top) and mechanical (bottom) rotor angular position is shown on the
right.


Fig. 4.11.3. Speed before LPF (left, top), speed after LPF (left, bottom), electrical position (right, top)
and mechanical position (right, bottom).

To obtain the signal to noise ratio (SNR) of the speed measurement, the data is captured with the
oscilloscope, saved to a disk, transferred to a computer and imported into the Matlab workspace. For a
DC signal the SNR is given by the mean of the signal squared (power of the signal) divided by the
variance of the signal (power of the noise). The derivation of the SNR is shown in Appendix F. It was
found that the SNR without the LPF is dB 13.19 = SNR and the SNR with the LPF is
dB 30.28 = SNR [M1].

95
Chapter 5 Load torque observer

This chapter explains a method by which the dynamic response of the drive system can be improved.
The load torque applied to the machine is not measured directly, but is calculated using the measured
rotor speed and stator current. Based on this observed load torque, a compensation current for the load
torque is calculated and added to the current reference command. Two methods of determining the
load torque are given, namely the reduced state observer and the full state observer.

The reduced state observer is introduced first and it is shown by means of a simulation example that
the dynamic response of a RSM with constant field current control (CFCC) can be increased by using
the load torque observer and compensation current feedback. The method is then adapted for the RSM
and PMA RSM with constant current angle control (CCAC). Using an accurate model of the PMA
RSM with CCAC, a simulation is performed to test the reduced state observer with compensation
current feedback. In this simulation, a load torque step is applied to the machine and the dynamic
response is evaluated. The proposed scheme is implemented practically and practical results are
shown. In the practical tests, the load torque is not a perfect step. Appendix H shows simulation
results where the simulated load torque matches the practical load torque.

The full state observer is then introduced. It has several advantages over the reduced state observer.
Using an accurate model of the PMA RSM with CCAC, a simulation is performed to test the full state
observer with compensation current feedback. The proposed scheme is implemented practically and
practical results are shown. Simulation results where the simulated load torque matches the practical
load torque are shown in Appendix H.


5.1. Why is the load torque observer needed?

A common control system structure for electrical machines is a speed loop with an inner current loop,
as shown in Fig. 5.1.1. The speed controller is designed by assuming that the load torque is zero [B9,
p.201 206]. A PI speed controller will ensure that the rotor speed follows a constant reference speed
with zero steady state error. The PI speed controller gives a certain response time to the speed loop
it can be designed to be slow or fast, the only constraint being that it should be much slower than the
inner current loop (see subsection 3.14). Unfortunately the load torque disturbance is compensated for
with the response time of the speed loop. Therefore if the speed loop is designed to be very slow, the
load torque compensation is also very slow.


96
I ref V ref I Tem
Speed
current loop
speed ref Y(s)
T load
D1(s)
Speed PI
G2(s)
Mech Model
Kt G1(s)
Electric Model
D2(s)
Current control

Fig. 5.1.1. General control system structure for an electric machine.

The problem is depicted in a general block diagram, shown in Fig. 5.1.2. The equation for the output
can be derived by using the block diagram and the superposition principal. It is given by:

(

+
=
) ( ) ( 1
) (
) (
) ( ) ( 1
) ( ) (
) ( ) (
S G s D
s G
s W
s G s D
s G s D
s U s Y (5.1.1)
Y(s)
W(s)
U(s)
G(s)
Plant
D(s)
PI

Fig. 5.1.2. General control system structure.

From equation (5.1.1) it can be seen that the poles for the first and second term on the right are at the
same location, indicating that the characteristic response times are the same. For example, let
s
K s K
s
K
K s D
I P I
P
+
= + = ) ( and
10
10
) (
+
=
s
s G in Fig. 5.1.2. Then, using equation (5.1.1), the
output is given by:
) ( 10 ) 10 (
10
) (
) ( 10 ) 10 (
) ( 10
) ( ) (
I P I P
I P
K s K s s
s
s W
K s K s s
K s K
s U s Y
+ + +

+ + +
+
= (5.1.2)

From equation (5.1.2) it can be seen that the transfer functions for both inputs have the same
denominator and so they are characterized by the same poles.

97
First let 0 ) (
8
) ( = = s W and
s
s U . Using the final value theorem:
8
) ( 10 ) 10 (
) ( 10 8
lim ) ( lim ) ( lim
0 0
=
+ + +
+
= =

I P
I P
s s t
K s K s s
K s K
s
s s sY t y (5.1.3)
This means that for a step input with magnitude 8, the output follows the input exactly in the steady
state.
Secondly, let
s
s W and s U
8
) ( 0 ) ( = = . Using the final value theorem:
0
) ( 10 ) 10 (
10 8
lim ) ( lim ) ( lim
0 0
=
+ + +
= =

I P
s s t
K s K s s
s
s
s s sY t y (5.1.4)
This means that the output is not affected by the disturbance in the steady state. Now, let K
P
= 1 and
K
I
= 10 in equation (5.1.2). Apply the input
s
s U
8
) ( = at t = 1 sec and the disturbance
s
s W
8
) ( = at t
= 3 sec. The output ) (s Y is shown in Fig. 5.1.3. Note that the dynamic response for the input and the
disturbance is the same.

Fig. 5.1.3. Dynamic response of input and disturbance.

These findings motivate the use of a load torque observer with compensation current feedback, which
in effect adds another loop that has a time constant less than the speed loops time constant, but greater
than the current loops time constant. The torque loop will be used to compensate for the load torque
disturbance. First, a load torque observer for the RSM with constant field current control (CFCC) is
investigated. Then a load torque observer for RSM or PMA RSM with constant current angle control
(CCAC) is derived.

98
5.2. Load torque observer for RSM with CFCC

In this subsection, it is shown that it is possible to improve the dynamic response of a RSM which uses
CFCC, with a load torque observer and compensation current feedback. The load torque observer in
this section is a reduced state observer, because it only observes the state that is not measured
directly, namely the load torque. A simulation example is used to show that the concept works.

To construct a model for the RSM with CFCC, the torque equation of the RSM and differential
equation that describes the mechanical plant are used:

m eq
m
eq load d q q d em
B
dt
d
J T i i
p
T

+ + = = ) (
2
3
(5.2.1)
Flux linkage can be expressed as the product of linearized inductance (see Fig. 2.2.1) and current.
Equation (5.2.1) is then written as:

m eq
m
eq load q d q d
m eq
m
eq load d q q q d d em
B
dt
d
J T i i L L
p
B
dt
d
J T i i L i i L
p
T

+ + =
+ + = =
) (
2
3
) (
2
3
(5.2.2)
For CFCC, the D-axis current is kept constant and this implies that the D-axis self inductance is ideally
constant. The Q-axis self inductance is also constant since the Q-axis flux linkage does not go into
saturation (see Fig. 3.8.4). Thus for CFCC, equation (5.2.2) can be written as

d q d T
m eq
m
eq load q T em
i L L
p
K with
B
dt
d
J T i K T
) (
2
3
=
+ + = =

(5.2.3)

It is motivated by [A13], [A14] and [A15] that in order to observe the load torque, a load torque
observer such as the one shown in Fig. 5.2.1, be used.
I ref V ref I Tem
Speed
Tem
Te -T load
T load
Compensation current
speed ref Y(s)
T load
D1(s)
Speed PI
G2(s)
Mech Model
LPF
Low pass filter
1/G2(s)
Inv Mech Model
1/Kt
Kt
Kt G1(s)
Electric Model
D2(s)
Current control

Fig. 5.2.1. Control system with load torque observer and compensation current feedback.

99

In Fig. 5.2.1, the speed and current signals are measured, and based on these measurements an
approximation for the applied load torque is achieved by an observer. This observed load torque is
used to calculate a compensation current for the load torque disturbance. The compensation current is
calculated using equation (5.2.3) in the steady state and neglecting the friction coefficient, and is given
as:
T
load
q
K
T
i = (5.2.4)
Equation (5.2.4) gives the correct reference current to compensate for the load torque in the steady
state. An example for this scheme follows [S5]. The model for the RSM is given by the equations
below, assuming that mutual inductances and the variation of flux linkages due to the changing rotor
position are negligible. The derivatives of the flux linkages are expressed as the product of
instantaneous self inductances and the derivatives of the currents, while the flux linkages are expressed
as the product of linearized self inductances and currents as explained previously (see Fig. 2.2.1).
q q m
d
d d s d
i L p
dt
di
L i r v + =
'
(5.2.E1)
d d m
q
q q s q
i L p
dt
di
L i r v + + =
'
(5.2.E2)
m eq
m
eq load d q q d em
B
dt
d
J T i i
p
T

+ + = = ) (
2
3
(5.2.E3)

These machine equations can be represented in block diagram form as shown in Fig. 5.2.E1. Here it is
assumed that the speed voltage terms,
q q m
i L p in equation (5.2.E1) and
d d m
i L p in equation
(5.2.E2), are cancelled by a decoupling procedure, as explained in subsection 3.2. The decoupling
procedure is not shown here. A subsystem for this block diagram is created using the inputs and
outputs as shown. The subsystem is called Machine and is used in the current and speed loops as
shown in Fig. 5.2.E2.

Tem
3
Iq
2
Id
1
Wm
1
Jeq.s+Beq
1
Lq.s+Rs
1
Ld.s+Rs
Product
3*p/2*(Ld - Lq)
3
Vq
2
T l oad
1
Vd

Fig. 5.2.E1. RSM machine model referred to as Machine in the simulation block diagram.


100
The complete control system with PI speed controller, P current controllers and a load torque observer
is shown in Fig. 5.2.E2. The saturation block that follows the output of the PI speed controller
prevents the current reference command from exceeding the rated current. The integral term of the PI
speed controller is also limited to prevent integrator windup (see subsection 3.11).
i q
1
1/100s+1
speed LPF
Wm scope
Wm ref
Torque ref
T l oad observe
Saturati on
Vd
T load
Vq
Wm
Id
Iq
Machi ne
16
Kq
32
Kd
Iq scope
1
s
Id scope
Id ref
1/(Kt)
Kt
Beq
Jeq
3
4.5
du/dt
1
1/230s+1
torque LPF

Fig. 5.2.E2. Simulation block diagram of RSM with CFCC and load torque observer.

The parameters for the machine are chosen as follows:
100 =
d
I A gives a constant D-axis inductance, mH L
d
10 = .
5 =
q
L mH
15 =
s
r m
3 = p pole pairs
2 =
eq
J kg.m
2

1 . 0 =
eq
B Nm/(rad/sec)
Then, from equation (5.2.3), ( ) 25 . 2 100 005 . 0 01 . 0
2
3
3 = =
T
K Nm/A.

The rated current is chosen to be A I
rated
283 200 2 = = , and therefore the current reference is
limited to this value. Using the formulas for the current controller gains, equations (3.6.7) and (3.6.8):

16
32
=
=
q
d
K
K
(5.2.E4)
The PI speed controller is taken from subsection 3.10 and is given by:

s
s
C
5 . 4 3 +
= (5.2.E5)

101
Let the observer LPF be a first order LPF:

1
1
) (
+
=
s
s H (5.2.E6)
The response time of the torque loop is chosen by setting the cutoff frequency of the LPF. If the input
to the LPF is a step function, ) ( ) ( t u t x = , the Laplace representation is given by:

s
s X
1
) ( = (5.2.E7)
The output of this LPF in the time domain is then given by:
) ( 1 ) ( t u e t y
t
|
|

\
|
=

(5.2.E8)
Let the output signals value reach 90% of the input value, within 10 ms (this is the chosen response
time for the torque loop):

230
1
9 . 0 1
10
=
=

m
e
(5.2.E9)
The torque loop LPF is therefore designed to have a cutoff frequency of 230 rad/sec and this cutoff
frequency corresponds to a response time of 10 ms. There is a tradeoff between the response time of
the torque loop and the noise in the system: if a fast response time is chosen there will be more noise,
because the filter will then have a large bandwidth, allowing more noise to enter. There is also a lower
limit on the torque loops response time, namely the response time of the current loop. If a higher
order LPF is chosen, the noise might be reduced, but the derivation for the cutoff frequency, equation
(5.2.E6) through to (5.2.E9), will of course be different. Noise is not included in this simulation.

The simulation is performed first without compensation from the load torque observer and then with
compensation, in both cases with a speed reference of 500 rad/sec at t = 0 sec and a load torque step of
300 Nm at t = 5 sec. The speed reaches the reference of 500 rad/sec in a few seconds and then
maintains the speed. The load torque applied at t = 5s causes a change in speed, which is corrected
slowly in the case of no compensation and fast in the case of compensation. The following graphs
show the performance of the control system. Note that these graphs are the actual windows of the
scopes shown in Fig. 5.2.E2; the name of each scope is shown in the top left corner.


102

Fig. 5.2.E3. Speed response without compensation (left) and with compensation (right).


Fig. 5.2.E4. Observed load torque with time range 0 to 10 sec (left) and at t = 5 sec (right).

Fig. 5.2.E3 shows that the dynamic response of the drive system is increased using the compensation
current from the load torque observer. Fig. 5.3.E4 shows that the load torque is observed correctly
(left) and the torque loop has the response time designed for (right). In this simulation the parameters
are fixed, there is no noise, and the controllers are continuous, which are not realistic circumstances.

The example above shows that it is possible to improve the dynamic response of a RSM under
constant field current control (CFCC) using a load torque observer. Now, the next question is whether
it is possible to derive a load torque observer for the RSM (and PMA RSM) under constant current
angle control (CCAC). This is an important question, because CCAC is a much better method to
control the currents of the PMA RSM compared to CFCC (in terms of efficiency), and the main goal is
to increase the dynamic performance of the PMA RSM.



103
5.3. Reduced state observer for RSM (or PMA RSM) with CCAC

The scheme for increasing the dynamic performance introduced in subsection 5.2 is adjusted here for
the RSM (or PMA RSM) that uses CCAC. For the RSM with CCAC, equation (5.2.E3) can be written
in a suitable form as shown in equation (5.3.1). The reason for this form is explained afterwards.
d q d d
m eq
m
eq load s d
m eq
m
eq load s q d
m eq
m
eq load q d q d em
L of function a L L
p
L f with
B
dt
d
J T i L f
B
dt
d
J T i L L
p
B
dt
d
J T i i L L
p
T
2
) 2 sin(
) (
2
3
) (
) (
2
) 2 sin(
) (
2
3
) (
2
3
1
2
1
2

=
+ + =
+ + =
+ + = =
(5.3.1)

For CCAC, both D-axis current and Q-axis current vary as the load torque or speed reference
demands, but the ratio between the currents (or the current angle) stays constant. The problem is that
as the D-axis current varies, so does the D-axis inductance. In reality, the Q-axis inductance also
changes with variation in Q-axis current, but it is nearly constant and can be taken as constant without
making a significant error. A method can be adopted where ) (
1 d
L f , in equation (5.3.1), is calculated
real time with knowledge from finite element analysis (see Appendix E, Fig. E.12). In other words, it
would be possible to use a lookup table (LUT) that gives
d
L for any D-axis current and then to
calculate ) (
1 d
L f real time. A possible adaptation to the load torque compensation method described
in subsection 5.2, is therefore to replace
T
K with ) (
1 d
L f and replace
q
i with
2
s
i .

To avoid calculation of ) (
1 d
L f , we can use finite element analysis (FEA) to calculate the produced
torque, for a constant current angle, as a function of stator current
s
i (not
2
s
i ), and to store this result in
a LUT. Therefore, the method suggested is to have
m eq
m
eq laod s s em
B
dt
d
J T i i f T

+ + = = ) (
2
and
to store ) (
2 s
i f in a LUT. The same FEA result can be used to generate the inverse function
em
s
em
T
i
T f =

) (
1
2
. Then the load torque can be observed and the compensation current can be
calculated using the scheme in subsection 5.2 with a few alterations: CCAC replaces CFCC,
) (
2 s
i f replaces
T
K , ) (
1
2 em
T f

replaces
T
K
1
and
s
i replaces
q
i .


104
At this point, the new method will be no different for the PMA RSM than for the normal RSM (the
actual LUTs will be different of course). FEA results for the PMA RSM (see Appendix E) will be
used in LUTs for the proposed load torque estimator and current compensator. From Fig. 2.5.4, the
constant current angle is chosen as = 54 , because this angle produces the maximum torque per
current. Data obtained from FEA is shown in the form of graphs in Fig. 5.3.1. The data is used to
create LUTs that can be used in simulation and practical experiments.

Fig. 5.3.1. Torque as a function of stator current (left) and stator current as a function of torque (right)
for a constant current angle of = 54 .

The simulation model for the PMA RSM with CCAC from Chapter 3 is used to test the reduced state
observer with compensation current feedback [S6]. The speed controller, constant current angle
current controller and machine model are placed in a subsystem called Speed, CCAC, PMA RSM.
The inputs to this subsystem are the measured speed, speed reference, load torque and compensation
current. The outputs of the subsystem are the stator current, current angle, electrical speed and
generated torque. The current angle is needed to indicate the direction of the stator current vector: if
the current angle is less than 90, the machine is in motoring mode and the stator current is positive, if
the current angle is greater than 90, the machine is in generating mode and the stator current must be
negative. The measured speed and stator current are used for the estimation of the load torque. The
produced torque output
em
T of the subsystem can be used for comparison with the estimated torque
LUT T measured s
K i , but is not needed in the scheme. The complete system is shown in Fig. 5.3.2.

50 100 150 200 250 300
0
200
400
600
800
Torque vs. Stator current
Stator current [A peak]
T
o
r
q
u
e

[
N
m
]
0 200 400 600 800 1000
0
100
200
300
Stator current vs. Torque
S
t
a
t
o
r

c
u
r
r
e
n
t

[
A

p
e
a
k
]
Torque [Nm]

105
di fferenti ator
Is
Iangle
Is.
motor / gen
We ref
We fi l tered - We ref
We fi l tered
We actual
z
1
Uni t Del ay
Tl oad Tem
TL^ Is comp
TL^ / Is comp
T l oad observe
We in
We ref
T load
Is comp
Is out
Iangle out
We out
Tem
Speed, CCAC, PMA RSM
T load
Load system
den(z)
1.0e-003 * [ 0.3889 0.7779 0.3889]
LPF 2nd order
@ 200 rad/sec
num(z)
den(z)
LPF 1st order
@ 100 rad/sec
num(z)
den(z)
LPF 1st order
@ 100 rad/sec
Is peak l i mi t Is compensate
Is Tem
Is / Tem
Is
Jeq
Accel
1/T
Noi se source

Fig. 5.3.2. Simulation block diagram of load torque observer (reduced state observer) with
compensation current feedback.

In Fig. 5.3.2, the blocks TL^ / Is comp and Is / Tem are LUTs that incorporate the graphs of Fig.
5.3.1 with some modifications, because the LUTs are only one-sided. The compensation current is
limited to the rated current of the PMA RSM, namely A I 283
lim
=
+
and A I 283
lim
=

.

As shown in Fig. 4.11.3, the practical speed measurement contains noise. Noise sources include the
PWM switching noise and quantization noise within the DSP. Noise can be reduced by a digital LPF,
the order of which depends on the severity of the noise. It is found by practical measurement that a
first order LPF with a cutoff frequency at 100 rad/sec is sufficient (see Fig. 4.11.3). For the simulation
diagram shown in Fig. 5.3.2, a first order discrete LPF with a cutoff frequency at 100 rad/sec is used
for the speed measurement. The LPF is placed in the feedback path of the speed loop.

In Fig. 5.2.E2, the proposed observer differentiates the unfiltered speed measurement. The
differentiated speed measurement is multiplied by the inertia, and the friction term is added to give
load em
T T . The unfiltered current measurement is multiplied by the torque coefficient to give the
produced torque
em
T . Then,
load em
T T is subtracted from
em
T and filtered, to give the observed load
torque
load
T

.

106

If the unfiltered speed measurement is differentiated, the result is very noisy. The current
measurement also contains noise, which means it is not suitable to use as the index in a lookup table.
In Fig. 5.3.2, the filtered speed measurement (also used for feedback to the speed controller) is
differentiated and the filtered current (also a LPF with a cutoff frequency at 100 rad/sec) is used as the
index to the lookup table to obtain
em
T . Finally, a second order LPF with the cutoff frequency at 200
rad/sec is used to filter the observed load torque. The filtered observed torque is used as the index to a
lookup table that gives the correct compensation current.

As a first test, the noise source is set to zero, the speed reference is set to
sec / 377
60
2
1200 3
60
2
rad pv
mech e
= =

, where
mech
v is the mechanical speed in rev/min, the
acceleration limiter is set to
2
sec 5
377 rad
dt
d
e
=

, and a positive torque step of 700 Nm is applied at

sec 10 = t . The speed response, current response, observed load torque and compensation current,
without current compensation feedback using the load torque observer, is shown in Fig. 5.3.3. The
experiment is repeated with current compensation feedback and the simulation results are shown in
Fig. 5.3.4.

Note that the graphs shown in Fig. 5.3.3 and Fig. 5.3.4 are the actual windows of the scopes shown in
Fig. 5.3.2; the name of the scope is shown in the top left corner. The table below applies for Fig. 5.3.3.
and Fig. 5.3.4.

Position Description y-axis range x-axis range
Top, left Electrical speed response 0 to 500 rad/sec 0 to 15 sec
Top, right Observed load torque -500 to 2000 Nm 0 to 15 sec
Bottom, left Stator current -50 to 300 A 0 to 15 sec
Bottom, right Compensation current -150 to 300 A 0 to 15 sec

107

Fig. 5.3.3. Simulation results without compensation current feedback.

From Fig. 5.3.3, note that the observed load torque is incorrect during the acceleration period
sec 5 0 < < t . Compare the observed load torque in Fig. 5.3.3 with the observed load torque from
the example in Fig. 5.3.E2. In the example the torque equation is linear (equation 5.2.3) and the torque
coefficient is constant. The large error in observed torque in Fig. 5.3.3 may be due to the non-linear
torque equation (equation 5.3.1) where the torque coefficient is not constant. The acceleration limiter
plays an important part here: if the acceleration limit is set to a high value, the error in observed
torque is large. It is therefore important that when this reduced state observer is used, the acceleration
limiter should be set to an appropriate value.


108

Fig. 5.3.4. Simulation results with compensation current feedback.

Although the observed torque is not correct during speed changes, Fig. 5.3.4 shows that the torque
observer with current compensation improves the dynamic response of the system, because the speed
stays close to the speed reference even when the load torque changes suddenly.

The system can be tested for robustness against noise by introducing noise on the speed measurement.
The practically measured SNR is 13dB as shown in Fig. 4.11.3. It is anticipated that the performance
might not be so good due to the fact that the derivative of the noisy speed is taken to compute the load
torque. The simulation results with added noise to the speed measurement are shown in Appendix H.

The reduced state observer is implemented practically, and practical results with and without
compensation current feedback are shown next. In these results, it can be seen that the applied load
torque is not a step function, but may be approximated by a saturated ramp function. Note that in the
practical results shown, the electrical speed error is
ref e e err
= . Therefore, if the load torque
falls away suddenly, as in Fig. 5.3.P1, the speed (and speed error) rises momentarily before it is
corrected. Simulation results that can be compared to the practical results are given in Appendix H. In

109
this simulation, the applied load torque is not a step function, but rather a saturated ramp function.
These simulation results are very similar to the practical results, indicating that the simulation model is
accurate.

To summarize, the load torque observer with current compensation improves the dynamic response of
the drive system and decreases the effect of load torque variation. The load torque observer structure
introduced uses the filtered speed measurement and filtered current measurement, along with the
torque coefficient (a result from FEA), to estimate the load torque. The estimated load torque and the
inverse torque coefficient (also a result from FEA) are used to calculate the compensation current.
Feedback of the compensation current reduces the effect of load torque variation.


110
Practical results for reduced state observer WITHOUT current compensation.

Experiment A: Load torque from 700 Nm to 0 Nm


Fig. 5.3.P1. Electrical speed error
ref e e
(top) and observed torque (bottom).


Fig. 5.3.P2. Stator current (top) and compensation current (bottom).
Electrical speed error
Overshoot:
1 divisions
2 volts / division
20 rad/sec / volt
Therefore, 40 rad/sec
Settling time:
4 seconds
Observed load torque
Change in magnitude:
-1 division
2 volts / division
350 Nm / volt
Therefore, -700 Nm
Settling time:
1 second

Stator current
Change in magnitude:
-1 division
2 volts / division
100 A / volt
Therefore, -200 A
Settling time:
1 second
Compensation current
Change in magnitude:
-1 division
2 volts / division
100 A / volt
Therefore, -200 A
Settling time:
1 second


111
Experiment B: Load torque from 0 Nm to 700 Nm


Fig. 5.3.P3. Electrical speed error
ref e e
(top) and observed torque (bottom).


Fig. 5.3.P4. Stator current (top) and compensation current (bottom).
Electrical speed error
Overshoot:
-0.5 divisions
2 volts / division
20 rad/sec / volt
Therefore, -20 rad/sec
Settling time:
4 seconds
Observed load torque
Change in magnitude:
1 division
2 volts / division
350 Nm / volt
Therefore, 700 Nm
Settling time:
4 seconds

Stator current
Change in magnitude:
1 division
2 volts / division
100 A / volt
Therefore, 200 A
Settling time:
4 seconds
Compensation current
Change in magnitude:
1 division
2 volts / division
100 A / volt
Therefore, 200 A
Settling time:
4 seconds


112
Practical results for reduced state observer WITH current compensation

Experiment A: Load torque from 700 Nm to 0 Nm


Fig. 5.3.P5. Electrical speed error
ref e e
(top) and observed torque (bottom).


Fig. 5.3.P6. Stator current (top) and compensation current (bottom).
Stator current
Change in magnitude:
-1 division
2 volts / division
100 A / volt
Therefore, -200 A
Settling time:
1 second
Compensation current
Change in magnitude:
-1 division
2 volts / division
100 A / volt
Therefore, -200 A
Settling time:
1 second

Electrical speed error
Overshoot:
0.4 division
2 volts / division
20 rad/sec / volt
Therefore, 16 rad/sec
Settling time:
1 second
Observed load torque
Change in magnitude:
-1 division
2 volts / division
350 Nm / volt
Therefore, -700 Nm
Settling time:
1 second

113
Experiment B: Load torque from 0 Nm to 700 Nm

Fig. 5.3.P7. Electrical speed error
ref e e
(top) and observed torque (bottom).


Fig. 5.3.P8. Stator current (top) and compensation current (bottom).

Electrical speed error
Overshoot:
0 divisions
2 volts / division
20 rad/sec / volt
Therefore, 0 rad/sec
Settling time:
1 second
Observed load torque
Change in magnitude:
1 division
2 volts / division
350 Nm / volt
Therefore, 700 Nm
Settling time:
4 seconds

Stator current
Change in magnitude:
1 division
2 volts / division
100 A / volt
Therefore, 200 A
Settling time:
4 seconds
Compensation current
Change in magnitude:
1 division
2 volts / division
100 A / volt
Therefore, 200 A
Settling time:
4 seconds


114
5.4. Full state observer for RSM (or PMA RSM) with CCAC

The observer structure introduced in subsection 5.3 can be viewed as a reduced state observer if it is
compared to a full state observer that estimates both speed and load torque. Observers are explained
thoroughly in [B7, p687], [B9, p.541] and [A16]. The full state observer has several advantages over
the reduced state observer in this case: there is no speed derivative, the plant parameters do not have
to be very accurate since they are corrected automatically with feedback, and FEA results for the
relationship between torque and stator current are not needed to compute the compensation current.
The derivation of the full state observer structure is given in Appendix G and the result is shown in
Fig. 5.4.1.
Tem
Tload^
DSP
We^
Kt
Torque
coefficient
(result from FEA)
L2
J/p*L1
Beq/p
p/Jeq
K Ts
z-1
Discrete-Time
Integrator
K Ts
z-1
Discrete-Time
Integrator
2
We measured
1
Is measured

Fig. 5.4.1. Full state observer.

The full state observer shown in Fig. 5.4.1 estimates the electrical speed and the load torque. The
operation of this observer can be explained as follows: the difference in observed speed and measured
speed must be due to incorrect plant parameters or load torque variation, assuming that the torque
coefficient, Kt, is accurate. The two feedback paths through L1 and L2 work simultaneously to give
the correct plant model and load torque.

Equation (G.9) shows that the gain L2 is negative. This seems correct: if the measured speed is
greater than the observed speed, a likely cause is a decrease in load torque. Since the estimated load
torque is subtracted at the input to the plant, this double negative sign could be replaced by a positive
sign. Then the speed error is actually the input to a PI controller: the proportional gain given by
1
L
p
J

and the integral gain given by
2
L . A new block diagram for the full state observer is shown in Fig.
5.4.2.


115
Actual machine
DSP
= Torque load in steady state
3
Torque load ^
2
We^
1
We
measured
Torque load
Kt
Torque
coefficient
(result from FEA)
Kt
Torque
coefficient
(actual)
p
Jeq.s+Beq
Real plant
Kp
Ki
K Ts
z-1
Discrete-Time
Integrator
G(z)
Discrete model
of real plant
2
Is measured
1
Is
(actual)

Fig. 5.4.2. Full state observer.

Inspecting Fig. 5.4.2, it is noted that after a certain time (which can be chosen), the input to the PI
controller is zero and at that stage the output of the PI controller must be equal to the load torque. This
structure gives one the freedom to choose a response time for the estimated load torque. To design the
PI controller, the block diagram for the DSP section of Fig. 5.4.2 is drawn in a slightly different way,
as shown below.
2
Torque load ^
1
We^
Kt
Torque constant
(result from FEA)
Kp
Ki
K Ts
z-1
Discrete-Time
Integrator
G(z)
Discrete model
of real plant
2
We
measured
1
Is measured

Fig. 5.4.3. PI controller design setup for load torque observer.

Because of the negative sign in the loop, the block diagram of Fig. 5.4.3 can be reduced to that in Fig.
5.4.4, if the measured current is taken as zero. The current is taken as zero to design the PI controller,
because in this loop it is a disturbance input and it will be compensated for within the settling time that
the PI controller is designed for.

116
2
Torque load ^
1
We^
-G(z)
negative
Discrete model
of real plant
Kp
Ki
K Ts
z-1
Discrete-Time
Integrator
1
We
measured

Fig. 5.4.4. Simplified PI controller design setup for load torque observer.

The speed measurement, We measured, is very likely to contain noise. Another advantage of this
observer structure is that the LPF for the speed can be included in the design of the PI controller. In
this way, stability can be guaranteed even if a high order LPF is used. Note that the controller does not
necessarily have to be a PI controller. Any controller structure can be chosen as long as there is an
integrator in the loop so that the input to the controller is eventually zero and the output is eventually
equal to the load torque. As shown in Fig. 5.4.4, the observed load torque should be taken from the
integrator branch of the controller. Alternatively, the observed load torque can be obtained by low
pass filtering the output of the controller.

The full state observer with LPF for the speed measurement is shown in Fig. 5.4.5. The controller can
now be designed while taking into account the LPF.
Actual machine
DSP
3
Torque load ^
2
We^
1
We
measured
Torque load
Kt
Torque
coefficient
(result from FEA)
Kt
Torque
coefficient
(actual)
p
Jeq.s+Beq
Real plant
LPF
LPF
LPF
C(z)
G(z)
Discrete model
of real plant
2
Is measured
1
Is
(actual)

Fig. 5.4.5. Full state observer with LPFs for speed and observed torque.

The full state observer structure discussed so far, gives the observed torque within a certain time that
can be chosen. This observed load torque can be used to compute the compensation current for the
load torque, using the inverse of the torque coefficient. For this FEA can be used to obtain the torque

117
coefficient and its inverse for any stator current. A revelation is that in the full state observer structure
it is not necessary to have this information to compute the compensation current. The reason for this is
explained next.

Consider again the differential equation that describes the produced torque and the mechanical system:
m eq
m
eq L em
B
dt
d
J T T

+ + = (5.4.1)
For constant current angle control (CCAC), equation (5.4.1) can be written in terms of the torque
coefficient as follows:

m eq
m
eq L T s
B
dt
d
J T K i

+ + = (5.4.2)
The stator current can be written as the superposition of a current that provides for the speed, and
another current that provides for the load torque:

L T sL
m eq
m
eq T s
m eq
m
eq L T sL s
T K i and
B
dt
d
J K i with
B
dt
d
J T K i i
=
+ =
+ + = +
,
, ) (

(5.4.3)

Now, consider again the full state observer, this time with the current represented as the superposition
of a torque current and a speed current:
Actual machine
DSP
T load
3
IsL*Kt = T load ^
2
We^
1
We
measured
Kt
Torque
coefficient
(result from FEA)
Kt
Kt
Torque
coefficient
(actual)
p
Jeq.s+Beq
Real plant
LPF
LPF
LPF
C(z)
G(z)
Discrete model
of real plant
3
IsL
(actual)
2
Is=IsL+Isw
(measured)
1
Is=IsL+Isw
(actual)

Fig. 5.4.6. Full state observer with torque current and speed current.


118
By simply omitting the torque coefficient at the input to the plant model, the feedback term is scaled
accordingly. Therefore the compensation current is available at the output of the PI controller and
information about the torque coefficient is not necessary. This is shown in Fig. 5.4.7.
Actual machine
DSP
T load
3
IsL = Is compensate
2
We^
1
We
measured
Kt
Kt
Torque
coefficient
(actual)
p
Jeq.s+Beq
Real plant
LPF
LPF
LPF
C(z)
G(z)
Discrete model
of real plant
3
IsL
(actual)
2
Is=IsL+Isw
(measured)
1
Is=IsL+Isw
(actual)

Fig. 5.4.7. Full state observer with current compensation output.

The block diagram used to design the controller that gives the load torque compensation, taking into
account the LPF for the speed measurement, is shown in Fig. 5.4.8. The LPF is chosen to be a 1
st

order Butterworth filter. The filter can be designed using Matlabs digital filter design tool or the
function butter. The discrete model of the mechanical plant is obtained using the bilinear
transformation. The LPF and discrete plant are in series and can be combined to form the complete
plant model. Note the negative sign for the discrete plant as explained for Fig. 5.4.4. Matlabs root
locus design tool can be used to design a stable controller.

2
Is compensate
1
We^
-G(z)
negative
discrete model
of real plant
H(z)
LPF
H(z)
LPF
C(z)
Controller
1
We

Fig. 5.4.8. Full state observer with LPF and current compensation output.


119
For the design of the controller, the plant parameters are given the following values and it is reassuring
to know that these parameters need not be perfect as they are automatically corrected by feedback:
Beq = 0.1;
Jeq = 2.75;
p = 3;

Create the continuous model for the plant:
G = tf([-p],[Jeq Beq])
Transfer function:
-3
------------
2.75 s + 0.1

Discretize the plant model using a bilinear transform. The sampling frequency is taken as 5000 Hz.
The Matlab function c2d is used with function parameters being the continuous plant, sampling time
and bilinear parameter:
Gz = c2d(G,1/5000,tustin)
Transfer function:
-0.0001091 z - 0.0001091
------------------------
z - 1
Sampling time: 0.0002

Design the 1
st
order Butterworth LPF with a cutoff frequency of 100 rad/sec:
[B,A] = butter(1,100/(2*pi*5000/2),low);
Hz = tf(B,A,1/5000)
Transfer function:
0.009901 z + 0.009901
---------------------
z - 0.9802
Sampling time: 0.0002

To implement the LPF in a simulation model, a discrete transfer function block can be used. The
block requires an array that describes the numerator and another array that describes the denominator.
These arrays are given by B and A, with reference to the Matlab code above.

Combine the discrete plant and LPF so that it can be imported into rltool:
plant = Gz*Hz;
rltool

The root loci of Hz, Gz and the combination are shown in Fig. 5.4.9. Note that to display the root
locus of the plant correctly, the controller gain should be negative. In Fig. 5.4.9, the root locus for the
discrete plant shows a single pole at 1 (this means that there is already an integrator in the loop) and a
single zero at -1. The root locus for the 1
st
order Butterworth LPF has a single pole near 1 and a single
zero at -1. The combination of these poles and zeros shows that stable closed loop poles result for any

120
negative controller gain. Since there is already an integrator in the loop due to the plant pole, the
controller for the estimated torque can be only a negative constant. If that is the case however, the
closed loop poles will be far away from the real axis, which means a highly undamped response. In
order to get the closed loop poles closer to the real axis, i.e. to get a damped response, the controller
can take on the structure of a lead-lag filter. A lead-lag filter has a zero at a lower frequency compared
to the pole. This means that as frequency increases from 0 Hz, there is phase lead due to the zero and
at higher frequencies, a phase lag due to the pole.


Fig. 5.4.9. Root locus for discrete plant (top left), LPF (top right) and combination (bottom)



121
The proposed controller structure is shown on the root locus with the combined plant (LPF and
mechanical plant) in Fig. 5.4.10. It is convenient to speak of the load torque observer with
compensation current feedback as the torque loop. The response time of the torque loop should be
greater than the current loop response time and less than the speed loop response time. The current
controllers are designed so that the response time of the current loops varies according to the
instantaneous inductances (see chapter 3). The response times for the current loops are always less
than 10 ms. The response time of the speed loop is chosen to be 2 s. A suitable response time for the
torque loop is therefore 20ms.

The proposed controller for Fig. 5.4.7 is therefore a lead-lag filter. The gain of the controller is
adjusted so that the settling time to a step input is 20ms. The controller is shown with the root locus of
the combined plant (LPF and negative mechanical plant) in Fig. 5.4.10.

Fig. 5.4.10. Complete root locus of torque loop: LPF, negative mechanical plant and load torque
observer controller (left). The region around 1 (right).

The controller can now be exported to the workspace. This is given by:
C
Zero/pole/gain:
-12089.2034 (z-0.9695)
----------------------
(z-0.2028)
Sampling time: 0.0002



122
To create a simulation block for the controller, a discrete transfer function block can be used. The
practical implementation of the LPF, discrete plant and controller in the control program is of course
different: it should be implemented using a Direct Form I or II structure. Using the accurate
simulation model of the PMA RSM with CCAC and speed control (chapter 3), the full state observer
with load torque compensation current can be tested [S7]. The machine model with current and speed
control is grouped together as a subsystem, in the same way as the subsystem in Fig. 5.3.2. The
complete system with full state observer and current compensation is shown in Fig. 5.4.11.

we
we^
Is
Iangle
Is.
motor / gen
We^
We ref
We filtered
We actual
Tload
Tem
T load observe
We in
We ref
T load
Is comp
We out
Is out
Iangle out
Tem
Speed, CCAC, PMA RSM
T load
Load system
den(z)
1.0e-003 * [ 0.3889 0.7779 0.3889]
LPF 2nd order
@ 200 rad/sec
num(z)
den(z)
LPF 1st order
@ 100 rad/sec
num(z)
den(z)
LPF 1st order
@ 100 rad/sec
Is.
Is peak limit Is compensate
Is Tem
Is / Tem
den(z)
0.0001091*[ 1.00 1.00]
G(z)
num(z)
z-0.2028
C(z)
lead lag filter
Band-Limited
White Noise
Accel

Fig. 5.4.11. Full state observer with compensation current feedback simulation block diagram.

In Fig. 5.4.11, note that a first order LPF is used for the speed measurement and a 2
nd
order LPF is
used for the compensation current. The 2
nd
order LPF is obtained in exactly the same manner as the 1
st

order LPF and it is not shown here. To view the observed load torque, a result from FEA is used in the
block Is / Tem, but this is not necessary to compute the compensation current.

As a first test, the noise power in the measured speed is set to zero, the speed reference is set to
sec
377
rad
e
= , the acceleration limit is set to
2
sec 5
377 rad
dt
d
e
=

, and a positive torque step of 700

Nm is applied at sec 10 = t . The speed response, current response, observed load torque and
compensation current, without compensation feedback using the load torque observer, is shown in Fig.
5.4.12. These results can be compared to the results for the reduced state observer in Fig. 5.3.4. The
simulation results for the full state observer, this time with the compensation, is shown in Fig. 5.4.13.
These results can be compared to the results for the reduced state observer in Fig. 5.3.5.


123
Position Description y-axis range x-axis range
Top, left Electrical speed response 0 to 500 rad/sec 0 to 15 sec
Top, right Observed load torque -500 to 2000 Nm 0 to 15 sec
Bottom, left Stator current -50 to 300 A 0 to 15 sec
Bottom, right Compensation current -150 to 300 A 0 to 15 sec


Fig. 5.4.12. Simulation results without compensation current feedback.

It can be seen from Fig. 5.4.12 that the observed load torque during the acceleration period
sec 5 0 < < t is nearly perfect. The observed load torque from the full state observer is much more
accurate during the acceleration period than the reduced state observer. Note that there is still a
dependence on the acceleration limiter and it is still important to keep the level of acceleration low.
The torque loop is now closed so that the compensation current is applied. The results are shown in
Fig. 5.4.13.


124

Fig. 5.4.13. Simulation results with compensation current feedback.

The system with the full state observer can be tested for robustness against noise by introducing a
small amount of noise on the speed measurement. The simulation results with noise added to the
speed measurement are shown in Appendix H. The full state observer is implemented practically and
practical results with and without compensation current feedback are shown next. Simulation results
that can be compared to the practical results are shown in Appendix H. The practical- and simulation
results are very similar, indicating that the simulation model is accurate.

125
Practical results for full state observer WITHOUT current compensation.

Experiment A: Load torque from 700 Nm to 0 Nm

Fig. 5.4.P1. Electrical speed error
ref e e
(top) and observed torque (bottom).


Fig. 5.4.P2. Stator current (top) and compensation current (bottom).
Electrical speed error
Overshoot:
1 division
2 volts / division
20 rad/sec / volt
Therefore, 40 rad/sec
Settling time:
4 seconds
Observed torque
Change in magnitude:
-1 division
2 volts / division
350 Nm / volt
Therefore, -700 Nm
Settling time:
1 second
Stator current
Change in magnitude:
-1 division
2 volts / division
100 A / volt
Therefore, -200 A
Settling time:
1 second
Compensation current
Change in magnitude:
-1 division
2 volts / division
100 A / volt
Therefore, -200 A
Settling time:
1 second


126
Experiment B: Load torque from 0 Nm to 700 Nm



Fig. 5.4.P3. Electrical speed error
ref e e
(top) and observed torque (bottom).


Fig. 5.4.P4. Stator current (top) and compensation current (bottom).
Stator current
Change in magnitude:
1 division
2 volts / division
100 A / volt
Therefore, 200 A
Settling time:
4 seconds
Compensation current
Change in magnitude:
1 division
2 volts / division
100 A / volt
Therefore, 200 A
Settling time:
4 seconds

Electrical speed error
Overshoot:
-0.5 divisions
2 volts / division
20 rad/sec / volt
Therefore, -20 rad/sec
Settling time:
4 seconds
Observed torque
Change in magnitude:
1 division
2 volts / division
350 Nm / volt
Therefore, 700 Nm
Settling time:
4 seconds


127
Practical results for full state observer WITH current compensation

Experiment A: Load torque from 700 Nm to 0 Nm


Fig. 5.4.P5. Electrical speed error
ref e e
(top) and observed torque (bottom).


Fig. 5.4.P6. Stator current (top) and compensation current (bottom).
Electrical speed error
Overshoot:
0.4 division
2 volts / division
20 rad/sec / volt
Therefore, 16 rad/sec
Settling time:
1 second
Observed torque
Change in magnitude:
-1 division
2 volts / division
350 Nm / volt
Therefore, -700 Nm
Settling time:
1 second
Stator current
Change in magnitude:
-1 division
2 volts / division
100 A / volt
Therefore, -200 A
Settling time:
1 second
Compensation current
Change in magnitude:
-1 division
2 volts / division
100 A / volt
Therefore, -200 A
Settling time:
1 second


128
Experiment B: Load torque from 0 Nm to 700 Nm


Fig. 5.4.P7. Electrical speed error
ref e e
(top) and observed torque (bottom).


Fig. 5.4.P8. Stator current (top) and compensation current (bottom).
Electrical speed error
Overshoot:
-0.2 divisions
2 volts / division
20 rad/sec / volt
Therefore, -8 rad/sec
Settling time:
1 second
Observed torque
Change in magnitude:
1 division
2 volts / division
350 Nm / volt
Therefore, 700 Nm
Settling time:
4 seconds

Stator current
Change in magnitude:
1 division
2 volts / division
100 A / volt
Therefore, 200 A
Settling time:
4 seconds
Compensation current
Change in magnitude:
1 division
2 volts / division
100 A / volt
Therefore, 200 A
Settling time:
4 seconds


129
5.5. Summary for load torque observer with compensation current

In this chapter, a method for improving the dynamic response or decreasing the effect of varying load
torque is explained. The method requires a load torque observer. Two types of observers are
presented, namely the reduced state observer and the full state observer.

For the reduced state observer, results from FEA are needed to compute the compensation current.
The load torque observer with compensation current feedback forms the torque loop. Low pass
filters are needed for the speed measurement and current measurement, because the speed is
differentiated and the current is used as the index to a LUT. The resultant observed load torque is
filtered by another 2
nd
order LPF and is used as the index to a LUT that provides the compensation
current.

The full state observer gives an estimate of the speed and the load torque. The difference between the
observed speed and the actual speed is used to determine the load torque. LPFs are also needed in this
scheme due to the noise in the speed measurement, but these filters can be included in the design of the
torque loop controller. The settling time for the torque loop can be chosen and stability can be ensured
by choosing the correct type of controller. For a 1
st
order speed LPF the controller structure can be a
lead-lag filter with a gain that gives the required settling time. The full state observer does not need
results from FEA to calculate the compensation current.

Practical results for the reduced state observer with compensation current feedback are similar to
practical results for the full state observer with compensation current feedback. Simulation results for
both schemes also agree with the practical results. It is the opinion of the author that the full state
observer is a better choice, since it can be designed always to give a stable response and no LUTs are
needed to compute the compensation current.


130
Chapter 6 Position sensorless control

This chapter explains one amongst many methods of position sensorless control, namely the high
frequency voltage injection method [A17][A18]. This method can be used for any machine which has
the magnetic anisotropy property. The reluctance synchronous machine (RSM) has a rotor that is
specifically designed to be anisotropic and is therefore perfect for this scheme. Fig. 2.4.1 shows that
for the 110 kW RSM (and PMA RSM), the D-axis self inductance is typically mH L
d
1
'
> , while the
Q-axis self inductance is typically mH L
q
1
'
< . Therefore, for the RSM and PMA RSM:
' '
q d
L L > .

The objective is to control the PMA RSM without a position sensor in the entire speed range (from
zero speed, through rated speed to high speed), with a high dynamic response and in an energy
efficient way. This chapter provides one approach to the problem and could possibly provide a basis
for future research.

The derivation that follows in this chapter is based on the transformation from the stationary
reference frame to the synchronously rotating 0 QD reference frame, which depends on the electrical
rotor position . The transformation is illustrated in Fig. 6.1:

D
Q
e

Fig. 6.1. The reference frame and the 0 QD reference frame.


131
6.1. High frequency voltage injection

Consider the effect of injecting a high frequency, small signal voltage. As a first step, consider a
rotating voltage in the QD0 plane: when the voltage is in the D-axis direction, the current response is
slow (the inductance is large) and therefore the current will reach only a small value; when the voltage
is in the Q-axis direction, the current response is fast (the inductance is small) and therefore the current
will reach a large value. The current response for a rotating voltage in the QD0 plane is therefore an
ellipse. This is true for a positive rotating voltage
+
in
V , and a negative rotating voltage

in
V . If these
responses are superposed (added), the total response is flat on the D-axis. This is shown in Fig. 6.1.1.
D
Q
+
in
V
in

+
resp
I

in
V

resp
I
) cos( t U V
in in
=
( )
in resp
V f I =
Q
Q
Q
Q Q
D
D D
D D
in

in

in

+
+
= =

Fig. 6.1.1. Injected voltage (left) and current response (right).

The objective is to inject an alternating voltage in the D-axis direction as shown in the bottom row of
Fig. 6.1.1. To be able to do this, the location of the D-axis is needed, i.e. knowledge of the rotor
position is needed. However, it is exactly the rotor position that is not known. The only possibility is
to inject the voltage in an observed D-axis direction. If the measured current response is then only in

132
the observed D-axis direction (no Q-axis component), it means that the observed QD0 axes and the
actual QD0 axes are aligned there is thus no error. If the measured current response has a Q-axis
component in the observed QD0 axes, then it means that there is a difference between the observed
QD0 axes and the actual QD0 axes. Note that it is not possible to measure a Q-axis current component
if only a D-axis voltage is injected, unless the observed QD0 axes are misaligned with the actual QD0
axes.

The voltage is injected in the observed D-axis direction (observed field coordinates are indicated by a
hat):
( ) t U V
in
QD
in
cos
) 0 (
^
= (6.1.1)
The injected voltage can be transformed to the stationary reference frame (as shown in Fig. 6.1.)
with the observed rotor angle:
( )
^ ^ ^
cos
) 0 ( ) (

j
in
j QD
in in
e t U e V V = = (6.1.2)
The injected voltage in the actual QD0 reference frame is given by the transformation of equation
(6.1.2) to the actual QD0 reference frame using the actual rotor angle:
( )
( )
) (
) ( ) 0 (
^
^
cos
cos

=
=
=
j
in
j j
in
j
in
QD
in
e t U
e e t U
e V V
(6.1.3)
The actual current response in the actual QD0 reference frame is given by equation (6.1.4). Here it is
assumed that the error in position estimation,

, is time invariant. This is only a reasonable

assumption if the error is very small, i.e. if a good tracking scheme is used.
( )
|
|

\
|

=
=
=

^
^
) sin(
1
cos
1
1
'
) (
'
) 0 (
'
) 0 (
j
in
in DQ
j
in
DQ
QD
in
DQ
QD
resp
e t
L
U
dt e t U
L
dt V
L
i
(6.1.4)
Care should be taken using the complex form in (6.1.4), especially for the inductance term - it is more
clearly written as:
(

\
|
+
|

\
|

=
|

\
|
+
|

\
|
=

^
'
^
'
' '
^
'
^
'
) 0 (
sin cos
) sin(
sin ) sin(
1
cos ) sin(
1
D Q
in
in
Q D
in
in Q
in
in D
QD
resp
jL L
t
L L
U
t
L
U
j t
L
U
i
(6.1.5)


133
It equations (6.1.4) and (6.1.5) it is assumed that the error in rotor position is small enough so that the
term
|
|

\
|

^
j
e is nearly time invariant. Furthermore, if the error is very small, then

|

\
|

^ ^
sin and
1 cos
^

\
|
. (6.1.6)
Equation (6.1.5) gives the current response in the actual QD0 reference, but the current can only be
measured in the observed QD0 frame. The current response in the observed QD0 frame is given by:

) ( ) 0 ( ) 0 (
^ ^

=
j QD
resp
QD
resp
e i i (6.1.7)

Substituting equation (6.1.5) into equation (6.1.7) and then using equation (6.1.6) to simplify, the
current response in the observed QD0 reference frame is obtained as:

( )
( )
(

\
|
+

(
(

\
|
+
|

\
|
+

=
(

\
|

(

\
|
+

=
' '
^
'
' '
) 0 (
' '
^
2
^
' '
' '
) 0 (
^ ^
' '
' '
) 0 (
) sin(
) sin(
1 1
) sin(
^
^
^
Q D Q
in
in
Q D
QD
resp
Q D D Q
in
in
Q D
QD
resp
D Q
in
in
Q D
QD
resp
L L j L
t
L L
U
i
L L j L L
t
L L
U
i
j jL L
t
L L
U
i

(6.1.8)

From equation (6.1.8) it can be clearly seen that the Q-axis component (imaginary component) of the
current response is zero when either the observed QD0 axes are aligned to the actual QD0 axes or if
there is no anisotropy, i.e.
' '
D Q
L L = .

134
6.2. Analysis of the current response and the rotor angle tracking scheme

In this subsection, the current response in the observed QD0 reference frame is analyzed and then a
rotor position tracking scheme is presented. Keeping in mind that for the RSM (and PMA RSM) the
D-axis inductance is greater than the Q-axis inductance, i.e. 0
' '
>
Q D
L L , the following statements
can be deduced from equation (6.1.8):

if 0 ) sin( > t
in
then
0 Re
) 0 (
^
>
)
`

QD
resp
i and
if 0
^
>
|

\
|
then { observed angle too big}
0 Im
) 0 (
^
>
)
`

QD
resp
i (a) {Q-axis current positive}
else if 0
^
<
|

\
|
then {observed angle too small}
0 Im
) 0 (
^
<
)
`

QD
resp
i (b) {Q-axis current negative}
else if 0 ) sin( < t
in
then
0 Re
) 0 (
^
<
)
`

QD
resp
i and
if 0
^
>
|

\
|
then { observed angle too big}
0 Im
) 0 (
^
<
)
`

QD
resp
i (c) {Q-axis current negative}
else if 0
^
<
|

\
|
then {observed angle too small}
0 Im
) 0 (
^
>
)
`

QD
resp
i (d) {Q-axis current positive}

If the Q-axis current is used for feedback, then result [(a) and (b)] corresponds to positive feedback,
while [(c) and (d)] corresponds to negative feedback.

135
From equation (6.1.8) it can be seen that the Q-axis current amplitude is an indication of the error in
estimated rotor position. The Q-axis current can be used for correction of the estimated rotor position
in a phase locked loop (PLL) manner, but then the sign of the Q-axis current needs to be unambiguous,
i.e. we have to choose either [(a) and (b)] or otherwise [(c) and (d)]. It will be shown that it is
possible to use the D-axis current to rectify the Q-axis current.

The following example illustrates the concept of rectifying the Q-axis current using the sign of the D-
axis current. From equation (6.1.8), the D-axis and Q-axis currents can be written as:
) sin(
1
t K i
in d
=
( ) ) sin(

2
t K i
in q
= , (6.1.9)
where
1
K and
2
K are positive constants. Fig. 6.2.1 shows the two possible scenarios.

Fig. 6.2.1. Q-axis current rectified by the sign of the D-axis current.

Tracking of the rotor angle can be achieved in a phase locked loop (PLL) manner. A PLL has three
parts namely the phase detector, loop filter and voltage controlled oscillator [B4, p.429]. The
following scheme is proposed:
the observed rotor angle is an input to the transformation from the stationary reference frame
to the rotating reference frame (phase detector)
the output of this transformation is the D-axis and Q-axis current components, which are
filtered at the high frequency band in which the information lies (loop filter)
the sign of the Q-axis current component is rectified using the sign of the D-axis current
component

136
the Q-axis current is the input to a voltage controlled oscillator (VCO), which can possibly be
a PI controller
the output of the VCO is the rotational speed, which can be integrated to obtain the estimated
rotor position

The Q-axis current can be used to correct the estimated position in a PLL manner only if [(c) and (d)]
is chosen, because negative feedback is necessary. This means that when the D-axis current is
positive, the Q-axis current should be inverted and when the D-axis current is negative, the Q-axis
current should not be inverted. This algorithm is shown in block diagram form in Fig. 6.2.2.

1
|i q|
Si gn
Product
-1
2
Iq
1
Id

Fig. 6.2.2. Rectification of the Q-axis current using the D-axis current.

Now, when the estimated rotor position is too big, the Q-axis current is negative, which means that the
output of the VCO will be reduced and the estimated angle will therefore also reduce. If the estimated
rotor position is too small, the Q-axis current is positive, which increases the output of the VCO and
therefore also increases the estimated rotor position.

The complete algorithm is shown in block diagram form in Fig. 6.2.3. The PI current controllers
should only respond to the fundamental current component. Low pass filters (LPFs) in the feedback
paths of the current loops are needed, because the current controllers, ) (z D and ) (z Q , will try to
compensate for the high frequency current components if the LPFs are omitted. The LPFs should
attenuate the high frequency current components sufficiently. The stability of the current loops should
be considered again, since the LPF adds a 90 phase shift.

In Fig. 6.2.3, the Park transform is used to transform from the synchronously rotating reference frame
to the stationary reference frame in ABC rather than - the result is the same. Also note that the
decoupling procedure explained in chapter 3 is omitted in Fig. 6.2.3, but should be included in
practice. The block diagram in Fig. 6.2.3 can be used for simulation.

This chapter explained one method of position sensorless control, which is promoted at this time by
the Bergishe Universitt Wuppertal in Germany. A collaboration between the University of

137
Stellenbosch and Bergishe Universitt Wuppertal allowed the author to visit the Electrical Machines
and Drives group in Wuppertal during September 2005. During this time the position sensorless
algorithm was discussed and the author constructed a simulation model similar to Fig. 6.2.3. The
detail designs of the simulation blocks (low pass filters, band pass filters, voltage controlled oscillator
etc) and the simulation results are however not shown here. Practical tests were performed on a RSM
(small in size compared to the RSM discussed in this thesis) to prove that the sensorless scheme is
valid. These practical results are shown in Fig. 6.2.4 and a practical demonstration video is also
included on the CDROM [V1].
Iq*
Iq
Vq*
Vd* i d ^
i q ^
Id
RSM model
i n ABC
reference frame
recti fi ed i q
1
El ectri cal Speed
Vd (HF)
VCO(z)
Si gn
Vd
Vq
theta
Id
Iq
RSM model
Product
theta
A
B
C
D
Q
Park
A
B
C
theta ^
D
Q
Park
butter
LPF W
butter
LPF Q
butter
LPF D
A*
B*
C*
A
B
C
Inverter model
theta
D
Q
A
B
C
Inverse Park
D
Q
theta ^
A
B
C
Inverse Park
-1
D(z)
Q(z)
K Ts
z-1
Di screte-Ti me
Integrator
butter
BPF Q
butter
BPF D
2
Iq Ref
1
Id Ref

Fig. 6.2.3. Block diagram representation of proposed sensorless position control method.



138

Position Description
Top, left D-axis current reference (top) and response (bottom)
Top, right Q-axis current reference (top) and response (bottom)
Bottom, left Speed reference (top) and speed response (bottom)
Bottom, right Estimated (top) and measured (bottom) electrical rotor position



Fig. 6.2.4. Practical results for RSM with position sensorless control.



139
Chapter 7 Summary, conclusions and recommendations

This thesis explains the operation of an electrical drive system and focuses on the design of a digital
control system that ensures good dynamic response and energy efficient operation. The components of
the drive system are a 110 kW permanent magnet assisted reluctance synchronous machine (PMA
RSM), a rectifier, an inverter, a load system and a digital controller.

The current in the PMA RSM is controlled in such a way that the current angle in the rotating QD0
reference is constant, as this assures high efficiency operation. The current angle of 54 is maintained
to give the maximum torque per current ratio. The current controllers are proportional controllers with
gains such that stability is always ensured.

A simple proportional integral speed controller is used for speed control of the PMA RSM. The
response time for the speed loop is chosen to be much longer than the response time for the current
loop, because then it can be assumed that the current follows its reference instantaneously, the speed
controller becomes parameter independent and a low pass filter can be used in the feedback path of the
speed loop.

A load torque observer with compensation current feedback is used to improve the dynamic
performance of the drive system. If the response time of the observer is increased, the amount of noise
in the system is also increased. Two observer structures, namely the reduced state observer and full
state observer, give similar practical results. Simulation results indicate that the observed load torque
of the full state observer is more accurate than that of the reduced state observer during speed changes.

It is shown that position sensorless control of the PMA RSM is possible. From this work, several
conclusions are reached and some recommendations are made for future research.


7.1. Conclusions

From theoretical analysis, finite element analysis results, simulation results using Matlabs Simulink
and in most cases practical implementation, the following conclusions are reached:
The currents of the reluctance synchronous machine (RSM) can be controlled by two methods,
namely constant field current control (CFCC) and constant current angle control (CCAC).
Both methods provide equal torque generation capabilities for the motoring and generating

140
mode of operation. The response time for torque generation using CFCC is slightly less than
the response time using CCAC, but CCAC is much more energy efficient.
The torque generation capability of the RSM is increased by adding permanent magnets in the
rotor. This machine can be called a PMA RSM. The currents of the PMA RSM should be
controlled using CCAC in order to provide equal torque generation in the motoring and the
generating mode of operation. CFCC should not be used, as the torque capability in
generating mode of operation is severely reduced.
Stable current controllers for the RSM or PMA RSM can be designed in the W-plane. The
current controllers are simple proportional controllers with a gain given by an analytic
formula, which can easily be adjusted for any stability margin.
The speed controller can be designed as a simple proportional integral (PI) controller that is
parameter independent, but then the response time must be much slower than the current
controllers response time. The load torque is a disturbance input in the speed loop and is
compensated for with the same response time as the speed loop.
A load torque observer with compensation current feedback can be used to improve the
dynamic response of the drive system. The reduced state observer requires results from FEA
and the derivative of the noisy speed signal to compute the compensation current. In the full
state observer there is no differentiation and it does not require any result from FEA to
compute the compensation current.
Position sensorless control of the PMA RSM is possible using a high frequency voltage
injection method. The scheme is based on the magnetic anisotropic property of machines,
which makes it ideal for a RSM or PMA RSM.


7.2. Recommendations

This study was subject to a few limitations for which recommendations are made. The work also
raises some interesting questions for future research:
The hydraulic load system can be replaced by a fast responding electrical load system to test
the proposed method of dynamic control on the PMA RSM.
The proposed method of dynamic control can be implemented and tested on a RSM (three-
phase and six-phase), comparing CFCC and CACC.
Position sensorless control of the PMA RSM in the entire speed range (at zero speed, and from
zero speed to high speed), with constant current angle control, can be implemented and tested.


141
Appendix R References

R.1. Articles on CDROM
Path: D:\articles
[A1] Kamper M.J. and Mackay A.T.:Optimum Control of the Reluctance Synchronous Machine
with Cageless Flux Barrier Rotor, SAIEE Trans., June 1995, Vol. 86, No. 2, pp. 49-56
[A2] Fratta A. and Vagati F.:Synchronous Reluctance vs Induction Motor: a comparison,
Proceedings Intelligent Motion (Nrnberg), April 1992, pp. 179-186
[A3] Bombela X.B., Jackson S.K. and Kamper M.J.: Performance of Small and Medium Power
Flux Barrier Rotor Reluctance Synchronous Machine Drives, ICEM (Istanbul), Sept. 1998,
Vol. I, pp. 95-99.
[A4] Germishuizen J., Van der Merwe F.S. Van der Westhuizen K. and Kamper M.J.:
Performance comparison of reluctance synchronous and induction traction drives for
electrical multiple units, IEEE IAS Conference (Rome), October 2000
[A5] Guglielmi P., Pastorelli M., Pellegrino G. and Vagati A.: Position-Sensorless Control of
Permanent-Magnet-Assisted Synchronous Reluctance Motor, IEEE Transactions on Industry
Applications, Vol. 40, No. 2, 2004, pp. 615-622
[A6] Morimoto S., Sanada M. and Takeda Y.: Performance of PM-Assisted Synchronous
Reluctance Motor for High-Efficiency and Wide Constant-Power Operation, IEEE
Transactions on Industry applications Vol.37, No.5, pp. 1234-1239
[A7] Boldea I., Tutelea L. and Ilie Pitic C.: PM-Assisted Reluctance Synchronous
Motor/Generator (PM-RSM) for Mild HybridVehicles: Electromagnetic Design, IEEE
Transactions on Industry Applications, Vol. 40, No. 2, 2004, pp. 492-498
[A8] Lipo T.A.: Synchronous Reluctance Machines A Viable Alternative for AC Drives?
Electrical Machines and Power Systems, Vol. 19, pp. 659-671, 1991
[A9] Xu L., Xu X., Lipo T.A. and Novotny D.W.: Vector Control of a Synchronous Reluctance
Motor Including Saturation and Iron Loss, IEEE transactions on Industrial Applications, Vol.
27, No. 5, pp. 977-985
[A10] Matsuo T. and Lipo T.A.: Field Oriented Control of Synchronous Reluctance Machine,
Proc. IEEE-PESC, 1993, pp. 425-431
[A11] Vagati A., Pastorelli M., Guglielmi P. and Canova A.: Synchronous Reluctance Motor
based Sensorless Drive for General Purpose Application
[A12] Arefeen M.S., Ehsani M., Lipo T.A.: Sensorless Position Measurement in Synchronous
Reluctance Motor, IEEE Transactions on Power Electronics, Vol. 9, No.6, pp. 624-630
[A13] Matsui N., Makino T., Satoh H Autocompensation of Torque Ripple of Direct Drive Motor
by Torque Observer. IEEE Trans on Industry Applications Vol. 29, No. 1, 1993, pp. 187-194

142

[A14] Zhao J., Kamper M.J., Van der Merwe F.S., On-line Control Method to Reduce
Mechanical Vibration and Torque Ripple in Reluctance Synchronous Machine Drives.
Proceedings of the IECON'97. New Orleans, USA, 1997, Vol 1, pp. 126-131.
[A15] Yamada K., Komada S., Ishida M. and Hori T., Analysis and Classical Control Design of
Servo System Using High Order Disturbance Observer. Industrial Electronics, Control and
Instrumentation, 1997. IECON 97. 23
rd
, Vol.1, pp. 4 9
[A16] Gopinath B., On the control of Linear Multiple Input-Output Systems. The Bell System
Technical Journal, Vol. 50, No. 3, March, 1971
[A17] Linke M., Kennel R., Holtz J., Sensorless position control of Permanent Magnet
Synchronous Machines without Limitation at Zero Speed
[A18] Linke M., Kennel R., Holtz J., Sensorless Speed and Position Control of Synchronous
Machines using Alternating Carrier Injection
[A19] Koziol, J.Jr., An Approximate Root Locus Method in the S-Plane for Sampled-Data
Systems, IEEE Transactions on Automatic Control, Issue 1, Feb 1971, Vol.6, pp. 101-102
[A20] Senjyu T., Shingaki T., Uezato, K. Sensorless Vector Control of Synchronous Reluctance
Motors with Disturbance Torque Observer, IEEE Transactions on Industrial Electronics,
Issue 2, April 2001, Vol. 48, pp. 402-407
[A21] Fratta A., Vagati A., Villata F., Permanent magnet assisted synchronous reluctance drives
for constant-power applications: Drive Power Limits, Intelligent Motion proceedings, 1992,
pp. 196-173

R.2. Books
[B1] Wildi T, Electrical Machines, Drives, and Power Systems, Prentice Hall, 2002
[B2] Kamper M.J., Design Optimisation of Cageless Flux Berrier Rotor Reluctance Synchronous
Machine, P.hD. Dissertation, University of Stellenbosch, December 1996
[B3] Fick P.D., Evaluation of the constant current angle controlled Reluctance Synchronous
Machine drive, M.Sc.Ing Dissertation, University of Stellenbosch, March 2002
[B4] Ziemer R.E., Tranter W.H., Principals of Communications Systems Modulation and
Noise, Joahn Wiley & Sons, Inc., 2002
[B5] Sibande, S.E., Design and Perfomance Evaluation of a PM-Assisted Rotor of a 110kW
Reluctance Synchronous Machine, M.Sc.Eng Dissertation, University of Stellenbosch, 2004
[B6] Proakis J.G. and Manolakis D.G., Digital Signal Processing, Prentice Hall, 1996
[B7] Ogata K., Discrete Control Systems, Prentice Hall, 1987
[B8] Krause, Paul C., Analysis of Electric Machinery, Chapter 3 Reference Frame Theory
[B9] Franklin G., Powell J. and Emami-Naeini A. Feedback control of Dynamic Systems.
Prentice Hall, 2002

143
[B10] Mohan, Undeland, Robbins, Power Electronics Converters, Applications and Design,
Third Edition, John Wiley & Sons Inc, 2003
[B11] Nilsson J.W., Riedel S.A., Electric Circuits, Sixth Edition, Prentice Hall, 2001
[B12] Phillips C.L., Troy Nagle H., Digital Control System Analysis and Design, Third Edition.
Prentice Hall, 1995
[B13] Wolf W., Computers as components, Morgan Kaufman publishers, 2001
[B14] Peebles P.Z., Probability, Random Variables and Random Signal Principles, 4
th
edition,
McGraw-Hill Inc, 2001
[B15] Kuo, Analysis and Synthesis of Sampled-Data Control Systems, Prentice Hall, 1963
[B16] Steward J., Calculus Fourth Edition, Books/Cole publishing company, 1999
[B17] Germishuizen J., Comparitive Study of Reluctance Synchronous and Induction Machine
Drives for Tail Traction, M.Sc.Ing Dissertation, University of Stellenbosch, 2000

R.3. P-CAD 2000 circuit diagram on CDROM
[C1] Circuit diagram for dumping circuit and warning light circuit with PCB layout
Path: D:\P-CAD 2000 circuit

R.4. Datasheets / Brochures on CDROM
Path: D: \data sheets
[D1] 2N2219a transistor
[D2] ICL7667 driver for IGBT
[D3] LA 55-P LEM current sensor
[D4] LM7815 +15V regulator
[D5] LM7915 -15V regulator
[D6] LV 100 LEM voltage sensor
[D7] OPA177 operational amplifier
[D8] SKM 50GB 121D Semikron dual IGBT module
[D9] SRI 5VDC SD C Songle relay
[D10] W4M diode bride rectifier
[D11] TRF 4283 230VAC,50Hz 12VAC transformer
[D12] LT 505-S LEM current sensor
[D13] TMS 320V C33 Texas Instruments DSP
[D14] EPI C6Q 240C8 Altera Cyclone FPGA
[D15] 6 SE 71, 3 AC Siemens Simovert Master Drive
[D16] resolver solutions Texas Instruments



144
R.5. Finite element analysis results on CDROM
[F1] Graphs for flux linkage and inductance comparison between RSM and PMA RSM
Path: D:\FEA\Comparison
[F2] Graphs for torque comparison between RSM and PMA RSM
Path: D:\FEA\Torque 360
[F3] Three-dimensional torque and flux linkage surfaces
Path: D:\FEA\Surface

R.6. Matlab 7 analysis on CDROM
[M1] Signal to noise ration of measured speed signal
Path: D:\Signal to noise ratio

R.7. Programs on CDROM
Path: D:\programs
[P1] control program for PMA RSM
[P2] finite element analysis program for RSM or PMA RSM
[P3] step-by-step guide for operator of the drive system

R.8. Matlab 7 simulation files on CDROM
[S1] Current control for PMA RSM without inverter model
Path: D: \simulation files\Current control\PMA RSM without inverter
[S2] Current control for PMA RSM with inverter model
Path: D: \simulation files\Current control\PMA RSM with inverter
[S3] Speed control for PMA RSM without inverter model
Path: D: \simulation files\Speed control\PMA RSM without inverter
[S4] Speed control for PMA RSM with inverter model
Path: D: \simulation files\Speed control\PMA RSM with inverter
[S5] Load torque observer for RSM example
Path: D:\simulation files\Torque load observer\Example
[S6] Reduced state observer for PMA RSM
Path: D:\simulation files\Torque load observer\Reduced state observer
[S7] Full state observer for PMA RSM
Path: D:\simulation files\Torque load observer\Full state observer


145
R.9. Videos on CDROM
[V1] demonstration of position sensorless control on the Wuppertal RSM
[V2] demonstration of rated conditions, dumping circuit and load torque observer on the 110 kW
PMA RSM

R.10. Websites
[W1] http://www.admotec.com/TT02.html Understanding Resolvers and Resolver-to-Digital
Conversion, October 2005
[W2] http://www.altera.com/literature/lit-cyc.jsp Altera Cyclone FPGA. October 2005
[W3] http://dsp.ti.com/ Texas instruments DSP village, October 2005
[W4] http://www.nist.gov/dads/HTML/finiteStateMachine.html National Institute of Standards and
Technology, October 2005
[W5] http://www.schenck-ind.com/balance2.html Schenck hydraulic dynamometer, October
2005


146
Appendix A Torque equation

The definition of the Park transform is:
(
(
(

+ +
=
(
(
(
(

+
+
=
=

1 ) 120 sin( ) 120 cos(

1 ) 120 sin( ) 120 cos(
1 ) sin( ) cos(
] [
2
1
2
1
2
1
) 120 sin( ) 120 sin( ) sin(
) 120 cos( ) 120 cos( ) cos(
3
2
] [
] ][ [ ] [
1
0





s
s
abc s qd
K
K
f K f
(A.1)
Use Park to transform the reluctance synchronous machines (RSMs) voltage equations in the ABC
reference frame (stationary with respect to the stator) to the QD0 reference frame (stationary with
respect to the rotor). In the ABC reference frame:
] [ ] [ ] [
abc abc s abc
dt
d
i r v + = (A.2)

In equation (A.2), ] [
abc
v is the supply voltage, ] [
abc s
i r is the loss component and ] [ ] [
abc abc
dt
d
e = is
the induced voltage. Now use equation (A.1) to transform equation (A.2) to the QD0 reference frame:
] [ ] [
0 0 0
0 0 1
0 1 0
] [
]) [ ( ] ][ [ ] )[ ] [ ]( [ ] [
]) [ ] ([ ] [ ] [
] [ ] [ ] [ ] [ ] ][ [ ] [
0 0 0
0
1
0
1
0
0
1
0
0
qd qd e qd s
qd s s qd s s qd s
qd s s qd s
abc s abc s s abc s qd
dt
d
i r
dt
d
K K K
dt
d
K i r
K
dt
d
K i r
dt
d
K i r K v K v

+
(
(
(

+ =
+ + =
+ =
+ = =

Hence,
q d e q s q
dt
d
i r v + + =
,
d q e d s d
dt
d
i r v + = (A.3)
where
e
is the electrical speed.

147
In equation (A.3), the derivative of the flux linkages can be expanded into products of partial
derivatives of the flux linkages with respect to variables on which they depend, and the derivatives of
those variables. Taking the derivative of the D-axis flux linkage for example:

e
e
d
q
q
d d
d
d d
dt
di
i dt
di
i dt
d

= (A.4)
By analogy, the derivative of the Q-axis flux linkage is given by:

e
e
q q
q
q
d
d
q q
dt
di
i dt
di
i dt
d

= (A.5)
The partial derivatives of the flux linkage with respect to the currents can be written as self and mutual
inductances. These inductance terms are referred to as instantaneous inductances or transient
inductances. Note however, that they are not linearized inductances:

e
e
d
q
d
d
d
d
dt
di
M
dt
di
L
dt
d

+ + =
' '

e
e
q q
q
d
q
q
dt
di
L
dt
di
M
dt
d

+ + =
' '
(A.6)
Substituting equations (A.4) and (A.5) into equation (A.3), result in:

e
e
q q
q
q
d
d
q
d e q s q
dt
di
i dt
di
i
i r v

+ + =

e
e
d
q
q
d d
d
d
q e d s d
dt
di
i dt
di
i
i r v

+ = (A.7)
The terms in equation (A.7) that contribute towards power output (and torque output) of the machine
are the speed voltages and the change in flux linkages due to the rotor position. Equation (A.7) with
only the power contributing terms is:

e
e
q
d e q
e

+ =

e
e
d
q e d
e

+ = (A.8)
The change in flux linkage due to the rotor position is assumed to be very small (generally techniques
such as skewing are used to minimize the change in flux linkage due to rotor position). Thus
neglecting the change in flux linkage due to the rotor position, equation (A.8) becomes:

(
(
(

=
0
] [
0 q e
d e
qd
e

(A.9)

148
The power in the ABC reference frame must be equal to the power in the QD0 reference frame,
therefore:
] [
3 0 0
0
2
3
0
0 0
2
3
] [
] [ ] [ ] [ ] [
]) [ ] ([ ]) [ ] ([
] [ ] [
1 1
1 1
qdo
T
qdo
qdo s
T
s
T
qdo
qdo s
T
qdo s output
abc
T
abc output
i e
i K K e
i K v K P
i e P
(
(
(
(
(
(

=
=
=
=


(A.10)
Using the transpose of equation (A.9) and substituting it into equation (A.10):
( )
d q q d e
qdo q e d e output
i i
i P


=
(
(
(
(
(
(

=
2
3
] [
3 0 0
0
2
3
0
0 0
2
3
] 0 [
(A.11)

The output torque,
output
T , is related to the output power,
output
P , by the mechanical speed,
m
as:
output
m
output
P T

1
= (A.12)

The mechanical speed,
m
, is related to the electrical speed,
e
, by the number of pole pairs, p , as:
e m
p

1
= (A.13)

Using equations (A.11) through to (A.13), the output torque is given by:
) ( )
2
3
(
d q q d output
i i p T = (A.14)


149
The relationship between flux linkage, current and linearized inductance is given by i L = . Using
linearized inductances, equation (A.14) can be written as:
) 2 sin( )
2
1
( ) ( )
2
3
(
sin cos ) ( )
2
3
(
) ( )
2
3
(
) ( )
2
3
(
2

s q d
s s q d
q d q d
d q q q d d
i L L p
i i L L p
i i L L p
i i L i i L p T
=
=
=
=

) 2 sin( ) ( )
4
3
(
2

s q d
i L L p = , (A.15)
where
s
i is the current magnitude and is the current angle.

Adding Permanent Magnets (producing a constant flux linkage) we can substitute
) (
PM q q

dt
d
dt
d
dt
d
q
PM q
q

= ) (
in equation (A.2):
d PM q e d s d
q d e q s q
dt
d
i r e
dt
d
i r e


+ =
+ + =
) (
(A.16)
and in equation (A.12):
) 2 sin( ) ( )
4
3
(
) ( )
2
3
(
) ( )
2
3
(
) ) ( ( )
2
3
(
2

s
q
PM
q d
q d
q
PM
q d
d q
q
PM
d q q q d d
d PM q q d
i
i
L L p
i i
i
L L p
i i
i
i i L i i L p
i i p T
+ =
+ =
+ =
=
(A.17)


150
Appendix B Zero order hold and the Z-transform

The output of a digital controller is followed by a sample and hold element that serves as the digital-to-
analogue converter (DAC). The hold element is usually a zero order hold (ZOH) element, which
keeps the sampled value constant for one sampling period T . This section analyses the effect of the
ZOH and in order to do this, the transfer function of the ZOH is derived. A discrete equivalent for the
combination of the ZOH and plant is then obtained using the Z-transform.

To be absolutely clear about the difference between a continuous-time and discrete-time signal, the
definitions are given here:

Continuous-time signals [B6, p.8] are defined for every value of time and mathematically these signals
can be described by functions of a continuous variable, e.g. ) (t x y
a a
= where
a
y is the analogue
signal,
a
x is the function describing the signal and t is the continuous variable.

Discrete-time signals, however, are defined only at certain specific values of time. Such a signal can
be obtained by taking samples of a continuous-time signal. There are many ways to sample an
analogue signal, but the most common method is periodic and uniform sampling [B6, p.23]. In this
method, the sampling instants are equidistant and the time interval between samples is referred to as
the sampling time T . If the discrete-time signal is represented in terms of the sampling time it
becomes only a function of an integer variable, e.g. ... 2 , 1 , 0 ), ( ] [ = = = n n T x n x y
a d d

Therefore the discrete-time signal
d
y can be represented as a series of numbers ] [n x
d
, where n will
be referred to as the time index.

An example of a continuous-time input signal to a ZOH element and the output signal of the ZOH is
shown in Fig. B.1.

Fig. B.1. Input and output waveforms of a ZOH element.


151
In the figure above, let the continuous-time input signal be ) (t x
a
and the continuous-time output
signal be ) (t x
a
. The output signal can be described in terms of the input signal using the unit step
function ) (t u as follows:
[ ]
[ ]
[ ] ... ) 3 ( ) 2 ( ) 2 (
) 2 ( ) ( ) (
) ( ) ( ) 0 ( ) (
+
+
+ =
T t u T t u T x
T t u T t u T x
T t u t u x t x
a
a
a a
(B.1)
Note that in equation (B.1), only samples of the continuous-time input signal at the sampling instants
are used. Therefore the output signal defined in equation (B.1) ) (t x
a
is the same for a continuous-
time input signal ) (t x
a
and a discrete-time input signal ] [n x
d
that is obtained by sampling the
continuous-time input signal at the sampling instants nT , where n is the integer time index, and T is
the inverse of the sampling frequency
s
f :

[ ]
[ ]
[ ] ... ) 3 ( ) 2 ( ] 2 [
) 2 ( ) ( ] 1 [
) ( ) ( ] 0 [ ) (
+
+
+ =
T t u T t u x
T t u T t u x
T t u t u x t x
d
d
d a
(B.2)

The Laplace transform [B11, p.585] is given by:
{ }

= =
0
) ( ) ( ) ( dt e t x t x L s X
st
a a a
(B.3)
The operational transform for translation in the time domain [B11, p598] is given by:
{ } ) ( ) ( ) ( s X e a t u a t x L
a
as
a

= (B.4)
Note that the Laplace transform is only defined for causal continuous-time signals. In equation (B.4)
the function that is being transformed is multiplied by the unit step function to ensure that the resulting
function is defined for all positive time.

Using equation (B.3) and (B.4), the Laplace transform of the output signal described by equation (B.1)
is given by:

[ ]
(

=
+ + +
(

=
+
(

+
(

=

=

s
e
s
e nT x
e T x e T x x
s
e
s
s
e
s
e
T x
s
e
s
x s X
Ts
n
nTs
a
Ts
a
Ts
a a
Ts
Ts Ts
a
Ts
a a
1
) (
... ) 2 ( ) ( ) 0 (
1
... ) (
1
) 0 ( ) (
0
2
2
(B.5)


152
Consider a discrete-time signal ) (nT x
a
that is obtained by sampling the continuous-time input signal
) (t x
a
using periodic and uniform sampling [B6, p.23]. The discrete time signal is only defined at the
sampling instants nT . If the Laplace transform of this discrete-time signal is taken, the integral sign
in equation (B.3) can be replaced by a summation sign and continuous variable t can be replaced by
the discrete variable nT . Then the term in the first bracket on the right hand side of equation (B.5) is
recognized as the Laplace transform of the discrete-time input signal:

=
0
) ( ) (
n
nTs
a
e nT x s X (B.6)
In fact, if the ZOH element follows the output of a discrete controller, the input to the ZOH is a
discrete-time signal. Since the first term on the right hand side of equation (B.5) is the Laplace
transform of the input signal to the ZOH and the term on the left hand side of equation (B.5) is the
Laplace transform of the output signal of the ZOH, the second term on the right hand side must be the
transfer function of the ZOH. Therefore:

(

s
e
s H
Ts
zoh
1
) ( (B.7)
The effect of the ZOH can be analyzed by looking at a Bode diagram of the transfer function. To
construct a Bode diagram, equation (B.5) is manipulated into a suitable form:

)
2
sin( 2
2
) ( 2 1
) (
2
2 2 2
T
e
j
e e e
j
e
j H
j
T
j
T
j
T
j
T
Tj
zoh

= (B.8)

In equation (B.8) the sampling time T can be written in terms of sampling frequency:
s s s
f
T

= = =
2
2
2
1
2
(B.9)
Substituting equation (B.9) into equation (B.8) and writing it in terms of magnitude and angle:
s
s
s
zoh
T j H

=
) sin(
) ( (B.10)
From equation (B.10) it is easy to construct a Bode diagram, and this is shown in Fig. B.2.

153

Fig. B.2. Bode diagram for ZOH.

From equation (B.10) and Fig. B.2, it can be seen that the ZOH distorts the amplitude and also adds
phase delay. For frequencies less than the Nyquist frequency
2
s

, the amplitude is only attenuated

slightly. The phase lag introduced by the ZOH is however, more serious. At the Nyquist frequency a
phase lag of 90 has already been introduced. To ensure a stable control system, this phase lag has to
be taken into account.

To take the ZOH into account, the transfer function of the ZOH should be inserted between the
controller and plant. Since the controller is discrete, but the transfer functions of the ZOH and the
plant are continuous, a choice has to be made to obtain a homogenous structure. The transfer function
of the ZOH, equation (B.7), is rather peculiar in the continuous domain, so it is preferable to combine
the transfer functions of the ZOH and the plant, and then to obtain a discrete equivalent. The discrete
equivalent is obtained using the inverse Laplace transform and then the Z-transform. This method
corresponds to the step-invariance method [B7, p.323].

The Z-transform [B6, p.152] of a discrete-time signal ] [n x
d
is defined as the power series:
{ } ] [ ] [ ) ( n x Z z n x z X
d
n
n
d d
= =

(B.12)
This is the transformation for the time-domain signal ] [n x
d
into its complex-plane representation
) (z X
d
. From a mathematical point of view the Z-transform is simply an alternative representation of
a signal. The exponent of the complex variable z gives information about the time index n and the
coefficient of z gives the value of the signal at time index n.


154
If the time series ) (n x
d
is obtained by sampling the continuous-time signal ) (t x
a
at the sampling
instants nT , equation (B.12) can be written as:
[ ] ) ( ) ( ) ( t x Z z nT x z X
a
n
n
a d
= =

(B.13)
Notice that the notation for the Z-transform has changed from curly brackets in equation (B.12) to
square brackets in equation (B.13). Assuming a causal signal (only positive time values), equation
(B.13) can be written as:
[ ] ) ( ) ( ) (
0
t x Z z nT x z X
a
n
n
a d
= =

(B.14)
Equation (B.14) is also known as the one sided Z-Transform [B7, p.42]. Comparing the one-sided Z-
transform in equation (B.14) and the Laplace transform in equation (B.6) the following relationship is
obtained:

sT
n nTs
e z
z e
=
=

(B.15)
This is an important relationship and will be used shortly.

The continuous-time signal ) (t x
a
in equation (B.14) can be obtained by the inverse Laplace transform
of ) (s X
a
. The following equation establishes the notation that is used:
{ } { } [ ] { } [ ] [ ] ) ( ) ( ) ( ) ( ) (
1
s X Z s X L Z t x Z nT x Z n x Z
a a a a d
= = = =

(B.16)

In equation (B.16) the notation plays an important part: curly brackets are used where the strict
definition of the Z-transform is applied, while square brackets are used where the definition is relaxed.
It is however important to remember that the Z-transform can only be performed on a series of
numbers.

The digital controller, DAC and continuous plant are in series, as illustrated in Fig. B.3. The DAC is a
ZOH element and the continuous transfer function for the ZOH is derived and shown in equation
(B.7). The transfer functions for the ZOH and the plant are now combined and a discrete equivalent is
obtained by using the Z-transform.

Fig. B.3. Discrete controller, ZOH and continuous plant.


155
Using equation (B.16),
(

s
e
s G z G
Ts
1
) ( ) ( . (B.17)
Using the relationship in equation (B.15), equation (B.17) is written as:

(

\
|
=
(

=

s
s G
z
z
s
s G
z z G
) ( 1 ) (
) 1 ( ) (
1
(B.18)
To evaluate equation (B.18) for a specific plant ) (s G , it is advisable to use a Z-transform
table such as [B12, p.676]. An example is shown below:
Laplace transform ) (s X Time function ) (t x Z-transform ) (z X
( ) a s s
a
+

at
e

1 ( )
( )( )
aT
aT
e z z
e z

1
1


Using the table above and equation (B.18), if
s
s G
s X
) (
) ( = then
aT
aT
e z
e
z G

=
1
) ( .



156
Appendix C The rotating magnetic field

Consider a stator with 3-phase winding, as shown in Fig. C.1. The three phases are represented by
three pairs of concentrated coils, marked A, A, B, B, C and C. Phase A current will flow into A and
out at A, phase B current will flow into B and out at B, and phase C current will flow into C and out
at C. Currents flowing in are marked with a cross and currents flowing out are marked with a dot.
Using the right-hand rule, the magnetic field pattern can be obtained.

A
A'
B'
B C
C'
phase A
magnetic
phase C
magnetic
phase B
magnetic

Fig. C.1. Stator with 3-phase winding.

Three-phase currents (ABC sequence) flow in the coils and create a rotating magnetic field. This can
be seen by looking at the magnetic field pattern at time T0 and time T1 (indicated in Fig. C.2). At
time T0 phase A is at a positive maximum, while both phase B and C are at half the maximum and
negative. At time T1 phase B is at a negative maximum, while both phases A and C are at half the
maximum and positive.


157

Fig. C.2. Three phase currents. Phase A (red), phase B (blue), phase C, (green).

At time T0, the magnetic field pattern is in the direction indicated in Fig. C.3 below:
A
A'
B'
B C
C'
phase A
magnetic
phase C
magnetic
phase B
magnetic

Fig. C.3. Magnetic field direction at time T0.

At time T1, the magnetic field pattern is in the direction indicated in the figure below:
A
A'
B'
B C
C'
phase A
magnetic
phase C
magnetic
phase B
magnetic

Fig. C.4 Magnetic field direction at time T1.
It should thus be clear that as time goes by, the magnetic field rotates anti-clockwise.

158
Appendix D Park transform in the time domain

Consider three-phase currents
abc
i in the stationary ABC reference frame and the transformation of
these currents to the synchronously rotating QD0 reference frame using the Park transform. The Park
transform requires the electrical rotor angle
e
, which can be written as a function of time using the
electrical rotor speed:
t
e e
= (D.1)
Let us say that the fundamental frequency of the three-phase currents in the ABC reference frame is
1
in the steady state. If the ABC sequence is chosen, the currents can be written as:

( )
(
(
(
(
(
(
(

\
|
+
|

\
|
=
(
(
(

3
2
cos
3
2
cos
cos
1
1
1

t I
t I
t I
i
i
i
p
p
p
c
b
a
(D.2)
Using the Park transform (given in Appendix A), and the definition of the ABC currents in equation
(D.2), the currents in the QD0 reference frame is given by:

( )
( )
( )
(
(
(
(
(
(
(

\
|
+
|

\
|

(
(
(
(
(
(
(

\
|
+
|

\
|

|

\
|
+
|

\
|

=
(
(
(

3
2
cos
3
2
cos
cos
2
1
2
1
2
1
3
2
sin
3
2
sin sin
3
2
cos
3
2
cos cos
3
2
1
1
1
0

t I
t I
t I
t t t
t t t
i
i
i
p
p
p
e e e
e e e
d
q
(D.3)
From equation (D.3), the Q-axis current is given by:

( ) ( )
( ) ( )
( )
( )
(
(
(
(
(
(
(

+
|

\
|
+ + +
+ +
|

\
|
+ +
+ + +
=
(

\
|
+
|

\
|
+ +
|

\
|

|

\
|
+ =
t t
t t
t t
I i
t t t t t t I i
e e
e e
e e
p q
e e e p q
1 1
1 1
1 1
1 1 1
cos
3
4
cos ...
... cos
3
4
cos ...
... cos cos
2
1
3
2
3
2
cos
3
2
cos
3
2
cos
3
2
cos cos cos
3
2

( ) t I i
e p q
=
1
cos (D.4)

159
Similarly, the D-axis current is given by:

( ) ( )
( ) ( )
( )
( )
(
(
(
(
(
(
(

+
|

\
|
+ + +
+ +
|

\
|
+ +
+ + +
=
(

\
|
+
|

\
|
+ +
|

\
|

|

\
|
+ =
t t
t t
t t
I i
t t t t t t I i
e e
e e
e e
p d
e e e p d
1 1
1 1
1 1
1 1 1
sin
3
4
sin ...
... sin
3
4
sin ...
... sin sin
2
1
3
2
3
2
sin
3
2
sin
3
2
sin
3
2
sin sin sin
3
2

( )
t I i
t I i
e p d
e p d

\
|
=
=
2
cos
sin
1
1

(D.5)

The D-axis and Q-axis currents in equations (D.4) and (D.5) can be represented in phasor notation as
follows:

2

=
=
p d
p q
I i
I i
(D.6)
Equation (D.6) should be interpreted as follows: the D-axis current phasor and the Q-axis current
phasor have the same magnitude, both phasors rotate anticlockwise with a rotational speed of
( )
1

e
, and the D-axis current phasor lags 90 behind the Q-axis current phasor. These phasors are
shown in a phasor diagram in Fig. D.1. The projection of these rotating phasors onto the real axis gives
the time domain equations (D.4) and (D.5).
Im
Re
1

e
d
i

q
i

Fig. D.1. D-axis and Q-axis current phasors.


160
If the fundamental frequency of the current is equal to the electrical rotational speed of the machine,
i.e.
e
=
1
, then equations (D.4) and (D.5) reduce to:

1
1
, 0
,


= =
= =
e d
e p q
i
I i
(D.7)
Equation (D.7) shows that the currents in the QD0 reference frame are DC quantities, given the
specific condition. Furthermore, the D-axis current is zero, given the specific condition. This result
might be surprising, but it does not mean that the D-axis current is always zero when
e
=
1
. The
inverse Park transform has to be considered.

Let the QD0 reference frame currents be DC quantities:
q q
K i =
d d
K i =
0
0
= i (D.8)
Now consider the inverse Park transform on equation (D.8):

( ) ( )
(
(
(

(
(
(
(
(
(
(

\
|
+
|

\
|
+
|

\
|

|

\
|
=
(
(
(

0
1
3
2
sin
3
2
cos
1
3
2
sin
3
2
cos
1 sin cos
d
q
e e
e e
e e
c
b
a
K
K
t t
t t
t t
i
i
i

(D.9)
From equation (D.9), the phase-A current is given by:

( ) ( )
( )
|

\
|
+ =
+ =
2
cos cos
sin cos

t K t K i
t K t K i
e d e q a
e d e q a
(D.10)
The phase-A current can be written in phasor notation as:

( )
q
d
d q a
d q d q a
d q a
d q a
K
K
K K i
jK K jK K i
jK K i
K K i
arctan

2
0

2 2
+ =
=
=
+ =

(D.11)
From equation (D.11), the phase-A current in the time domain is:

|
|

\
|
+ =
q
d
e d q a
K
K
K K i arctan cos
2 2
(D.12)


161
Similarly, the phase-B and phase-C currents are given by:

|
|

\
|
+ =
q
d
e d q b
K
K
K K i arctan
3
2
cos
2 2

(D.13)

|
|

\
|
+ + =
q
d
e d q c
K
K
K K i arctan
3
2
cos
2 2

(D.14)

Equations (D.12) through to (D.14) show that the currents in the ABC reference frame have identical
amplitude and frequency. The frequency is given by the electrical rotor speed and the amplitude is
given by the vector sum of the D-axis and Q-axis current components. The phase offset
q
d
K
K
arctan is
equal to the current angle , in the QD0 reference frame.

To summarize, the Park transform takes currents from the ABC reference frame to currents in the QD0
reference frame. If the currents in the ABC reference frame are in an ABC sequence (not ACB), then
the D-axis current lags 90 behind the Q-axis current in a phasor notation. It is shown that if the
frequency of the currents in the ABC reference frame is equal to the electrical rotational speed, the
currents in the QD0 reference frame are DC quantities.

If
1
=
e
, a new set of stationary axes can be chosen for phasor notation of the D-axis and Q-axis
currents. From Fig. D.1, the new set of axes is shown in Fig. D.2:
0
1
=
e
d
i

q
i

d-axis
q-axis
d
i

q
i

d-axis
q-axis

Fig. D.2. Phasor diagram for D-axis and Q-axis currents with the condition that the frequency of the
ABC currents is equal to the electrical rotational speed of the rotor.

162
Appendix E Lookup tables

This appendix shows how finite element analysis (FEA) results are used to create suitable lookup
tables (LUTs) for the torque and flux linkages of the PMA RSM. The LUTs are used to create an
accurate model of the PMA RSM for simulation. Some of the LUTs are used for the current
controllers and load torque observer in simulation and practical implementation.

The torque, D-axis flux linkage and Q-axis flux linkage of the PMA RSM are functions of the D-axis
current and Q-axis current. The torque and flux linkages can therefore be represented as 3D surfaces.
To gain better understanding of a 3D surface, a contour plot of that surface can be used. The surfaces
and contour plots of the torque and flux linkages are shown next and explained afterwards. These
graphs provide insight into the physical nature of the PMA RSM [F3].

After the explanation of the 3D graphs, a method is selected by which to obtain the LUTs. The LUTs
that are used for simulation and practical implementation of the current controllers and the reduced
order observer are shown. This is followed by a brief explanation of an alternative method to select
data for the LUTs. Finally, the difference between instantaneous and linearized inductance is
demonstrated by means of an example.


163

-500
0
500
-300
-200
-100
0
100
200
300
-1500
-1000
-500
0
500
1000
1500
D-axis current [A]
Torque surface for PMA RSM
Q-axis current [A]
T
o
r
q
u
e

[
N
m
]

Fig. E.1. Torque surface for PMA RSM
D-axis current [A]
Q
-
a
x
i
s

c
u
r
r
e
n
t

[
A
]
Torque contour [Nm]
-300 -200 -100 0 100 200 300
-300
-200
-100
0
100
200
300
-1000
-800
-600
-400
-200
0
200
400
600
800
1000

Fig. E.2. Torque contour for PMA RSM.



164
-400
-200
0
200
400
-400
-200
0
200
400
-1
-0.5
0
0.5
1
D-axis current [A]
D-axis flux linkage surface for PMA RSM
Q-axis current [A]
D
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]

Fig. E.3. D-axis flux linkage surface
D-axis current [A]
Q
-
a
x
i
s

c
u
r
r
e
n
t

[
A
]
D-axis flux linkage contour [Wb turns]
-300 -200 -100 0 100 200 300
-300
-200
-100
0
100
200
300
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8

Fig. E.4. D-axis flux linkage contour.

165
-400
-200
0
200
400
-400
-200
0
200
400
-0.6
-0.4
-0.2
0
0.2
D-axis current [A]
Q-axis flux linkage surface for PMA RSM
Q-axis current [A]
Q
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]

Fig. E.5. Q-axis flux linkage surface.
D-axis current [A]
Q
-
a
x
i
s

c
u
r
r
e
n
t

[
A
]
Q-axis flux linakge contour [Wb turns]
-300 -200 -100 0 100 200 300
-300
-200
-100
0
100
200
300
-0.5
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0

Fig. E.6. Q-axis flux linkage contour.


166
The torque surface of Fig. E.1 has the shape of an unbalanced saddle. There is a large positive torque
in the first quadrant ( 0 >
d
i and 0 >
q
i ), a large negative torque in the second quadrant ( 0 <
d
i
and 0 >
q
i ), a small positive torque in the third quadrant ( 0 <
d
i and 0 <
q
i ) and a small negative
torque in the forth quadrant ( 0 >
d
i and 0 <
q
i ). The contour plot contains all the information of the
surface plot. Each contour represents one torque value. The distance between the contours gives the
slope of the surface: contours close to each other represent a steep slope, while contours far apart
represent a gentle slope. The steepest slope represents the maximum torque per ampere.

Imagine that the torque surface in Fig. E.1 is a mountain and that a mountaineer wishes to reach the
highest possible point in the shortest time. The mountaineer travels at the same speed regardless of the
slope, but can only travel in one direction. The mountaineer possesses a contour plot of the mountain
the likes of Fig. E.2. The mountaineer sees that the highest point lies in the first quadrant and he plots
two possible paths by which he can travel as shown in Fig. E.7. The mountaineer measures the
distances between the contour lines for the path given by = 34 and also measures the distances
between the contour lines for the path given by = 54 . The mountaineer deduces that the path
given by = 54 is steeper than the path given by = 34 , because the sum of the distances between
the contours for the two paths are the same, but a higher point is reached by traveling the path given by
= 54 .

Fig. E.7. First quadrant of the torque contour plot.

It was shown in subsection 2.5 that the current angle = 54 leads to the maximum torque per
ampere. The explanation above supports the result of subsection 2.5.

167
The D-axis flux linkage surface of Fig. E.3 has the shape of a curved ribbon. The surface has the
steepest gradient at the origin. The steep gradient is echoed in the contour plot Fig. E.4, where the
lines close to each other at the origin. The Q-axis current has a demagnetizing effect on the D-axis
flux linkage: the magnitude of the D-axis flux linkage tends to be less if the magnitude of the Q-axis
current is large. This effect can be observed from the contour plot as well, and is explained next.

Consider a selection of contour lines for positive D-axis currents (Fig. E.8 left) and a selection of
contour lines for negative D-axis currents (Fig. E.8 right). A straight line for 55 =
d
i A is drawn on
the left hand plot: the region to the right of the line corresponds to more positive flux linkage, while
the region to the left of the line corresponds to more negative flux linkage. Now it can be seen that if
the Q-axis current increases in magnitude (positive), the flux linkage tends to be more negative (the
straight line is to the left of the contour). Similarly, a straight line for 132 =
d
i A is drawn on the
right hand plot: the region to the right of the line corresponds to more positive flux linkage, while the
region to the left of the line corresponds to more negative flux linkage. Now it can be seen that if the
Q-axis current increases in magnitude (negative), the flux linkage tends to be more positive (the
straight line is to the right of the contour).


Fig. E.8. Positive (left) and negative (right) selections of D-axis flux linkage contour lines.

D-axis current [A]
Q
-
a
x
i
s

c
u
r
r
e
n
t

[
A
]
D-axis flux linkage contour [Wb turns]
35 40 45 50 55 60 65 70 75
-250
-200
-150
-100
-50
0
50
100
150
200
250
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
D-axis current [A]
Q
-
a
x
i
s

c
u
r
r
e
n
t

[
A
]
D-axis flux linkage contour [Wb turns]
-200 -180 -160 -140 -120 -100 -80 -60
-250
-200
-150
-100
-50
0
50
100
150
200
250
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8

168
The Q-axis flux linkage surface of Fig. E.5 also has the shape of a curved ribbon. It is however much
flatter compared to the D-axis flux linkage surface. The surface in Fig. E.5 has the steepest gradient at
the origin. The steep gradient is echoed by the contour plot Fig. E.6, which has lines close to each
other at the origin. The D-axis current has a demagnetizing effect on the Q-axis flux linkage: the
magnitude of the Q-axis flux linkage tends to be less if the magnitude of the D-axis current is large.
This effect can be observed from the contour plot as well, and is explained again for the sale of
completeness.

Consider a selection of Q-axis flux linkage contour lines for positive Q-axis currents (Fig. E.9 left) and
a selection of contour lines for negative Q-axis currents (Fig. E.9 right). A straight line for 90 =
q
i A
is drawn on the left hand plot: the region above the line corresponds to more positive flux linkage,
while the region below the line corresponds to more negative flux linkage. Now it can be seen that if
the D-axis current increases in magnitude (positive), the flux linkage tends to be more negative (the
straight line is below the contour). Similarly, a straight line for 87 =
q
i A is drawn on the right hand
plot: the region above the line corresponds to more positive flux linkage, while the region below the
line corresponds to more negative flux linkage. Now it can be seen that if the D-axis current increases
in magnitude (negative), the flux linkage tends to be more positive (the straight line is above the
contour).

Fig. E.9. Positive (left) and negative (right) selections of Q-axis flux linkage contour lines.

D-axis current [A]
Q
-
a
x
i
s

c
u
r
r
e
n
t

[
A
]
Q-axis flux linakge contour [Wb turns]
-200 -100 0 100 200
75
80
85
90
95
100
105
110
115
120
125
-0.5
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
D-axis current [A]
Q
-
a
x
i
s

c
u
r
r
e
n
t

[
A
]
Q-axis flux linakge contour [Wb turns]
-200 -100 0 100 200 300
-130
-120
-110
-100
-90
-80
-70
-60
-50
-0.5
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0

169
From the explanation above, it is clear that the values of the torque and the flux linkages are very
much dependent on the position in the complex QD0 current plane. Equations (2.1.6) through to
(2.1.10) express the derivatives of the flux linkages as the products of partial derivatives (
'
d
L ,
'
d
M ,
'
q
L
and
'
d
M ) and derivatives of the currents (
dt
di
d
and
dt
di
q
). Graphs for the partial derivatives as
functions of currents are shown in Fig. 2.4.1. Here it can be noted that different graphs are obtained
for different working points on the QD0 current plane. It is also noted that the values for mutual
inductances are relatively small compared to the self inductances.

In chapter 3, a decoupling procedure is suggested in Fig. 3.2.2. For the decoupling procedure, LUTs
are needed for the flux linkages and mutual inductances. The current controllers are derived in
subsection 3.6; equations (3.6.7) and (3.6.8) show that LUTs are needed for the self inductances.

One method to obtain these LUTs is to assume rated working conditions for a constant current angle of
= 54 . In the FEA program, first the Q-axis current is kept constant at ) 54 sin( 200 2 =
q
i A, and
flux linkages are recorded for the D-axis currents 300 300 < <
d
i A. From these flux linkages it is
possible to obtain ) (
1 d d
i f = , ) (
2 d q
i f = , ) (
3 d
d
d
i f
i
=

and ) (
4 d
d
q
i f
i
=

, where
n
f is a
function of
d
i . Secondly, the D-axis current is kept constant at ) 54 cos( 200 2 =
d
i A, and flux
linkages are recorded for the Q-axis currents 300 300 < <
q
i A. From these flux linkages it is
possible to obtain ) (
5 q q
i f = , ) (
6 q d
i f = , ) (
7 q
q
q
i f
i
=

and ) (
8 q
q
d
i f
i
=

, where
n
f is a
function of
q
i . These results are shown in Fig. E.10.

170

Fig. E.10. FEA results of flux linkages and inductances for rated working conditions.

In Fig. E.10, the flux linkages in the top row are obtained from FEA. These flux linkages are then
differentiated to give the inductances in the second row. Similarly, the flux linkages in the third row
are obtained from FEA and differentiated to give the inductances of the fourth row. Some of the
-300 -200 -100 0 100 200 300
-1.5
-1
-0.5
0
0.5
1
1.5
D-axis flux linkage for Iq = 229 A
D
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]
D-axis current [A]
-300 -200 -100 0 100 200 300
0
0.2
0.4
0.6
0.8
1
Q-axis flux linkage for Iq = 229 A
Q
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]
D-axis current [A]
-300 -200 -100 0 100 200 300
0
2
4
6
8
x 10
-3
D-axis self inductance
D
-
a
x
i
s

s
e
l
f

i
n
d
u
c
t
a
n
c
e

[
H
]
D-axis current [A]
-300 -200 -100 0 100 200 300
-2
0
2
4
x 10
-4
Q-axis mutual inductance
Q
-
a
x
i
s

m
u
t
u
a
l

i
n
d
u
c
t
a
n
c
e

[
H
]
D-axis current [A]
-300 -200 -100 0 100 200 300
-1.5
-1
-0.5
0
0.5
1
1.5
Q-axis flux linkage for Id = 166 A
Q
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]
Q-axis current [A]
-300 -200 -100 0 100 200 300
0
2
4
6
8
x 10
-3
Q-axis self inductance
Q
-
a
x
i
s

s
e
l
f

i
n
d
u
c
t
a
n
c
e

[
H
]
Q-axis current [A]
-300 -200 -100 0 100 200 300
0
0.2
0.4
0.6
0.8
1
D-axis flux linkage for Id = 166 A
D
-
a
x
i
s

f
l
u
x

l
i
n
k
a
g
e

[
W
b

t
u
r
n
s
]
Q-axis current [A]
-300 -200 -100 0 100 200 300
-2
0
2
4
x 10
-4
D-axis mutual inductance
D
-
a
x
i
s

m
u
t
u
a
l

i
n
d
u
c
t
a
n
c
e

[
H
]
Q-axis current [A]

171
graphs in Fig. E.10 appear to be noisy: this is because the FEA is a numerical approach and there are
quantization errors. Note that the inductances are supposed to be smooth graphs. Polynomial fitting
can be used to approximate the graphs for the flux linkages and these polynomials can then be
differentiated to obtain the inductances. However, polynomial fitting also introduces errors, so this
approach is not followed here.

LUTs for the D-axis flux linkage, Q-axis flux linkage and D-axis inductance are needed for the current
controllers. Practical implementation of the LUTs in the control program requires that the measured
current be used as an index to an array (vector). For example, 81 samples of the D-axis flux linkage
are stored in an array. The index to the array is therefore from 0 to 80. The entry at index 0
corresponds to the D-axis flux linkage for 300 =
d
i A, and the entry at index 80 corresponds to the
D-axis flux linkage for 300 + =
d
i A. The D-axis current is measured and the index is then calculated
as follows:
5 . 7
300 +
=
d
i
index . The index is a floating point number, but is rounded to the nearest
integer to be used as an index in an array.

Lookup tables for the torque as a function of stator current, with a constant current angle as well as the
inverse function are still needed. FEA results are obtained for a constant current angle of = 54 and
for 300 0 < <
s
i A. The results are shown in Fig. E.11.

Fig. E.11. Torque and stator current conversions for a constant current angle.

The graphs in Fig. E.11 are implemented as LUTs in the control program in a similar fashion as
explained for the D-axis flux linkage. Additional logic is needed however, because the graphs are only
one-sided. The torque is an odd function of the stator current: for a given positive stator current, the
positive torque is given by 0 ), (
9
> =
+
s s em
i i f T ; for a negative stator current of the same magnitude,
the negative torque is given by 0 ), (
9
> =

s s em
i i f T .
50 100 150 200 250 300
0
200
400
600
800
Torque vs. Stator current
Stator current [A peak]
T
o
r
q
u
e

[
N
m
]
0 200 400 600 800 1000
0
100
200
300
Stator current vs. Torque
S
t
a
t
o
r

c
u
r
r
e
n
t

[
A

p
e
a
k
]
Torque [Nm]

172

The flux linkage and inductance LUTs are obtained by assuming rated working conditions for a
constant current angle of = 54 . The Q-axis current is kept constant at rated conditions when the D-
axis flux linkage graph is obtained; the D-axis current is kept constant at rated conditions when the Q-
axis flux linkage graph is obtained. This method however is not truly accurate when constant current
angle control (CCAC) is used to control the PMA RSM. Although the LUTs were obtained using the
rated condition method described above, it is suggested that the flux linkage graphs be obtained as a
functions of the stator current with a constant current angle for future reference. This method is
explained briefly.

Equations (2.1.6) through to (2.1.10) express the derivatives of the flux linkages as the product of
partial flux linkage derivatives and derivatives of the independent variables. For example:

dt
d
dt
di
i dt
di
i dt
d
e
e
d
q
q
d d
d
d d

= (E.1)
It is assumed that the change in flux linkage due to the rotor position
e
is negligible:

dt
di
i dt
di
i dt
d
q
q
d d
d
d d

(E.2)
Equation (E.2) can also be written as:

dt
di
M
dt
di
L
dt
d
q
d
d
d
d ' '
+

(E.3)
Equation (E.3) expresses the derivative of the flux linkage in terms of independent variables in
Cartesian form. An alternative approach, in polar form ) , (
s d
i f = , is shown next. Assuming that
the change in flux linkage due to the rotor position is negligible:

dt
d
dt
di
i dt
d
d s
s
d d

(E.4)

For CCAC however, the current angle is kept constant; therefore equation (E.4) reduces to:
=

= 54 ,

dt
di
i dt
d
s
s
d d
(E.5)
The D-axis current controller in subsection 3.7, however requires
'
d
L . The stator current in equation
(E.5) is written again in terms of Cartesian components:

173

dt
di
i dt
d
dt
i
d
i dt
d
d
s
d d
d
s
d d
|
|

\
|

=
=
|
|

\
|

54 cos
1
54 ,
cos
(E.6)
Now, the D-axis self inductance is defined as:

s
d
d
i
L

cos
1
'
(E.7)
FEA can be used to obtain the result for equation (E.7) and this result can be used in the D-axis current
controller of subsection 3.7.

Note that if CCAC is used and the flux linkages are obtained as functions of the stator current by
keeping the current angle constant, there is no approximation. With accurate
d
and
q
, the speed
voltage terms of equations (3.2.1) and (3.2.2) can be calculated accurately. In this scheme it is not
necessary to know the mutual inductance, because the flux linkage graphs incorporate this knowledge.
This would make the decoupling procedure suggested in subsection 3.2 very accurate.

The mutual inductances are very small and therefore the method of obtaining the LUTs by assuming
rated working conditions is acceptable. For a PMA RSM that is controlled with CCAC and has large
mutual inductances, the second approach of obtaining LUTs should be used.

As a final remark, the difference in instantaneous self inductance and linearized self inductance is
pointed out by taking the D-axis self inductance as an example. The difference is clearly shown in
Fig. E.12.
-400 -300 -200 -100 0 100 200 300 400
0
1
2
3
4
5
6
7
8
x 10
-3
D-axis self inductance
D
-
a
x
i
s

s
e
l
f

i
n
d
u
c
t
a
n
c
e

[
H
]
D-axis current [A]
instantaneous
linearized

Fig. E.12. Illustration of the difference between instantaneous and linearized D-axis self inductance.

174
Appendix F Signal to noise ratio

The signal to noise ratio (SNR) is the ratio of the power in the signal to the power in the noise.
The representation of a measured quantity can be represented by:
) ( ) ( ) ( t n t x t q + = , (F.1)
where ) (t x is the desired signal and ) (t n is the noise. For a signal ) (t x with power spectral
density ) ( f S the total power is given by [B4, p.50]:

= =
T
T
T
s
dt t x
T
df f S P
2
) (
2
1
lim ) ( (F.2)
The noise ) (t n is a random process. The power in a random process is given by the
autocorrelation function evaluated at the origin [B14, p.194]:
[ ] ) ( ) 0 (
2
t n E R P
nn n
= = , (F.3)
where ) ( )] ( [ t n t n E = denotes the expected value or mean of the variable ) (t n . It is assumed
that the mean of the random process is zero (sometimes ) (t n is positive, but at other times it is
negative, so the average value or mean is zero):
0 ) ( )] ( [ = = t n t n E (F.4)
The definition of variance is given by [B14, p.82]:
( ) [ ]
2 2
) ( ) ( t n t n E
n
= (F.5)
Combining equations (F.3) through to (F.5), the noise power is given by the variance. If ) (t x
is a DC signal, equation (F.2) shows that the power in the signal is given by the mean squared.
Since the variance of a DC signal is zero and the mean of the noise is zero, the SNR for the
measured quantity is given by:

)} ( var{
)} ( { )] ( [
2
t q
t q mean t q E
SNR
q
= =

(F.6)


175
Appendix G Full state observer

The derivation of the full state observer is shown here. Observer structures are discussed in [B7,
p687], [B9, p.541] and [A16]. The states that will be observed are the electrical speed and the torque
load. Observed states (variables) are indicated with a circumflex accent ^. The states are represented
collectively by x and the measured states by y.

Start with the differential equation for the produced torque and mechanical system:
(

=
(

=
+ + =
+ + =
p
B T i K
J
p
p
B T T
J
p
dt
d
p
B
p dt
d
J T
B
dt
d
J T T
e eq L s T
e eq L em
e
e eq
e
eq L
m eq
m
eq L em
1
1
1 1

(G.1)
It is assumed that the torque load is constant, although this is not necessary for the proper functioning
of the observer:
0 =
dt
dT
L
(G.2)
Equation (G.1) and (G.2) can be written in state space form [A20]:
[ ]
(

(
(
(

+
(

(
(
(

=
(

L
e
s
T
eq
L
e
eq eq
eq
L
e
T
y
i
K
J
p
T
J
p
J
B
T

0 1
0 0 0
&
&
(G.3)
Equation (G.3) shows that only the speed can be measured. The states which are chosen to be
observed are speed and torque load. Equation (G.3) can be written as:

Hx y
Gu Fx x
=
+ = &
(G.4)
A very important question is whether it is possible to observe the states that have been chosen. The
observability matrix provides the answer. If the inverse of this matrix exists [B9, p.547], or
equivalently, the rank of the matrix is equal to the number of states to be observed [A16], then it is
indeed possible to construct an observer.


176
The observability matrix is given by [B9, p.547]:

(
(
(
(

=
1
...
n
HF
HF
H
O (G.O1)
Substituting equation (G.3) into equation (G.o1), the observability matrix is given by:

(
(

=
eq eq
eq
J
p
J
B
O
0 1
(G.O2)
From equation (G.o2), all the rows of the matrix are linearly independent, i.e. the matrix is non-
singular and therefore the inverse exists. Equivalently, the rank of the matrix is equal to the number of
states to be observed. In fact, the inverse of the matrix is given by:

(
(

p
J
p
B
O eq eq
0 1
1
(G.O3)

The observer structure is given by equation (G.5). This equation shows that the output of the real
plant, Hx , is compared to the output of the plant model, x H , and the error is used to correct the model
described by F and G. The time constant in which the correction of the plant should be done is chosen
by selecting an appropriate gain vector L.
( ) x H Hx L Gu x F x

+ + = & (G.5)
Equation (G.5) can be written as:
LHx Gu x LH F x + + = ) (

& (G.6)
The dynamic response of the system, described by equation (G.6), is given by its characteristic
equation (G.7). In equation (G.7), is the determinant of .

[ ]
2 1
2
1
2
1
0 1
0 0
0
0
) (
L
J
p
L
J
B
s s
s L
J
p
L
J
B
s
L
L
J
p
J
B
s
s
LH F sI
eq eq
eq
eq eq
eq
eq eq
eq

|
|

\
|
+ + =
+ +
=
(

+
(
(
(

+
(

=
(G.7)


177
The characteristic equation (G.7) can be assigned to any second order polynomial in s:

2 2
2 1
2
n n
eq eq
eq
s s L
J
p
L
J
B
s s + +
|
|

\
|
+ + (G.8)
Choose for example a natural undamped frequency 100 =
n
and a damping coefficient 8 . 0 = .
Substituting these values into equation (G.8) and solving for L:

6666
160 2
2
2
1
=
=
n
eq
eq
eq
n
p
J
L
J
B
L

(G.9)
These values can be verified using Matlab: first the state space matrixes are created and then the
acker formula is used to calculate the gains.
J = 2;
p = 3;
B = 0.1;
F = [-B/J -p/J; 0 0];
G = [-p/J 0];
H = [1 0];
pe = roots([1 2*0.8*100 100^2]);
L = acker(F,H,pe);
L
1.0e+003 *
0.1599
-6.6667


From equation (G.5):
( )
e e s
T
eq
L
e
eq eq
eq
L
e
L
L
i
K
J
p
T
J
p
J
B
T

0 0

2
1

+
(
(

+
(

(
(
(

=
(
(

&
&
(G.10)
The State Space form can be transformed to time domain equations:

( )
( )
e e
eq
s T L e
eq
e
eq
e e s T
eq
L
eq
e
eq
eq
e
L
p
J
i K T
p
B
dt
d
p
J
L i K
J
p
T
J
p
J
B

1
1
+ +

=
+ +

= &
(G.11)

( )
( )

=
=
dt L T
L T
e e L
e e L

2
2
&
(G.12)


178
Equation (G.11) and (G.12) can be represented in block diagram form, shown in Fig. G.1.
Tem
Tload^
We^
1
s
1
s
Integrator
L2
J/p*L1
Beq/p
p/Jeq Kt
2
We measured
1
Is measured

Fig. G.1. Full state observer.



179
Appendix H Additional simulation results

The simulation results shown in this appendix are comparable to the practical results. The
load torque in the simulation is adjusted to resemble the practical load torque and noise is
added to the speed signal.

Simulation results for the system with the reduced state observer are shown first (Fig. H.1 to
Fig. H.4) and this is followed by simulation results for the system with the full state observer
(Fig. H.5 to Fig. H.8). In both cases, simulation results with and without compensation
current feedback are shown.

From the simulation results, note that the stator current has more noise when the
compensation current is used for feedback. This result is also found in the practical
measurements. The speed overshoot, when no compensation is used, also matches the
practical result. This indicates that the simulation model is accurate and may be used to test
the performance of the system under various load conditions e.g. a step load change, a load
pulse etc.

All the figures in this appendix adhere to the table below. The scales described in the table
are the same scales used for the practical measurements in Chapter 5.

Position Description y-axis range x-axis range
Top, left Electrical speed error -80 to 80 rad/sec 10 to 30 sec
Top, right Observed load torque -1500 to 1500 Nm 10 to 30 sec
Bottom, left Stator current -400 to 400 A 10 to 30 sec
Bottom, right Compensation current -400 to 400 A 10 to 30 sec




180

Fig. H.1. Reduced state observed without compensation current feedback, noise in the speed
measurement and load torque from 700 Nm to 0 Nm within 1 second.

Fig. H.2. Reduced state observed with compensation current feedback, noise in the speed
measurement and load torque from 700 Nm to 0 Nm within 1 second.

181

Fig. H.3. Reduced state observed without compensation current feedback, noise in the speed
measurement and load torque from 0 Nm to 700 Nm within 4 seconds.

Fig. H.4. Reduced state observed with compensation current feedback, noise in the speed
measurement and load torque from 0 Nm to 700 Nm within 4 seconds.

182

Fig. H.5. Full state observed without compensation current feedback, noise in the speed
measurement and load torque from 700 Nm to 0 Nm within 1 second.

Fig. H.6. Full state observed with compensation current feedback, noise in the speed
measurement and load torque from 700 Nm to 0 Nm within 1 second.

183

Fig. H.7. Full state observed without compensation current feedback, noise in the speed
measurement and load torque from 0 Nm to 700 Nm within 4 seconds.

Fig. H.8. Full state observed with compensation current feedback, noise in the speed
measurement and load torque from 0 Nm to 700 Nm within 4 seconds.