Modeling and Testing of Line Start

Permanent Magnet Motors

TOMAS MODEER

Licentiate Thesis
Stockholm, Sweden 2007
The front cover page graph illustrates a measured start-up and
synchronization of a Line Start Permanent Magnet Motor.

TRITA-EE 2007:049 Electrical Machines and Power Electronics

ISSN 1653-5146 School of Electrical Engineering, KTH
Teknikringen 33
ISBN 978-91-7178-749-1 SE-100 44 Stockholm
SWEDEN

Akademisk avhandling som med tillstånd av Kungl Tekniska

högskolan framlägges till offentlig granskning för avläggande
av teknologie licentiatssexamen i elektriska maskiner och effek-
telektronik fredagen den 28 september 2007 klockan 10.15 i H1,
Teknikringen 33, Kungl Tekniska högskolan, Valhallavägen 79,
Stockholm.

c Tomas Modeer, September 3, 2007

Printed by Universitetsservice US AB
Abstract

This licentiate thesis describes the modeling and measurements

performed with the aim of developing design guidelines for Line
Start Permanent Magnet Synchronous Motors, LSPMSMs.
LSPMSMs can offer higher efficiency than standard induction
motors used in the industry today, especially for small motor sizes.
The increase in efficiency results in lower environmental impact
and reduced electricity cost. The LSPMSM has, however, several
drawbacks, among the most important is the reduced starting ca-
pability compared to induction motors. Furthermore, the rotor con-
struction is complex and the added cost of magnet material makes
the LSPMSM a comparably expensive motor type.
The design of a LSPMSM is a trade-off between starting ca-
pability and steady state performance. The thesis discusses these
trade-offs and the models that can be used as a basis for design
and optimization. The models make use of different motor param-
eters, and a number of measurement methods for measuring these
parameters are described and compared. Among these is a step re-
sponse adapted measurement method that provides most of the pa-
rameters of interest.
The development and setup of a brake bench for measuring both
start-up and steady state performance is presented. Furthermore
the start-up behavior and steady-state performance is calculated
using measured parameters. The calculated performance is com-
pared to measured performance and found to correlate well for
nominal operating conditions. Thus, design guidelines can be based
on the models proposed.

Keywords: line start permanent magnet, flux integration, step re-

sponse measurement.
Acknowledgment

First, I would like to thank my supervisors Prof. Chandur Sadaran-

gani and Dr. Juliette Soulard for the help and support during the
project.
Secondly, I am greatly indebted to Grundfos and the people tak-
ing care of me during my stay in Bjerringbro. For the professional
and open welcome I would like to thank Finn Jensen, Henrik Oer-
skov, Jan Harding Gliemann and Pierre Vadstrup.
Sincere gratitude to Prof. Hans-Peter Nee for all the inspiring
discussions and for his positive and academic interest.
Thanks goes to my fellow, and former, PhD colleagues at the de-
partment. Thanks to Jörgen for being a good friend and helping me
out with latex and stuff. Special thanks to Dmitry Svechkarenko
for sharing the office peacefully and teaching me the details of the
Swedish language.
I would like to thank my friends at the electronics lab ELAB,
especially Johan Uusijärvi who has been a very good friend and
patient listener during our discussions.
Finally, I would like to thank my family, especially my mother
who has supported me in my decisions and helped me in numerous
ways.
Contents

1 Introduction 1
1.1 Drive systems . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The Line Start Permanent Magnet Motor . . . . . . . 4
1.3 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Main Contributions . . . . . . . . . . . . . . . . . . . . 7

2 Models and Design 9

2.1 Sizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Steady state models . . . . . . . . . . . . . . . . . . . . 11
2.3 Start-up models . . . . . . . . . . . . . . . . . . . . . . 19

3 Equivalent Circuit Parameter Measurement 23

3.1 Static measurements . . . . . . . . . . . . . . . . . . . 24
3.2 Synchronous measurements . . . . . . . . . . . . . . . 42
3.3 Comparison . . . . . . . . . . . . . . . . . . . . . . . . 46

4 Measured Performance 51
4.1 Starting performance . . . . . . . . . . . . . . . . . . . 51
4.2 A nominal start . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Starting Capability . . . . . . . . . . . . . . . . . . . . 67
4.4 Magnet Braking Torque . . . . . . . . . . . . . . . . . 68
4.5 Starting Current and Fuse Concerns . . . . . . . . . . 71
4.6 Steady state performance . . . . . . . . . . . . . . . . 74

5 Calculated Performance and Comparison to Mea-

surements 79
5.1 Starting Performance . . . . . . . . . . . . . . . . . . . 79
5.2 Synchronization . . . . . . . . . . . . . . . . . . . . . . 83
5.3 Steady state performance . . . . . . . . . . . . . . . . 85

6 Conclusions and Future Work 91

A List of Equipment 93
B List of Symbols 95

C Abbreviations 97

Bibliography 99

References 100
Chapter 1

Introduction

The energy consumption in the world is increasing. As of 2004 the

global power consumption was 15 TW [3] and most of this energy
comes from fossil fuels. This consumption of fossil fuels has nega-
tive environmental impact and as an effect of this, the environmen-
tal concerns are also increasing. Measures to reduce the impact on
our environment are being taken both on national scales and on
larger scales such as the Kyoto Protocol update to the UNFCCC1 .
Only a fraction, or 1.8 TW [2], of the global power consumption is
electrical energy. Distribution of energy by electrical means is in
principle environmentally friendly, despite the fact that the con-
version to and from other energy types has limited efficiency. Still,
the overall efficiency including the transport of energy can be com-
parably high. A large part of the electric energy produced is used
to provide physical motion and thus is used by electrical motors in
various drive systems. The efficiency of such drive systems varies
greatly. It is of course of great interest, not only for environmental
concerns but also economical, to improve the overall efficiency. The
purpose of this study is to make a contribution to the improvement
of the efficiency of electrical motors in drive systems.

1.1 Drive systems

Today there is an increasing number of electrical drives used
throughout society. An electrical motor has several advantages
compared to other primary movers such as hydraulic and pneu-
matic actuators and combustion engines. Some of the major advan-
tages of electrical drives compared to other drive alternatives are:

• Low maintenance, modern electrical motors can be expected

1 United Nations Framework Convention on Climate Change
2 1. Introduction

to operate for several years without need for refueling, lubri-

cation or cleaning.

• Controllability, an electrical drive offers control to a degree

not attainable by other means.

• Cleanliness, an electrical drive has no exhaust gases and gen-

erally no exposed lubricated surfaces and can thus be used in
applications sensitive to contamination.

The drive systems can be classified as either variable or constant

speed drives. Variable speed drives are used in applications such as
robotics, traction, processing, manufacturing etc. Constant speed
drives are used mainly for various types of mass transfer such as
fans, pumps and conveyor belts. For each type of application there
are several alternative motor types that can be used.
There are motor types that are normally only used for one of the
drive systems and others which are common both in variable and
constant speed drives. Two of the more common motor types used
in constant speed drives are induction, or asynchronous, motors
and synchronous motors. Induction motors have traditionally been
used only in constant speed drives but are gaining popularity in
variable speed drives thanks to their ruggedness. Induction motors
dominate the market for constant speed drives of moderate power.
For high power installations synchronous motors are more popular
as they allow the reactive power to be controlled.

Torque Production
The models and theories used to describe the force phenomena ob-
served between multiple current conducting coils or between coils
and permanent magnets can take many forms. The choice of which
to use or which to consider "correct" is a matter of taste.
A common approach when analyzing electrical machines is to
consider one set of the coils or permanent magnets as providing a
magnetic field in which another set of coils produce a force when
current is passed through them. The former coils are then called
field coils and the latter, force producing coils are called armature
coils. As there is a counterforce to the force acting on the armature
that acts on some other part of the machine the division between
field and armature is often a synthetic one.

Induction Motors
In most electrical machine types the magnetic field, in which the
armature will produce torque, is provided either by a separate field
Drive systems 3

winding or by a permanent magnet. In an induction motor, IM,

however, the field arises from currents induced in the rotor wind-
ing, which in a way can be viewed as a field winding. The rotor
winding currents are induced by transformer action from the ar-
mature winding. The armature winding thus carries both armature
and field current components.
The field current component increases the total current magni-
tude and there will be larger resistive losses than if only the arma-
ture current component was present. The magnitude of the needed
field current component depends on the reluctance of the flux path
linking the stator and rotor winding. A large part of this reluctance
is due to the air-gap between the rotor and stator. To get a low re-
luctance the air-gap must be very small, which in turn means tight
mechanical tolerances. Furthermore, to achieve high efficiency, a
high grade electrical steel is needed that can provide a high per-
meability while keeping the iron losses acceptable. Improving ei-
ther of these factors increase the production costs of the induction
machine and there is, naturally, a trade-off between efficiency and
total cost of the motor.
High efficiency motors reduce the electricity cost for the opera-
tor and thus the higher acquirement cost can be accounted for in
a matter of months or years depending on the motor efficiency and
cost and the duty ratio of operation. There are however obstacles
to introduce higher efficiency motors. One major reason is that the
extra investment cost is not taken from the same account, or even
the same company when subcontractors manage new installations,
and thus the saving on the electricity bill is of no concern to the
purchaser. Furthermore the classification of motor efficiency can
vary between different manufacturers and the actual efficiency in
operation is very dependent on local environment factors such as
temperature, line voltage and load torque. To improve this situa-
tion in the EU the DG TREN2 with cooperation from CEMEP3 , has
developed a database of high efficiency three-phase squirrel cage
induction motors, EuroDEEM 4 , which not only provides a common
database of a large selection of motors available in Europe but also
the efficiency of the motors measured in a consistent manner.
EuroDEEM sorts the motors into three different efficiency clas-
sifications, EFF1, EFF2 and EFF3, where the limit efficiencies de-
pend on the output power and number of poles. The required effi-
ciency is lower for lower power and pole number motors. The lowest
power included in the classification is 1.1 kW at which the require-
2 Directorate-Generalfor Energy and Transport of the European Commission
3 European Committee of Manufacturers of Electrical Machines and Power Elec-
tronics
4 European Database of Energy Efficient Electric Motors
4 1. Introduction

ment for a four pole motor is ≥ 83.8 % for a EFF1 classification and
76.2 % for a EFF2.
The efficiency of small induction motors is low and it is costly
to improve. This is to a large degree due to the fact that there is a
limit on how small the air-gap can be allowed to be. The tolerances
on the die stamping of the laminations is fairly low, so to improve
matters the rotor is generally machined as a final step before as-
sembly. Boring stator would improve tolerance of the stator but it
is prohibitively costly and machining problems due to the open sta-
tor slots may arise. Even if such measures are taken there must
still be a substantial air-gap to allow for flexing of the shaft and
to limit the windage losses. The efficiency thus drops with reduced
motor size as is shown in [5], a fact that is also evident in the Eu-
roDEEM classifications. For small induction motors the efficiency
is thus limited and other motor types can provide superior perfor-
mance.

Permanent Magnet Synchronous Motors

Permanent magnet synchronous motors, PMSM, can have larger
air-gaps without affecting the efficiency to the same degree as in
induction motors. With larger air-gaps the required amount of mag-
net material, and the associated cost, increases but for small motor
sizes the magnet cost is not so troublesome. For small motor sizes
the PMSM can provide a high efficiency at a competitive motor cost.
The PMSM however requires a variable frequency drive, VFD, for
the start sequence. This is generally considered too expensive for
constant speed applications and also reduces the overall efficiency.

1.2 The Line Start Permanent Magnet Motor

The idea of combining the high efficiency of the PMSM with sim-
plicity and starting ability of an IM drive dates back to the 1950’s
[7]. The combination is generally known as Line Start Permanent
Magnet Synchronous Motor, LSPMSM, due to the ability to start
when connected directly to the line (mains).
The LSPMSM has had limited market penetration, probably
due to a number of factors, some of which are:

- The extra cost of magnet material compared to IMs.

- The complex rotor construction and hence increased produc-

tion cost.

- The intricate machine design, making it hard to optimize.

The Line Start Permanent Magnet Motor 5

A major reason is probably that the motor market is fairly conser-

vative and that there has been little incentive to develop high effi-
ciency motors. The cost of high performance magnets has decreased
since the introduction of neodymium magnets in the 1980’s but it
is not until the last decade that LSPMSMs have become available
in a commercial setting.

General characteristics
The LSPMSM can be considered a combination of a permanent
magnet synchronous motor and an asynchronous, induction motor.
At steady state operation the LSPMSM differs little from the op-
eration of other permanent magnet synchronous motors. The rotor
winding functions as a damper, or amortisseur.
At start-up the LSPMSM is accelerated by an asynchronous
torque in very much the same way as an induction motor at startup.
The LSPMSM however suffers from braking torque due to cur-
rents induced by the magnets. Furthermore the magnetic saliency,
which can be quite significant, negatively affects the induction mo-
tor properties of the LSPMSM. Both these effects reduce the start-
ing capability.
Initial designs of LSPMSMs are often based on analytical mod-
els. These models utilize simplifications making the predicted be-
havior and performance inaccurate. More accurate modeling using
Finite Element Methods can include spatial effects and non-linear
magnetic properties in a very powerful way. The drawback is sim-
ulation time and model complexity. An analytical model can give
more insight into what needs to be changed to meet specifications.

Design Specifications
LSPMSMs are mainly an alternative to induction motors and in
many respects there are similar limitations and obstacles affecting
both types. Accommodating a wide supply voltage range and main-
taining high efficiency over the range sets costly requirements on
the dimensions and machining of the motor.

Efficiency and Power Factor

High efficiency is the major reason for choosing a LSPMSM so
naturally there are requirements on the efficiency. High efficiency
is attractive as the motor can be potentially more environment
friendly but the main driving reason is reduced total cost. High
efficiency reduces the electricity cost for the end-consumer but if
the power factor is low the cost can still be unnecessarily high.
6 1. Introduction

Current harmonics can generally be kept acceptably low to meet

regulatory standards and not affect the power factor considerably.
On the other hand, accommodating a wide supply voltage range it
is hard to keep the displacement power factor high over the whole
voltage range. The efficiency and displacement power factor, DPF,
are interrelated as the losses in the motor to a large degree are
current-dependent. A DPF lower than one means that the current
is larger than needed and that there are associated, unnecessary
losses in the stator winding.

Starting Capability
The motor must be able to start under all specified starting con-
ditions. A typical load specification includes a rotational inertia as
well as one or more speed dependent torque components.

Starting Current
The high current levels during starting must not blow any fuses.

Torque Pulsations
Torque pulsations put the mechanical system the motor is con-
nected to under extra stress. Axles and impellers risk breaking due
to repetitive strain. Furthermore, the risk of cavitation in pump ap-
plications is increased. The torque pulsations during the early start
need however not necessarily be a disadvantage. In certain applica-
tions, such as impeller pumps subject to debris, the pulsations can
help free the impeller from obstacles that would otherwise prevent
a successful start.

1.3 Thesis Outline

The work described in this report aims to improve the understand-
ing and thereby performance of LSPMSMs. The performance of a
LSPMSM can be predicted by different models which are discussed
in Chapter 2. To verify the analytical models the equivalent circuit
parameters of a small LSPMSM have been measured and used to
predict its performance. There are numerous ways to measure the
parameters, the methods and results of these measurement are
presented in Chapter 3.
The measured performance of the motor is presented in Chapter
4 and compared to predicted performance in Chapter 5.
A significant amount of time has been spent on Finite Element,
FE, simulations, some aimed at calculating equivalent circuit pa-
Main Contributions 7

rameters, others to verify measurement methods and lastly also

some to determine certain motor performance. These simulations
and their advantages and shortcomings depend a lot on the FE
software used and are not discussed further in this thesis.
During the project the equivalent parameters have been calcu-
lated from motor dimensions, material data etc. These calculations
are not described herein as there is little novelty about them and
the reader is better referred to existing literature.

1.4 Main Contributions

The main contributions of this thesis are:

- It is shown that static, i.e. not rotating, DC-excited measure-

ments can provide most of the equivalent circuit parameters
of interest. Measurements using the proposed static method
and measurements at synchronous speed show good correla-
tion while the static measurements provide significantly bet-
ter accuracy in the d-axis parameters for low current magni-
tudes.
- A method to measure the rotational inertia of a motor and
load setup is presented. It provides a high accuracy measure-
ment of the inertia even with significant friction present. Fur-
thermore, it is shown that the torsional stiffness of the setup
limits the amount of inertia that can be emulated by the load
motor. For reasonable shaft and coupling stiffness the usable
range of inertia is very limited.
- It is shown that measuring the braking torque of a short cir-
cuited motor can provide information on the transient magnet
braking torque, its impact on the synchronization capability
and the flux distribution between the different layers in the
motor.
Chapter 2

Models and Design

Various models can be used to predict performance and behavior of

the LSPMSM during the design phase of development. The accu-
racy and precision of the prediction has to be weighed against the
complexity of the model. Finite Element models can achieve very
accurate results and take into account numerous phenomena at the
price of high complexity and long simulation times. Simpler mod-
els based on equivalent circuits offer shorter calculation times and
rapid design iterations. To keep the model complexity low, different
models are used for different operating modes. Separate models are
used for predicting the start-up and the steady state behavior. The
different models can be used to investigate:
Steady state performance such as power factor, efficiency and
load torque capability.
Start-up capability and behavior such as allowable load iner-
tia and starting current magnitude.
Stability analysis such as evaluating angular stiffness and
damping factors.
The optimization of a motor design regarding these properties puts
incompatible requirements on the design. A finished design is thus
a trade-off between mainly the steady state performance, starting
capability and production cost.

2.1 Sizing
In general the space that is available for a LSPMSM is limited,
both in volume and shape. If this is not the case the motor design
is still limited in size by cost of materials. Applications that are not
limited by either space or cost are unknown to the author and in
10 2. Models and Design

Figure 2.1: Cross section of a line start permanent magnet syn-

chronous motor, LSPMSM. Scale 1:2, stator outer diameter is
130 mm.

general both are tightly constrained. Most LSPMSMs can be con-

sidered replacements for induction motors and as such will have to
fit given frame sizes, axle heights, flanges and mounts. The sizing
of a LSPMSM follows roughly the same methods as used for induc-
tion or synchronous motors. The efficiency of induction motors is
comparably low, especially for smaller motor sizes, typically in the
range 70 − 80% according to Gustavson [5]. The induction motor
size is thus limited in many applications by the losses that must be
accounted for, most importantly the rotor losses that have a large
thermal resistance to the ambient temperature. The LSPMSM,
having virtually no rotor losses, can therefore be of smaller size for
a given load torque. A LSPMSM replacing an IM of the same size
would have a greater torque rating than the IM, which in general
is no drawback but also seldom a benefit. The extra size margin
can instead be used for improved efficiency as it is the prime de-
sign target for a LSPMSM. The LSPMSM has however rotor losses
during the start sequence, and thus if the application includes a
multitude of large inertia starts in short time spans, this must be
taken into account when determining the required motor size. In
such applications the benefits of a LSPMSM are less pronounced.
Steady state models 11

2.2 Steady state models

For evaluating and analyzing the steady state performance an
equivalent circuit d-q model as used by Binns [19] , shown in [Fig-
ure 2.2], and the corresponding phasor diagrams (harmonic vec-
tors) provide a powerful tool [1] [11]. Evaluating the effect of supply
voltage, load torque and equivalent parameter variations on dis-
placement power factor and efficiency are among the prime uses.
In true steady state operation there are no currents induced in the

ωr Ψsq L′lrd
Rs Llsd
id
− +
+

usd Lmd R′rd

Llsq ωr Ψsd L′lrq

iq Rs
+ −
+

usq Lmq R′rq

Figure 2.2: Transient d-q equivalent circuit.

rotor cage and thus the rotor branches need not be included. Fur-
thermore, the voltages over the inductances are zero, but they are
included to show the coupling between the d- and q-axis:

Ψsd = Lmd id + Ψm (2.1)

Ψsq = Lmq iq (2.2)

Combining the leakage and magnetizing inductances into syn-
chronous inductances, Ls ,

Lsd = Llsd + Lmd (2.3)

Lsq = Llsq + Lmq (2.4)

yields a simplified equivalent circuit shown in [Figure 2.3]. A pha-
sor diagram for a typical LSPMSM operation is shown in [Figure
2.4].
In the following analysis the model is simplified further by ne-
glecting the stator resistance as it is generally much smaller than
the synchronous reactances.
12 2. Models and Design

Rs Lsd
id
+

−
usd +
ωr Lsq iq

Rs Lsq
iq
+

+
usq ωr (Lsd id + Ψn )
−

Figure 2.3: Simplified steady state d-q equivalent circuit.

jXq Iq

jXd Id
Us

Rs Is

δ
ϕ Is
q
d

Figure 2.4: Typical phasor diagram for a salient synchronous mo-

tor.

From the simplified phasor diagram [Figure 2.5] it is evident

that the difference between the terminal voltage vector, Us , and
the induced voltage, E, must thus be spanned by the voltage jXs is ,
regardless of the magnitudes of Us and E and their phase relation:

Us ≈ E + jXs Is (2.5)
Steady state models 13

|is | = const.
+

jXs Is
E

Us

Is
E
Us

ϕ
δ δ

q q
d d

Figure 2.5: Simplified phasor Figure 2.6: Phasor diagram,

diagram constant current magnitude
contours.

Current magnitude and torque production

An increase in the load angle, the angle between Us and E, gen-
erally implies an increase in the magnitude of the stator current
to satisfy Equation 2.5. In LSPMSMs in general the q-axis induc-
tance is larger than the d-axis inductance due to the low perme-
ability of the permanent magnets. Thus to span a voltage in the d-
direction requires less current than a corresponding voltage along
the q-axis. The supply voltage phasors that yield a certain sup-
ply current magnitude thus forms ellipses around E as shown in
[Figure 2.6]. The major axis of these equi-current ellipses are per-
pendicular to the internal voltage, E, as the q-axis reactance is the
larger than the d-axis reactance. If the effect of the stator resis-
tance is included the major axis will not be exactly perpendicular
to E, but tilted somewhat in the clockwise direction due to the re-
sistive voltage drop.
The synchronous torque, consisting of magnet and a reluctance
torque components:
TS = TM + TR (2.6)
is calculated using expressions described by Miller [14]. The ex-
pression for the magnet torque is
3 p Us E
TM = sin(δ) . (2.7)
2 ωs Xmd
14 2. Models and Design

+
|Tm | = const.
Trel = const.

E Us E
Us
+

δ δ

q q
d − + d
+ −

Figure 2.7: Phasor diagram, Figure 2.8: Phasor diagram,

constant magnet torque con- constant reluctance torque con-
tours. tours.

The factor Us sin(δ) shows that the magnet torque depends on the
projection of the stator voltage onto the d-axis and thus the magnet
equi-torque contours form vertical lines as shown in [Figure 2.7].
The expression for the reluctance torque is
 
3 p Us2 1 1
TR = − sin(2δ) (2.8)
2 ωs 2 Xmq Xmd

and the reluctance equi-torque contours have more complex ap-

pearance as shown in [Figure 2.8].

Efficiency and displacement power factor

The losses in a LSPMSM are due to loss phenomena of different ori-
gins and depends on numerous factors. The windage and friction
losses depend only on the rotor speed, the iron losses are mainly
dependent on the supply frequency and voltage and the lamina-
tion material. The copper losses, which are a large part of the total
losses, depend only on the supply current magnitude. The efficiency
is thus very dependent on the current magnitude and in turn also
Steady state models 15

the displacement power factor, DPF, i.e. a large part of the losses
are only current magnitude dependent and to minimize these, the
DPF should be maximized.
To have a low current consumption at no load torque the mag-
nitude of Us should equal the magnitude of E. Then no current
is needed for magnetization purposes and only current enough to
cover the losses is consumed. As the load torque increases so does
the load angle δ. Neglecting the losses the voltage vector connecting
Us and E is jXs is and the phase angle ϕ must necessarily be
δ
ϕ= (2.9)
2
Including the losses will decrease the angle between Us and Is and
increase the power factor, which may be obvious as the active power
is increased. The displacement power factor can be deduced from
[Figure 2.9]. As is evident the DPF drops for small load angles

cos ϕ = const.
lagging

leading

1.0 Us E

q
d

Figure 2.9: Phasor diagram, constant displacement power factor

contours.

unless the supply voltage is exactly matched to the EMF, which in

practice never is the case. At low loads the DPF is thus highly de-
pendent on the supply voltage as will be seen in chapter 4. As the
motor in a pump application will likely be operated at full load or a
large fraction thereof, the DPF at no load is not so interesting. The
DPF in the expected load range is more important. As the saliency
is Xq > Xd the load angle variation can be quite large. When the
load angle is increased the supply voltage vector follows a circular
path with radius Us around the origin. A unity power factor can
16 2. Models and Design

be obtained if the supply voltage magnitude is adjusted so that it

follows a circular path of radius E/2 centered around jE/2. The
supply voltage magnitude for a LSPMSM is general constant so a
high displacement power factor can not be obtained for all load an-
gles. For increasingly larger load angles the power factor is high
for a decreasingly smaller load angle interval, even if Us is opti-
mally adjusted for that load angle range. The discussion of Us vs E
is of course symmetric and during the design the E is adjusted to
a certain supply voltage. From this reasoning it would seem ben-
eficial to adjust the Us or E so that the supply voltage is slightly
lower than the magnet EMF. Then the displacement power factor
would go from leading to lagging with increasing load torque, peak-
ing with a unity power factor at some torque, ideally the nominal
load torque. There are however reasons for not dimensioning the
magnets like this. The magnet braking torque is proportional to
the square of the magnet EMF and the magnet weight, and cost,
is roughly proportional to the EMF. Thus a LSPMSM is generally
slightly under-magnetized due to cost and starting concerns.
A saliency Xd > Xq would reduce the load angle range and also
reduce the supply voltage dependency of the DPF.

Steel
For high efficiency a high resistivity (low loss) sheet steel is favor-
able to use in the stator. To allow this and keep losses low a rela-
tively "high" stator yoke should be used. The rotor laminations can
be made in a low resistivity steel as the rotor experiences little flux
variations under normal operation.

Air-gap
The supply can generally be considered a voltage source and so also
the magnet EMF. Any mismatch between the two is compensated
by an increased magnetizing current, the magnitude of which de-
pends on the magnetizing inductance. A small air-gap is important
to get large magnetizing inductances and thus keep the displace-
ment power factor and efficiency high over a wide voltage range.
For narrow voltage ranges a comparably large air-gap can be used
without reducing the efficiency as much as in induction motors.
This is most beneficial for small motor sizes were the air-gap must
be large enough to accommodate the factors discussed in Section
1.1. Thus the LSPMSM has a significant efficiency advantage over
induction motors for fractional horsepower motors.
For a small air-gap the harmonic content of the air-gap flux den-
sity is however not smoothed as much as for a larger air-gap so an
Steady state models 17

increase in iron loss can be expected. This should however be off-

set by lower copper losses due to the reduced need for magnetizing
current.

Magnets

The choice of magnets today is mainly between neodymium and

ferrite magnets. Samarium-cobalt magnets are probably consid-
ered too expensive for LSPMSM although the high temperature
range and chemical resistance are important benefits. As the mag-
nets, out of necessity, must be placed in the d-axis main flux path
and the magnet relative permeability is close to one the choice of
magnet greatly affects the d-axis magnetizing inductance. High en-
ergy density magnets, i.e. neodymium, allow a larger d-axis induc-
tance than what is possible with ferrite magnets, resulting in a
high power factor and high efficiency over a larger supply voltage
range.
Flux concentration arrangements allow air-gap flux densities
beyond the remanent flux density of the magnet material and also
allow better magnet utilization as the magnet operating point can
be closer to the energy maximum. Increasing the air-gap flux den-
sity is important when using ferrite magnets but not so when using
neodymium magnets as a fairly low air-gap flux density is used to
achieve high efficiency.
Very thin magnets, besides from being brittle causing manufac-
turing problems, also are more sensitive to demagnetization as the
magnet reluctance is low and stray fields could more easily cause
demagnetization.
The magnets are free to be placed anywhere along the d-axis
main flux path. However, there are benefits and drawbacks to cer-
tain magnet placements. Overall interior magnet placements have
good protection from the cage, while also thermally not too close.
There are designs with magnets mounted in the top of the rotor
slots [22] and surface/inset mounted magnets with a conductive
sleeve acting as rotor winding. These designs are more sensitive
to demagnetization and thermal concerns. The magnet size is cho-
sen to achieve the desired flux density in the air-gap. As the EMF
is chosen to be only a fraction of the supply voltage due to mag-
net braking concerns, the flux when operating will be significantly
larger than the flux from the magnets alone. Thus the air-gap flux
density due to the magnets should be chosen comparably low to al-
low an increased air-gap flux density without to much iron losses
in the stator.
18 2. Models and Design

Slots and Windings

Slot number combinations follow the same concerns as in induc-
tion motors and will not be discussed further here, the reader is
referred to induction machine literature such as "Harmonic Fields
Effects in Induction Machines" by Heller [6]. One thing to keep in
mind is that it is generally favorable to have the number of rotor
slots divisible by the number of poles, so that the poles are sym-
metric. Furthermore, it is good to make sure that the magnet brak-
ing torque and slot harmonic torque components do not contribute
maximally at one particular speed.
The air-gap flux density contains a large third harmonic due
to the rectangular magnet flux density shape. To avoid introduc-
ing this third harmonic into the line currents the stator winding is
almost always star-connected.

Rotor slots and winding

The rotor resistance should be kept low enough to facilitate syn-
chronization, i.e. cage torque should bring the rotor up to an as
high as possible speed before it is balanced by the load torque. A
torque peak at half the synchronous speed is good to counteract
the Görges phenomenon due to (d-q) asymmetry which can be ex-
pected to be substantial. Deep rotor bars, i.e. rotor bars that have
slot inductances so large that the resistance varies significantly
over the frequency range during start, are beneficial, but as there
are no efficiency concerns as in induction machines there is only a
trade-off between different speed ranges. Most notably it is a bal-
ance between early start (zero speed) torque and synchronization.
As the rotor is accelerated the cage torque will decline and
at some speed close to synchronous be balanced by the load and
magnet braking torque. For speeds close to synchronous both the
load and magnet braking torque are fairly independent of the rotor
speed while the cage torque declines approximately linearly with
increasing speed. The slope of linear decrease of the cage torque
depends inversely on the rotor resistance so the speed interval over
which the actual synchronization process must span is proportional
to the rotor resistance. However, during the synchronization cycle
the magnet flux penetration into the stator is slowed by a low re-
sistance rotor cage, affecting the synchronization ability negatively.
This effect should however not be very pronounced as the slip fre-
quency is very low so there is plenty of time for the magnet flux to
build up.
The quasi-static magnet braking torque as described by Hon-
singer [8] is only valid for very slow starts. The magnet braking
Start-up models 19

torque is highly dependent on the starting time. For fast starts the
rotor cage will maintain the magnet flux magnitude for a large part
of the starting sequence. The reaction currents in the stator wind-
ing will then not depend on the magnetizing reactances but rather
on the leakage reactances, which means that the braking torque
can be substantially larger. This additional braking will slow the
start-up until balanced by the cage torque.
The rotor winding can be made more resistive compared to IMs
as there are no efficiency concerns. Increasing the rotor resistance
moves the torque peak to lower speeds. A very resistive cage how-
ever influences the synchronization capability negatively and it
is often the synchronization capability that is the limiting factor
when considering starting capability.

Rotor slot shape

As the motor is voltage fed the starting current magnitude is de-
termined by the stator leakage inductance and to some degree also
the stator winding resistance. To limit the starting current the sta-
tor leakage inductance must be fairly large, i.e. the magnetomotive
force required to drive a certain flux must be kept low. A small air-
gap combined with closed rotor slots provides low reluctance flux
paths that limit the starting current magnitude without affecting
the synchronous performance too much.

2.3 Start-up models

The start-up is a fairly complex phenomenon, but a set of ma-
chine equations in a d-q reference frame as described by Binns
in [19] models the start-up fairly accurately. Large current magni-
tudes during starting cause saturation, nonlinearity and harmon-
ics, none of which are included in the model. Using comparably
coarse simplifications the LSPMSM during starting can be viewed
as a superposition of an induction motor and a short-circuited syn-
chronous motor. Combining this simplification with the assumption
that any change in speed is slow compared to the electromagnetic
response of the motor yields a quasi-static model.

Quasi static model

To have the quasi static model to produce usable results the change
in angular velocity ωr must be much slower than the time response
of the magnetic circuits in the model.
The average torque acting on the rotor is divided into cage
torque, magnet braking torque and load torque. Different equiv-
20 2. Models and Design

LSPMSM ≈ IM SM

∼ ∼

Figure 2.10: Quasi static start-up model approximation. During

starting the line start permanent magnet synchronous motor can
be modeled as an induction motor coupled to a short-circuited syn-
chronous motor.

alent circuits are used to calculate the cage and magnet braking
torques. In reality the flux producing the cage torque and the flux
from the magnets share the same flux paths and for the separation
and superposition of these torque components to be valid a linear
magnetic media must be assumed.

Cage torque
To calculate the cage torque a traditional steady state induction
motor equivalent circuit as described in for example [20] can be
used. The torque produced by the cage is the power dissipated in
s of the equivalent circuit in [Figure
the fictitious resistance Rr0 1−s
2.11] divided by the angular speed of the rotor. Some of the flux

′
Xlr
Rs Xs
+

R′r

Us Xm

1−s


R′r s

Figure 2.11: Quasi static induction motor equivalent circuit.

passing the air-gap will not link with the cage and is represented
by the leakage reactance Xlr 0
. The iron bridges covering the cage
bars saturate at relatively low currents and the flux lost is best
modeled as a voltage drop in the equivalent circuit. An extended
model taking into account the effect of iron bridges as described by
Nee in [16] is used.
Start-up models 21

′
Xlr
Rs Xs
+

R′r
s

Us Xm
+ ′
−
Urb

Figure 2.12: Quasi static induction motor equivalent circuit with

iron bridges represented by Urb

Magnet braking torque

The currents induced in the stator winding due to the permanent
magnets cause resistive losses acting as a braking torque which
is represented by the short-circuited synchronous motor in [Figure
2.10]. An expression for the quasi-static magnet braking torque can
be derived from the equivalent circuit in [Figure 2.2] as shown by
Herslöf [7]. Not taking into account the line impedance yields the
following expression:
! !
mp 2
Rs2 + Xsq (1 − s)2 Rs (1 − s)
Tb = − E02 .
2ωs Rs2 + Xsd Xsq (1 − s)2 Rs2 + Xsd Xsq (1 − s)2
(2.10)

Machine Equations for the transient model

For evaluating the transient behavior of the motor the rotor fixed
machine equations as described by Honsinger in [8] are used. These
are rewritten here with slightly different nomenclature:


 Ψ̇sd = usd − Rs isd + ωr Ψsq







 Ψ̇sq = usq − Rs isq − ωr Ψsd



Ψ̇rd = −Rr0 ird (2.11)







 Ψ̇rq = −Rr0 irq





T = mp 2 (Ψsd isq − Ψsq isd )
Chapter 3

Equivalent Circuit
Parameter Measurement

There are numerous alternative ways of measuring the parame-

ters used in the machine models. Some of the parameters are quite
easily measured with sufficient accuracy, such as the winding resis-
tance, while others are harder to obtain. Ideally the method used
should not affect measured values but in reality non-ideal prop-
erties not included in the models can affect the obtained values
appreciably. The different measurement methods can be divided
into measurements done with the rotor locked in position, here-
after called static measurements, and measurements where the
rotor is rotating at nominal synchronous speed, hereafter called
synchronous measurements. Measuring equivalent parameters of
a LSPMSM differs from measurements on synchronous machines
with a field winding as the magnetization can not be changed, i.e.
the normal no-load and short-circuit tests can not be used. The
measurement possibilities are also somewhat different from other
permanent magnet machines due to the cage winding having much
lower impedance than the damper windings or pole shoes of other
permanent magnet machines.
The measurements described in this chapter were made on a
fairly small 4-pole LSPMSM with nominal output of 750 W, the
cross section of which is shown in [Figure 2.1]. Some further data
on the test motor are given in Appendix A. The obtained values and
plots are presented in ordinary SI units, i.e. not in per-unit values,
as measurements were only made on one motor and the compara-
tive benefits of the per-unit system were not applicable.
24 3. Equivalent Circuit Parameter Measurement

3.1 Static measurements

For the static measurements the shaft of the machine was con-
nected to a rotary table to facilitate locking the rotor in an arbi-
trary position. The static measurements benefit from some advan-
tages compared to synchronous measurements. Most important is
that the parameter measurements are not dependent on the mag-
net EMF or load angle measurements. Coupled to this advantage is
of course the drawback that the magnet EMF can not be measured
in static tests. A more serious disadvantage is that the phase, pole
and slot saliency is not averaged as in synchronous measurements.
Yet another disadvantage is that the flux density distribution in
the stator part is not exactly the same as under synchronous mea-
surements which could cause some discrepancy.
The static measurements can be divided into alternating and
direct current excited measurements. As the LSPMSM has a rotor
winding, in general a cage winding, the flux produced by an alter-
nating stator current will not penetrate far into the rotor and thus
measuring the magnetizing reactances is not possible using alter-
nating current measurement methods.

DC-excited static measurements

Stator resistance
The stator resistance can easily be measured by feeding a direct
current through the windings and measuring the voltage at the
terminals. Although the stator resistance is independent of the cur-
rent it is advisable to use a direct current of magnitude comparable
to the nominal phase current to achieve good measurement accu-
racy. At the same time the self-heating of the winding must be kept
in mind for this measurement. The resistance measured phase to
phase at room temperature (20◦ C) and approximately 1 A DC was
4.6 Ω, i.e. the stator resistance is

Rs = 2.3 Ω (3.1)

per phase.

Synchronous reactances
The inductance of a coil or winding, is generally defined as a mea-
sure relating a flux linkage to a current

Ψ
L= (3.2)
i
Static measurements 25

but sometimes a differential definition is used

dΨ
L= (3.3)
di
which is equivalent if the inductance is constant, i.e. there is neg-
ligible saturation.

dΨ
L= di
saturation
Ψ

Ψ
L= i

Figure 3.1: Flux linkage vs. current and two alternative induc-
tance definitions.

The voltage over the inductance is equal to the rate of change of

flux linkage
dΨ
u= . (3.4)
dt
One way to measure inductances is to measure this voltage for a
known current excitation. A common choice is a sinusoidal current
excitation. Then the measured voltage will also be a sinusoid that
is out of phase with the current. However, the inductance does not
necessarily need to be measured using an alternating current. The
inductance can also be calculated if the flux linkage corresponding
to a certain DC current flowing through the winding can be mea-
sured. Measurement methods utilizing this fact are described in
[12], [13] and [10].
An important aspect of DC excited measurements is that the
synchronous inductance is measured at equivalently 0 Hz. During
steady state operation the flux linking with the the rotor, i.e. the
flux represented by the magnetizing inductance, has only a DC-
component. Even during transients the magnetizing flux has very
low frequency components so the flux distribution at 0 Hz is compa-
rable to the flux distribution during normal operation. Thus mea-
suring the magnetizing inductance by a DC-flux is well motivated.
The effect of the stator leakage flux on the other hand, is always
at line frequency during normal operation but in the DC-excited
26 3. Equivalent Circuit Parameter Measurement

measurement the inductance relating to a DC-flux is measured

which could be thought of as a disadvantage of this measurement
method.
In the following analysis a common T-model equivalent circuit
shown in [Figure 3.2] is used for each phase.

Rs Lls L′lr
+

uphase Lm1 R′r

Figure 3.2: Equivalent circuit of one phase.

The stator resistance Rs is the resistance of the stator winding.

The magnetizing inductance Lm1 represents the main flux, gener-
ally the fundamental flux crossing the air-gap, that links with the
winding. As the main flux also links with the other phase windings
the magnetizing inductance, Lm , used in a common two-axis d-q
model is larger:
n
Lm = Lm1 (3.5)
2
where n is the number of phases. The stator leakage Lls represents
the flux linking the stator winding that is not part of the main flux,
i.e. slot leakage, end winding inductance etc. The rotor leakage L0lr
similarly represents the rotor winding leakage but referred to the
stator side. Also the rotor winding resistance Rr0 is transferred to
the stator side.
For simplicity it would be beneficial to excite only one phase of
the machine but as the neutral node in a star connected machine is
generally not accessible from the outside, or if the machine is delta-
wound, this is not possible. Thus at least two of the phases must be
excited. There are an infinite number of possible combinations of
current magnitudes in the phases, however, two of special interest
are:

- peak current magnitude in one phase, hereafter called three-

phase-excitation.

- zero current in one phase, hereafter called two-phase-

excitation.

These two alternatives are snapshots of the sinusoidal three-phase

currents shown in [Figure 3.3]
Static measurements 27

Three-phase excitation
i
ir is it

Two-phase excitation

Figure 3.3: DC-excitation alternatives.

Three-phase-excited measurements
Considering the first alternative, shown in [Figure 3.4], a current I0
is driven through phase R, and due to the equal stator resistances
the current will divide evenly between the phases S and T. Given

Rext
it
+

R
ut

I0

Figure 3.4: DC excitation along the R-phase direction in a three-

phase-configuration.

that this symmetry is maintained throughout the measurement,

which is true whenever the magnetic circuit is symmetric around
the excitation axis, i.e. excitation along either d- or q-axis, a single
28 3. Equivalent Circuit Parameter Measurement

phase equivalent circuit as shown in [Figure 3.5] can be used for

analysis.

Rext Rs Lls L′lr

it
+

I0 ut Lm R′r

Figure 3.5: Single phase equivalent circuit.

Considering that the DC winding currents are a snapshot of the

sinusoidal three-phase currents in [Figure 3.3] at the instant when
iR (t0 ) = îphase (3.6)
the DC excitation current should be chosen to be equal to the peak
magnitude of the phase currents under normal operation:
I0 = îphase . (3.7)
The current I0 will cause a resistive voltage drop I0 Rs in the
R-phase and a resistive voltage drop of I0 Rs/2 in the S- and T-
phases, thus an equivalent stator resistance of 1.5 times the phase
stator resistance is appropriate to use in the equivalent circuit in
[Figure 3.5]. In a similar way, all the other circuit elements, i.e. the
rotor resistance Rr0 , the magnetizing inductances Lm and leakage
inductances Lls and Llr are also scaled by 1.5.
The magnetomotive forces produced by the winding currents
will be along the magnetic axes of the individual phases. The mag-
netomotive forces produced by the excitation currents
 
1 1
IR,S,T = 1, − , − I0 (3.8)
2 2
are shown in [Figure 3.7] and as can be seen the resultant magne-
tomotive force will lie on the axis of the R phase and will have a
magnitude of
 
1 1 1 1 3
F = FR + FS + FT = 1 + · + · · FR = FR . (3.9)
2 2 2 2 2

This resultant magnetomotive force drives a flux through the

machine along the R-phase magnetic axis. The measured magne-
tizing inductance, which due to the rotor saliency is angle depen-
dent, will depend on the reluctance of the flux paths in this direc-
tion. As the rotor can be locked in any position this can be chosen
to correspond to either the d- or q-axis.
Static measurements 29

R−

T+ S+
× ×
eS
FT
eR
eT 60◦ FR Ftot
S− T−
R+ FS
×

Figure 3.6: Magnetic axes of Figure 3.7: Mmagnetomotive

the three phase windings. forces of the three phases with
three-phase-excitation.

When the switch in [Figure 3.5] is closed the current through

the circuit will start to decay. [Figure 3.8] shows a typical step re-
sponse, i.e. the terminal current it and terminal voltage ut , from
measurements on the test motor1 with the excitation axis along
the d- and q-axis and an excitation current of approximately 1.0 A.
If the stator resistance is known in advance, the measured termi-
nal voltage and current can be used to calculate the voltage over
the stator leakage and the magnetizing inductance as

u = ut − Rs it . (3.10)

This voltage corresponds to the rate of change of flux linking the

stator winding, i.e. Lls and Lm :
dΨ
u= (3.11)
dt
Integrating this voltage over a certain time span will yield the dif-
ference in flux linkage
Z t2
u dt = ∆Ψ . (3.12)
t1

Integrating from the instant when the switch is closed (i = I0 ) to

infinity (i = 0) will yield the flux linkage due to the direct current
I0 . Using the inductance definition in (3.2)
Z
1 ∞
Ls = u dt . (3.13)
I0 0
1 Data on the test motor in Appendix A
30 3. Equivalent Circuit Parameter Measurement

Terminal voltage & current [V,A] it d-axis

0

-1 ut
-2

-3

-4

-5

-6

-7
1
it
Terminal voltage & current [V,A]

q-axis
0

-1 ut
-2

-3

-4

-5

-6

-7
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
Time [s]

Figure 3.8: Typical step response along d- and q-axis with three-
phase-excitation and I0 = 1.0 A.

The time-voltage area shown shaded in [Figure 3.9] is propor-

tional to both Ls and I0 . Although the rotor branch (L0lr and Rr0 )
will affect the shape of the step response it will not change the in-
tegrated voltage. The wavy appearance of the q-axis measurements
is due to mechanical oscillations interacting with magnetic system.
This mechanical oscillation has a high Q-factor and a comparably
low amplitude which means that little energy is lost through this
interaction and thus it should not affect the measurement appre-
ciably.
Measurements with excitation currents ranging from 0.5 A to
2.5 A were made and the reactance values obtained by integration
are show in [Figure 3.10]. For lower excitation currents the mea-
surement accuracy suffers and to indicate this some minimum and
maximum plausible reactance values were calculated. To do this
Static measurements 31

Terminal voltage & current [V,A] it d-axis

0
ut
-1 ut - Rs it
-2

-3

-4

-5

-6

-7
1
it
Terminal voltage & current [V,A]

q-axis
0
ut
-1 ut - Rs it
-2

-3

-4

-5

-6

-7
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
Time [s]

Figure 3.9: The shaded area is proportional to Lm + Lls and I0 .

Table 3.1: Measurement error estimates.

Current gain ±1 %
Current offset ±1 % full-scale∗
Voltage gain ±1 %
Voltage offset ±1 % full-scale∗
Emf ±1 %
Load angle ±0.5◦ mech.
∗
No offset error for AC measurements

limit intervals on the raw measurement data were estimated by

applying the error factors and terms in [Table 3.1].
These limits were then used to calculate limits on the reactance
values. Thus the reactances are shown as bands in [Figure 3.10]
and as expected the bands are wider for low excitation currents.
These bands should not be taken as actual errors or confidence in-
32 3. Equivalent Circuit Parameter Measurement

tervals but rather as an indication of typical error magnitudes.

30

Xsq
Impedance [Ω]

25

20

15 Xsd

10
0.5 1 1.5 2 2.5
Excitation, equivalent peak phase current [A]

Figure 3.10: Static measurement of synchronous reactances, three

phase excitation.

Two-phase-excited measurements
In addition to the three-phase-excited measurements, tests with
two-phase-excitation were made, mainly to provide an additional
set of data as the results should be very similar. However, the flux
density distribution in the stator is not the same as under three-
phase-excitation so this could yield a noticeable difference. Also, for
the slot combination and winding of the test motor the direct axis
rotor center tooth is aligned with a stator tooth when using two-
phase excitation and aligned with a stator slot when using three-
phase excitation. The direct current I0 flows only in phases R and S
as shown in [Figure 3.11] and represents the instantaneous value
of the three phase currents with peak magnitude
√
3
I0 = îphase . (3.14)
2
As with three-phase excitation a one phase equivalent circuit is
used but now all the equivalent parameters are scaled by a factor 2.
The magnetomotive forces as shown in [Figure 3.12] will produce a
Static measurements 33

resultant magnetomotive force which will be offset from the R-axis

by 30◦ and have the magnitude
√ √ !
3 3 √
F = FR + FS + FT = + + 0 · FR = 3FR . (3.15)
2 2

Rext
it
+

R
I0 ut

Figure 3.11: DC excitation in a two-phase-configuration.

FR
30◦

Ftot
FS

Figure 3.12: Magnetomotive forces under two-phase-excitation.

The same error estimates as used in the three-phase excited

measurements was used and the calculated reactance values are
shown in [Figure 3.13]. The correspondence between the two- and
three-phase excited measurements is good. The slight bump in the
q-axis reactance could be due to tooth to tooth flux paths going out
of saturation. Another possibility is that the torque produced by the
34 3. Equivalent Circuit Parameter Measurement

q-axis current aligns the rotor slightly to minimize the reluctance,

thereby increasing the inductance.

30

Xsq
Impedance [Ω]

25

20

15 Xsd

10
0.5 1 1.5 2 2.5
Excitation, equivalent peak phase current [A]

Figure 3.13: Static measurement of synchronous reactances, two

phase excitation.

Step response adapted parameters

Although the synchronous inductance along the d- and q-axes is
useful information, the step response contains information about
the other equivalent circuit elements as well. For common param-
eter values two fairly distinct time constants can be observed in
the step response. Initially, the current in the stator leakage in-
ductance decreases quickly while at the same time the current in
the rotor branch increases. This initial step response is determined
by a fast time constant depending on the leakage inductances and
the external stator and rotor resistances. A longer time constant in-
cluding the magnetizing inductance dictates the response following
the initial transient.

Optimized adaption of parameter values

Finding out exactly how each parameter affects the step response is
cumbersome, and more importantly it is hard to get any notion on
how measurement errors affect the calculated parameter values. A
Static measurements 35

different approach was used in which a simulated step response is

fitted to the measured data. The step response for a given set of
parameters can be simulated in S PICE, M ATLABTM etc.

Lls L′lr

+ +
+ + +
Rs + Rext i1 i2 R′r
Lm

Figure 3.14: Simplified circuit after the switch is closed.

Again, the one phase equivalent circuit [Figure 3.5] is used for
analysis. However, after the switch in [Figure 3.5] is closed the cir-
cuit can be reduced to [Figure 3.14].
Kirchoff ’s voltage law in the two loops gives two equations:

 di1
 (Rext + Rs )i1 + Lls dt + Lm ( di 1
dt −
di2
dt ) = 0
(3.16)
 di2
Rr0 i2 + Llr dt − Lm ( di 1
dt −
di2
dt ) = 0

or in matrix notation
     di1

Rext + Rs 0 i1 Lls + Lm −Lm dt
+ =0
0 Rr0 i2 −Lm Llr + Lm di2
dt
(3.17)
or
di
Ri + L =0 (3.18)
dt
with
     
Rext + Rs 0 i1 −Lm Lls + Lm
R= ,i= ,L= .
0 Rr0 i2 Llr + Lm −Lm
(3.19)
From an initial guess of the parameter values, the step response
can be calculated by solving this system of differential equations
yielding an expression for i1 of the form

i1 = C1 e−t/τ1 + C2 e−t/τ2 , (3.20)

were τ1 and τ2 are the time constants discussed above. The dif-
ferential equation solver ode45 included in M ATLABTM was used
and the result compared to the measured step response. The mea-
sured step response and the simulated step response for the initial
36 3. Equivalent Circuit Parameter Measurement

1
it
Terminal voltage and current [V, A]

-1

-2 ut

-3 Measurement

Initial simulation
-4
Optimized simulation

-5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
Time [s]

Figure 3.15: Adaption of simulated step response to measured

data. Excitation along the q-axis.

guess of parameter values are shown as solid and dash-dotted lines

in [Figure 3.15]. The error between the simulated and measured
step response was calculated and the parameter set optimized to
minimize the quadratic error sum over all samples. The step re-
sponse for the final fit of parameter values is shown dashed in
[Figure 3.15]. As there are 5 circuit elements but only 4 unknowns
(C1 ,τ1 ,C2 ,τ2 ) in the step response one of the circuit elements, or
more specifically, one of the inductances is superfluous as long as
only the terminal quantities can be observed. To address this the
equivalent circuit could be reduced to a Γ- or inverse Γ-model but
instead the T-model was retained and the ratio between the stator
and rotor leakage locked to an arbitrary value of 1:1.
The optimization could be done using any multidimensional nu-
merical optimization method, in this case the Nelder-Mead method
Static measurements 37

fminsearch included in M ATLABTM , was used. To keep optimiza-

tion time short without affecting accuracy too much the number of
samples was averaged down 16 times to 625 samples.
The impact of measurement errors on the different parameter
values is not obvious.
One way to get a notion of how good the accuracy of the param-
eter values are would be to calculate the possible minimum and
maximum amplitude measured step responses given certain error
estimates such as the ones in [Table 3.1] and then fit simulation
curves to this data. The difference in each parameter between the
different adaptations could then be interpreted as an indication of
the sensitivity of each parameter. This would however not guaran-
tee that the limits of each parameter value is found, in fact some
parameters might not change at all between the different curve fits.
Another approach would be to find how much each parameter
can be varied while keeping the step response within the step re-
sponse limits. This is however not possible if there is significant
measurement noise or other effects causing the step response to
have non-ideal shape such as the response in [Figure 3.15] as there
is no simulated response that would fit within the measured step
response limits.
Probably the simplest way to visualize the sensitivity of each
parameter would be to note how good the simulation fit was by
calculating an RMS error. Then each parameter is increased or de-
creased until this error increased by a certain factor. This is the
method used and the arbitrary RMS factor is 2. These minimum
and maximum values are shown as bands in [Figure 3.16] and [Fig-
ure 3.17].
The saliency can be seen in all the parameters but is of course
largest in the magnetizing inductance. The resistance values were
expected to show negligible saliency but as seen this was not the
case. The difference in the stator resistances is relatively small and
can be explained by and is probably due to thermal variations in
primarily the external resistance. The rotor resistance saliency is
probably coupled to the the large difference in the magnetizing in-
ductances and need not be as large as shown although a difference
can be expected due to the different slot leakages causing different
levels of skin effect.
The larger q-axis leakage is due to the rotor slot bridges being
unsaturated. The bridges for d-flux are already saturated by flux
from the permanent magnets. As the current increases also the q
bridges saturate and the leakage inductance falls towards the d-
axis value.
Both the stator and rotor resistances show a negative depen-
dence on the excitation current. Regarding the stator resistance
38 3. Equivalent Circuit Parameter Measurement

Rsq
3
Rs [Ω]

2
Rsd
1

30
Xm [Ω]

20 Xmq
Xmd
10

6 Xlsq, Xlrq
Xls = Xlr [Ω]

2 Xlsd, Xlrd

Rrq
Rr [Ω]

1
Rrd
0
0.5 1 1.5 2 2.5
Equivalent peak phase current [A]

Figure 3.16: Step response optimized parameter values, d-axis pa-

rameters in blue and q-axis in green, three-phase excitation.

it is probably due to the external resistance heating up and thus

increasing in resistance. As a constant external resistance is as-
sumed this increase is taken from the stator resistance. This also
affects the rotor resistance but to a much lesser degree as the ro-
Static measurements 39

Rsq

3
Rs [Ω]

2
Rsd
1

30
Xm [Ω]

20 Xmq
Xmd
10

6 Xlsq, Xlrq
Xls = Xlr [Ω]

2 Xlsd, Xlrd

Rrd
Rr [Ω]

1
Rrq
0
0.5 1 1.5 2 2.5
Equivalent peak phase current [A]

Figure 3.17: Step response optimized parameter values, two-

phase excitation.

tor resistance is adjusted to fit mainly the slower step response.

Another effect causing negative current dependence is the rotor
slot bridges saturating. With unsaturated bridges the rotor leak-
age inductance, especially the q-axis leakage inductance, is high
40 3. Equivalent Circuit Parameter Measurement

and could cause considerable skin effect thus increasing the rotor
resistance as long as the bridges are not saturated. For higher cur-
rents the bridges saturate and the rotor resistance drops together
with the leakage inductance.

Rsq

3
Rs [Ω]

2
Rsd
1

30
Xm [Ω]

20 Xmq
Xmd
10

6 Xlsq, Xlrq
Xls = Xlr [Ω]

2 Xlsd, Xlrd

Rrd
Rr [Ω]

1
Rrq
0
0.5 1 1.5 2 2.5
Equivalent peak phase current [A]

Figure 3.18: Step response optimized parameter values, two- and

three-phase excitation measurements overlayed.
Static measurements 41

AC-excited static measurements

Leakage inductances
Measurements of the leakage inductances can be made by mechan-
ically locking the rotor and exciting two or three of the stator wind-
ings with a sinusoidal voltage, i.e. applying a pulsating voltage in-
stead of a rotating. If the impedance of the cage branch can be as-
sumed to be much lower than the magnetizing inductance, which
is often the case at nominal frequency, a simplified equivalent cir-
cuit [Figure 3.19] without the magnetizing inductance can be used.
From the fundamental current phase and magnitude the imped-

Rs Lls L′lr
it
+

ut R′r

Figure 3.19: Equivalent circuit without magnetizing inductance.

ance observed at the motor terminals can be calculated.

15
10
5
ut
[V, A]

0
10 ⋅ it
-5
-10
-15
-20
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018
Time [s]

Figure 3.20: Static d-axis leakage inductance measurement, ter-

minal voltage and current with 2-phase excitation.

Measurements were done with the rotor locked in the positions

resulting in the smallest and largest impedances, corresponding
to the d- and q-axis. The impedances measured at 50 Hz and peak
phase current of approximately 1.3 A were
Zd = 4.51 + 4.34i Ω (3.21)
42 3. Equivalent Circuit Parameter Measurement

15
10
5
ut
[V, A]

0
10 ⋅ it
-5
-10
-15
-20
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018
Time [s]

Figure 3.21: Static q-axis leakage inductance measurement, ter-

minal voltage and current with 2-phase excitation.

Zq = 4.90 + 6.36i Ω (3.22)

corresponding to total leakage inductances of

Llsd + Llrd = 14 mH (3.23)

Llsq + Llrq = 20 mH (3.24)

Rotor cage resistance

The cage resistance can be calculated by subtracting the stator re-
sistance from the resistive part of the impedance in the preceding
section.
Rrd = Re{Zd } − Rs = 2.2 Ω (3.25)
Rrq = Re{Zq } − Rs = 2.6 Ω (3.26)

3.2 Synchronous measurements

Synchronous measurements have the advantage of being made
under conditions similar to the operating conditions of interest.
Among the disadvantages is the need for torque angle measure-
ment and the dependence on the back-EMF E0 .

Open circuit voltage

Rotating the rotor at synchronous speed with open terminals, and
thus no current, from the phasor diagram in [Figure 2.5] the ter-
minal voltage is equal to the back-EMF E0 . The line-to-line voltage
Synchronous measurements 43

measured with a True RMS multimeter was 134.4 V and thus

√
2 · 134.4
E0 = √ = 110 V (3.27)
3

150
Induced voltage [V]

100
50
0
-50
-100
-150
-200
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018
Time [s]

Figure 3.22: Measurement of the back-EMF, open circuit terminal

voltage line-to-line.

Synchronous reactances
If the back-EMF E0 is known, both in magnitude and phase relative
to the terminal voltage, the d- and q-axis synchronous reactances

Xsd = Xmd + Xls (3.28)

Xsq = Xmq + Xls (3.29)

can be calculated given that the stator resistance is known and that
the terminal current is measured.

Without angle measurement

To know the phase of the back-EMF requires that the mechanical
angle of the rotor is measured with fairly high accuracy. However,
the direct axis reactance can be measured without an angle mea-
surement in a no-load test if the losses are assumed to be small. In
that case angle between the terminal voltage and the magnet EMF
can be neglected. The d-axis synchronous reactance Xsd can then
be found as
Ut − E0
Xsd = (3.30)
It
44 3. Equivalent Circuit Parameter Measurement

where Ut and It are the terminal (phase) voltage and current. Two
line-to-line voltages and two phase currents were measured using
an oscilloscope. Each quantity was averaged over 5 periods and
the fundamental component extracted and then the synchronous
impedance of each phase was calculated. There was negligible dif-
ference between the different phases and the average is shown in
[Figure 3.23]. It can be seen that the resistive part of the imped-

Xd (FC)

15 Xd
Impedance [Ω]

10

5 Rd

Rd (FC)
0
1 1.5 2 2.5 3 3.5 4 4.5 5
Current [A]

Figure 3.23: Direct axis impedance from synchronous measure-

ments (motoring) without angle measurement. Without and with
friction compensation (FC).

ance is much larger than the stator resistance Rs as it includes the

iron and friction losses. This does however not influence the value
of Xsd significantly as the load angle remains small. Measurements
were also done with a synchronous motor connected to it compen-
sating for iron and frictional losses, marked (FC) in [Figure 3.23].
Considering the large dependence on the phase current, the dif-
ference between the two measurements is not so important. The
initial positive current dependency of the inductance could be due
to the iron bridges around the magnet going out of saturation. This
slight incline can also be seen in the static DC-excited measure-
ments but not in the synchronous measurements with the machine
running as a generator as the negative current will not bring the
bridges out of saturation.
Synchronous measurements 45

With angle measurement

Adding angle measurement makes it possible to calculate both
the direct and quadrature axis synchronous inductances. To mea-
sure either the direct or quadrature inductance, ideally for simple
calculations, only d- or q-current should excite the machine. Also
one would like to measure the inductances under different cross-
current biases.
However, arbitrary excitation, i.e. arbitrary current vector re-
quires vector control of the machine under test which was not avail-
able in the lab setup. As an alternative the machine was run as a
generator with an adjustable resistive load. This will not give sepa-
rate control of the d- and q-axis currents. The ratio of the currents
will depend on the current magnitude. The calculated reactance
values are thus related to, and valid only for, some combination of
d- and q-current.
The calculation of the synchronous inductances from the mea-
sured quantities is assumed to be a monotonic function over the
error intervals. Thus only combinations of the boundary values are
examined. The estimated error intervals are given in [Table 3.1].
The maximum and minimum calculated reactance values for
each measurement are shown in [Figure 3.24].

50

40
Impedance [Ω]

30

20

10

0
0
-1 -1
-2
-2 -3
q current [A] -3 -4 d current [A]

Figure 3.24: Synchronous reactances from measurement on ma-

chine run as generator with resistive load and load angle measure-
ment.
46 3. Equivalent Circuit Parameter Measurement

Assuming that the cross bias current does not affect the reac-
tance values considerably the calculated curves are projected onto
the d- and q-axis as shown in [Figure 3.25] and [Figure 3.26].

50

40
Impedance [Ω]

Xsq
30

20

10 Xsd

0
-5 -4.5 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5
d current [A]

Figure 3.25: Projection of [Figure 3.24] onto the iq = 0 plane.

3.3 Comparison
The synchronous reactances Xsd and Xsq measured using the dif-
ferent methods previously described are plotted against the equiv-
alent d- and q-currents in [Figure 3.27]. The figure shows that
there is no overlap between measurement regions of the different
measurement methods. An overlap would be beneficial, and should
be sought in future measurements, as it allows for a better compar-
ison between the different measurement methods. For clarity pro-
jections of [Figure 3.27] onto the iq = 0 and id = 0 planes are shown
in [Figure 3.28] and [Figure 3.29] respectively. The measured val-
ues do agree fairly well and an overall idea of the synchronous re-
actances can be seen.
Comparison 47

50

40
Xsq
Impedance [Ω]

30

20

10 Xsd

0
-3.5 -3 -2.5 -2 -1.5 -1 -0.5
q current [A]

Figure 3.26: Projection of [Figure 3.24] onto the id = 0 plane.

Static m. 3-phase

50

40
Impedance [Ω]

Static m. 2-phase
30

20 Sync. m. ind. load

Static m. 2/3-phase
10

0
2 Sync. m. res. load
1
0 0
-1 -2
-2
q current [A] -3 -4 d current [A]

Figure 3.27: Synchronous reactances measured with varius meth-

ods.
48 3. Equivalent Circuit Parameter Measurement

50

40
Reactance [Ω]

Xsq
30

20

10 Xsd

0
-5 -4 -3 -2 -1 0 1 2
d-current [A]

Figure 3.28: [Figure 3.27] projected onto the d-axis.

50

40 Xsq
Reactance [Ω]

30

20

10
Xsd

0
-5 -4 -3 -2 -1 0 1 2
q-current [A]

Figure 3.29: [Figure 3.27] projected onto the q-axis.

Conclusion
Measurements presented in this chapter show a promising poten-
tial for experimentally determining the equivalent circuit param-
Comparison 49

eters of a LSPMSM. Many of the parameters are measured with

sufficient accuracy to allow the performance of the motor to be pre-
dicted to a satisfactory level. However, most of the measurements
in this chapter could benefit from more accurate measurement data
and, more importantly, better treatment of measurement errors
and uncertainties. The step response adapted parameter values are
perhaps the measurements showing most room for improvement
but also the most promising. Using better data-capture equipment,
i.e. with better vertical resolution, it should be possible to measure
the equivalent circuit parameters with greater accuracy. The DC-
excitation also allows for accurate measurements where the exci-
tation is a very small current step but while the motor is biased
by a much larger DC-current. The bias current need not be along
the same axis as the excitation current, nor must it be pure d- or
q-current. If automated, the parameter values corresponding to a
large number of d- and q-current combinations could be measured
and the sparse plot in [Figure 3.27] replaced by an exhaustive sur-
face.
Chapter 4

Measured Performance

Typical applications for line start motors are fans and pumps. To
the end user, and thus also to the manufacturer, it is of interest
to know how large fan or pump can be attached, i.e. what kind of
load is acceptable for a certain motor. The performance that can be
expected for such loads is of interest as well. To the end user there
are mainly two characteristics that are important:
• The motor should start with the load applied without blowing
fuses, overheating etc.
• The steady state efficiency and power factor should be high,
or rather as specified.
Evaluating the steady state performance of line start motors dif-
fers little from evaluating the steady state performance of other
synchronous motors, so therefore focus in this chapter is on mea-
suring the starting performance.

4.1 Starting performance

To measure the starting performance of a line start motor there
is a need for a brake bench setup that can provide controlled and
repeatable starting conditions. The repeatability is important to be
able to compare measurements of similar nature. The conditions
that need to be controlled are:
• Line voltage.
• Line impedance.
• Line phase at turn-on.
• Load torque.
52 4. Measured Performance

• Load inertia.

• Starting position.

Although the line-phase at turn-on and the starting position are

highly interrelated they are not interchangeable. According to d-
q theory there should be no difference between different starting
positions if the phase is adjusted accordingly. This is however not
the case in reality, partly due to slot and saliency effects but also,
more importantly, that the initial electromagnetic transient de-
pends heavily on the phase at turn-on.
To evaluate a start certain specific quantities need to be mea-
sured during the start:

• Line current.

• Line voltage.

• Torque.

• Rotational speed.

These conditions and quantities can be divided into requirements

regarding the electrical supply and those belonging to the mechan-
ical load.

Electrical supply
On the electrical side there is need for a variable three-phase volt-
age source that can be turned on at a given instant. The voltage
supply in the test setup was realized by a variable three-phase
autotransformer connected to three-phase 400 V mains. The au-
totransformer has considerable output impedance which also can
vary with the voltage setting. Measuring both the line voltage and
current it is possible to calculate the total line impedance of the
autotransformer and the mains. The line impedance was found not
to depend so much on the voltage setting and thus a single value
was used for all voltage settings. The line impedance taking into
account only the first harmonic of the voltage and current mea-
sured over a few cycles during the start transient was found to be
ZL = 0.2 + 0.9i Ω. A variable line-impedance was deemed unneces-
sary, or at least not crucial to the test methodology and the initial
tests, and thus not used in the setup.
To control the voltage supply turn-on a computer controlled
three-pole relay was used. To synchronize the turn-on to the sup-
ply voltage the zero-crossing of one of the phase voltages was used
to trigger a variable delay after which the relay is energized. The
Starting performance 53

comparably long, and to some degree random, turn-on time makes

a relay perhaps not the ideal choice as switch element. Transistor
switches could provide much better timing precision, but the sim-
plicity and low on-state resistance of a relay makes it a good choice.
As the relay is energized by single phase 50 Hz all phase angles can
not be accommodated by the relay solution without changing the
phase supplying the relay. The variations in turn-on time from test
to test is small enough (< 1 ms) not to affect the starting conditions
too much and as a three-pole relay with common armature is used
all three phases are connected almost simultaneously.

Mechanical load
On the mechanical side there is a need for a shaft torque that repre-
sents the typical behavior of the load types considered for LSPMSM
applications. Here only impeller type pump loads are considered.
Such loads exhibit a torque mainly dependent on rotational speed
[9]. To some degree a torque component that depends on the angu-
lar position is motivated by phenomena such as vanes of the im-
peller passing either intake or output ports etc. In the pumps con-
sidered these angular position torque components are small com-
pared to torque components that depend on the rotational speed
and its derivative. The shaft torque can thus be divided into

dω
Tem = a + bω + cω 2 + J (4.1)
dt
The static friction coefficient, a, is only needed in certain cases and
the viscous friction, b, is comparably low and does not really need
to be included. Thus to model a typical pump torque there is only
a need for a load torque that has quadratic relation to the rota-
tional speed and an inertia component. The inertia is comprised of
the pump impeller and the fluid contained in the impeller cavity.
Equation 4.1 can be thus be reduced to

dω
Tem = cω 2 + J . (4.2)
dt
Some alternatives considered for the mechanical load in the test
setup were:

• A fan or centrifugal pump.

• Shaft mounted weights for inertia and various electromag-

netic devices for load torque such as:

– An eddy current brake.

54 4. Measured Performance

– A DC generator.

– A synchronous generator

• A vector controlled synchronous generator.

A fan or a centrifugal pump has inherently the right torque

characteristics as it is the intended load. To vary the torque and in-
ertia is however cumbersome as although the head can be changed
by throttling this will only affect the torque. To change the inertia
either size or number of impellers need to be changed.
The eddy current brake has a linear torque-speed relationship
and thus the field current would need to be varied to produce the
quadratic torque-speed relation desired. Other types of electromag-
netic brakes have different torque-speed characteristics but none
provide the required quadratic relation.
Both synchronous and DC generators also provide a linear
torque-speed curve for a given load and field excitation. Thus ei-
ther the load or the field excitation would have to change during the
start for the load to have a quadratic speed dependence. Changing
a variable resistive load quickly enough is not feasible as there is a
limit to the speed and acceleration of the sliding contacts. Varying
the field current is slow as the field coil of suitable generators have
inductances in the range of several henries. If a generator with a
low inductance field coil would be used then the field coil could be
excited by an additional DC generator connected to the same shaft,
thus automatically providing the quadratic torque-speed relation-
ship. The extra generator would add a linear torque component
which however small, would be unwanted. A better solution would
be to measure the rotational speed and use a computer controlled
amplifier to set the excitation current.
When considering a computer controlled setup a vector con-
trolled synchronous motor offers incomparable versatility and sev-
eral advantages. The inertial torque component can be produced by
a virtual inertia making the mechanical setup very simple. This is
the alternative used in the test setup.
Further advantages of a vector controlled setup is the possibil-
ity to measure the torque-load angle relationship synchronously
thus including the leakage inductances etc. and averaging tooth
and pole saliency. The load angle sweep time can be set indepen-
dently and must be comparably long to avoid inducing currents in
the cage. The torque for shorter sweep times are of interest for sta-
bility analysis.
Starting performance 55

Test setup
The schematic of the major components of the test setup are shown
in [Figure 4.1]. Both the electrical machines and the torque trans-

computer
control

resolver

3-phase auto- relay motor torque load inverter DC

AC trans- under trans- machine supply
supply former test ducer

Figure 4.1: Components of the starting performance measurement

setup.

ducer are visible in the photograph in [Figure 4.2] A permanent

Figure 4.2: Photograph of the setup. Motor under test to the left,
torque transducer in the middle and load machine to the right.

magnet synchronous motor with surface mounted magnets is used

to load the motor under test. The load machine is driven by an
IGBT inverter controlled by a DSP system from DSpace. The rotor
angle is measured by a resolver premounted in the load machine
56 4. Measured Performance

and the torque is measured by an optoelectronic torque transducer.

Details on the major components are presented in Appendix A.
For a starting capability test the procedure is as follows:

Test Procedure

1. The autotransformer is set to the required voltage.

2. The virtual inertia and load torque components are set.

3. The starting rotor position is set by the load machine.

4. The zero crossing of phase voltage ur is sensed and triggers a

delay before the starting relay is energized.

5. The currents, voltages and angular position are measured un-

til the rotor reaches a synchronous steady state.

To verify that the inertia of the system is as calculated and that

the friction torque is not too large tests were performed to measure
these quantities.

Rotational inertia
The rotational inertia, J, of the test system is approximately known
beforehand but can be measured to greater accuracy by accelera-
tion tests. The acceleration ω̇ when a known torque, T , is applied
is
T
J= . (4.3)
ω̇
The torque is supplied by the load machine as the characteristics
of the motor under test of course are not known. If there are other
torque components acting on the rotor system, which is generally
true, the measured value of the moment of inertia will be incorrect.
There will be for example iron and windage losses and mechanical
friction braking the rotor. The static friction can be considerable so
the speed range over which the acceleration takes place is chosen
not to include standstill.
Over a sufficiently small speed range the braking torque is ap-
proximately constant and by measuring both the acceleration and
deceleration over this speed range the effect of the unknown torque
can be minimized. Without the braking torque the acceleration
would be
T
ω̇ = (4.4)
J
Starting performance 57

but with addition of the unknown, but assumed constant, braking

torque TX the acceleration is slower:
T − TX
ω̇1 = (4.5)
J
The deceleration on the other hand is faster:
−T − TX
ω̇2 = (4.6)
J
Combining 4.5 and 4.6 and rearranging yields
2T
J= (4.7)
ω̇1 − ω̇2
The acceleration ω̇1 and deceleration ω̇2 are calculated by least
squares fitting of two straight lines to the measured rotational
speed as shown in [Figure 4.3]. The electromagnetic torque used

160
155
Speed ωr [rad/s]

150
145
140
135
130
125

0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

Time [s]

Figure 4.3: Inertia measurement, rotational speed when acceler-

ated and decelerated by a constant torque of 1 Nm.

was 1.0 Nm and the total inertia of the test setup measured with
this method was
J = 6.1 · 10−3 kgm2 (4.8)
which is close to what can be expected.
The asymmetry of [Figure 4.3] shows that the braking torque is
considerable:
−J (ω̇1 + ω̇2 )
TX = ≈ 0.34 Nm (4.9)
2
and that using only one of the slopes to calculate the moment of
inertia would be very inaccurate.
58 4. Measured Performance

Friction
The friction torque of the setup was evaluated both by a roll-out
test and by stationary measurements at a number of different
speeds. The measured friction torque includes the windage and
iron losses but as the measurement was performed only to verify
that these torque components were small enough not to affect the
other measurements too much this was of little concern.
For the stationary measurements the setup was driven by the
load machine at 49 different, constant, speeds ranging from close to
standstill up to nominal speed. At each speed both the torque mea-
sured by the torque transducer, Tmeas and the calculated torque
supplied by the load machine, Tref , were recorded and are shown
in [Figure 4.4]. The torque measured by the torque transducer

0.4
Troll
Friction torque [Nm]

0.35

0.3 Tref

0.25
Tmeas
0.2

0.15
0 50 100 150
Speed ωr [rad/s]

Figure 4.4: Friction torque, from stationary measurements, Tmeas

and Tref , and from roll-out test, Troll .

does not include the friction in the load machine so it should be

lower than the reference torque as is the case in [Figure 4.4]. The
torque transducer has, however, an arbitrary offset that has been
subtracted, hence the offset of Tmeas vs. Tref is of little significance,
and the fact that Tmeas < Tref at most speeds is merely a coinci-
dence.
For the roll-out test the setup is driven to nominal speed by the
load machine and then let to decelerate by the friction torque alone.
The speed then decreases as shown in [Figure 4.5]. Differentiating
the speed numerically and multiplying by the moment of inertia
measured in the previous section yields the uppermost curve, Troll ,
in [Figure 4.4]. The periodic ripple is caused by position dependent
torque, the spectrum of which is shown in in [Figure 4.6]. The spec-
Starting performance 59

150
Speed ωr [rad/s]

100

50

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Time t [s]

Figure 4.5: Roll-out test.

tral resolution is poor due to the fact that the roll-out test takes
a limited time that cannot be increased without altering the setup
or affecting the measurement. The reason for the relatively large,
but not troublesome, first harmonic is unknown but could be due
to shaft misalignment and the associated forces in the shaft cou-
plings.

0
Friction torque [dBNm]

-10

-20

-30

-40

0 1 2 3 4 5 6
Space harmonic number [ ]

Figure 4.6: Spectrum of position dependent friction.

In theory there are torque components, such as the windage

torque, that are proportional to the square of the rotational speed.
From [Figure 4.4] it is clear that any such components are masked
by much larger constant and linear components and also additional
position dependent friction. A linear fit to the roll-out data is thus
60 4. Measured Performance

reasonable and is shown as a dashed line in [Figure 4.4]

Tf riction = bω + c (4.10)

where
Nm
b ≈ 7 · 10−4 , c ≈ 0.3 Nm (4.11)
rad/s
The conclusion of this measurement is that there is a constant fric-
tion component with a magnitude of approximately 0.3 Nm and a
torque component varying linearly with speed contributing approx-
imately 0.1 Nm at nominal speed.

Virtual inertia
The mechanical setup can be represented by the inertias of the mo-
tor under test, J1 , and the inertia of the load machine, J2 connected
by a stiff shaft as shown in [Figure 4.7]. The sum of the electro-

T1 J1 J2 T2

Figure 4.7: The mechanical system with a stiff connection between

the two rotors.

magnetic torques, T1 and T2 , applied to the two rotors accelerate

the system. By modulating the torque of the load machine, T2 , ac-
cording to the acceleration of the system:

dωr
T2 = −JV (4.12)
dt
the load, as seen from the test motor, behaves as a pure inertial
load. The inertia of the load as experienced by the test motor is

JL = J2 + JV (4.13)

and the load machine thus emulates a virtual inertia of magnitude

JV . The virtual inertia control loop is shown in [Figure 4.8].
The amount of inertia that can be emulated depends on a few
important factors:

• The maximum torque that the load machine can supply.

• The resolution, noise and bandwidth of the angular measure-

ment, in this setup measured with a resolver.
Starting performance 61

• Limitations of the mechanical system.

The first factor is the least troublesome. An infinite moment of in-
ertia can be emulated by keeping the rotor locked in position, then
the load machine must be able to counteract the torque of the mo-
tor under test. The maximum torque of the motor under test can
be estimated in a few different ways depending on what is known
of the motor under test.

T1 1 ω̇
+
J1 +J2

T2

Jv

Figure 4.8: Simplified virtual inertia control loop.

The mechanical system is not as simple as presented in [Figure

4.7] as the shaft connecting the two rotors is not infinitely stiff and
acts as a torsional spring, thus affecting the dynamic behavior of
the system. The mechanical system including this spring is shown
in [Figure 4.9] where k is the torsional stiffness of the spring and
Tk is the spring torque. The relation between the torque supplied

Tk

k
T1 J1 J2 T2

Tk

Figure 4.9: The mechanical system with a torsional spring be-

tween the two rotors.

by the load machine, T2 , and the load machine rotor angle, θ2 , can
be derived in the following way: The rotational acceleration of the
load machine rotor is
T2 − Tk
θ̈2 = (4.14)
J2
and the rotational acceleration of the motor under test rotor is, as-
suming that the electromagnetic torque of the test motor is zero,
Tk
θ̈1 = . (4.15)
J1
62 4. Measured Performance

The spring torque is

Tk = k (θ2 − θ1 ) . (4.16)

Making use of the Laplace transform and combining the two latter
equations:
k (θ2 − θ1 )
s2 θ1 = (4.17)
J1
Rearranging yields an expression for the rotational angle of the
test motor:
k
θ1 = θ2 . (4.18)
J1 s2 + k
Inserting the spring torque expression (4.16) into (4.14) and using
(4.18)
T2 − k (θ2 − θ1 )
s2 θ2 = (4.19)
J2
 
k
2
J2 s θ2 = T2 − k θ2 − θ2 (4.20)
J1 s2 + k
The transfer function from load machine torque to load machine
rotor angle can then subsequently be found to be:

θ2 J1 s2 + k
= (4.21)
T2 J2 J1 s + (J2 + J1 ) ks2
4

or relating the acceleration to the torque:

θ̈2 J1 s2 + k
= 2
. (4.22)
T2 J1 J2 s + (J1 + J2 ) k

The system has poles in

s
(J1 + J2 )
s = ±j k (4.23)
J1 J2

and a corresponding resonance frequency of

s
1 (J1 + J2 )
fR = k . (4.24)
2π J1 J2

The stiffness of the shaft is mainly limited by the torque transducer

as it uses a torsional spring to convert torque to angular displace-
ment which is subsequently measured. The stiffness of the torque
transducer taken from the data in Appendix A is

k = 6.4 kNm/rad (4.25)

Starting performance 63

The calculated inertias of the two rotors including the couplings are
roughly equal and their sum given by the measurement described
in the preceding section, thus
J1 = J2 = 3.05 · 10−3 kgm2 (4.26)
and the calculated resonance frequency is
fR = 326 Hz . (4.27)
The resonance frequency was measured by an impulse response
test of the system. The impulse was generated by tapping the load
machine axle coupling in a tangential direction with a hammer.
The impulse response is shown in [Figure 4.10], clearly showing
the mechanical oscillations. The frequency spectrum of the same

8
Rotational speed [rad/s]

-2
0 0.05 0.1 0.15 0.2
Time [s]

Figure 4.10: Impulse response of the dual-mass mechanical sys-

tem showing significant oscillations.

test is shown in [Figure 4.11] where the marked peak is located at

290 Hz. The resonance frequency of 290 Hz differing from the cal-
culated value implies that at least one of the quantities used in
Equation 4.24 is inaccurate. As the sum of the inertias is known
with fairly good accuracy from the measurement described in the
preceding section and the assumption that the inertias J1 and J2
are equal yield the lowest resonant frequency this suggest that the
total torsional stiffness of the system is substantially lower than
the 6.4 kNm/rad of the torque transducer. Back calculations give a
total torsional stiffness of 5.1 kNm/rad and which would be the ef-
fect of an extra torsional spring with a stiffness around 25 kNm/rad
which is a plausible value for the combined stiffness of the machine
axles. Any other division of the inertia would of course mean that
the shaft has even lower stiffness.
64 4. Measured Performance

0
Relative rotational speed [dB]

-10

-20

-30

-40

-50

-60

-70
0 500 1000 1500
Frequency [Hz]

Figure 4.11: Frequency spectrum of the impulse response in [Fig-

ure 4.10].

The mechanical resonance limits the amount of virtual inertia

that can be used in this setup. The gain in the virtual inertia con-
trol loop [Figure 4.8] is
JV
|G| = (4.28)
J1 + J2
and at the mechanical resonance frequency the loop gain must be
low enough to avoid oscillation. To ensure this a low pass filter
is inserted before the Jvirtual gain block in [Figure 4.8]. To allow
large virtual inertias the low pass bandwidth has to be chosen very
small, causing significant phase shift and amplitude errors in the
filtered acceleration signal. This is not desirable but troublesome to
get around, the starting behavior with this type of load behavior is
still of interest as real world applications generally provide limited
stiffness if large inertias are present. Improved control strategies,
such as described in [17], can offer improved performance and as-
sociated increased virtual inertia range. Due to time limitations
such measures have not been implemented. The inertia values in
the following sections should thus not be taken as correct absolute
values but rather as relative measurements.

4.2 A nominal start

The typical start behavior is presented in this section, the different
quantities are measured during a cold start, i.e. motor at room tem-
perature, with nominal torque. The nominal output power of the
test motor is 750 W which corresponds to nominal torque of 4.8 Nm.
A nominal start 65

The actual torque used for the measurements in this section was
5.0 Nm but the starting behavior should not differ appreciably from
the nominal. No extra virtual inertia was used, only the physical
inertia of the system, measured to around 6.1 · 10−3 kgm2 , limits
the acceleration. The accelerating torque is measured by a torque
transducer but as a considerable part of the rotational inertia in
the setup is due to the rotor of the motor under test the measured
torque will differ from the electromagnetic torque developed by the
motor under test. If the inertia of the rotor is known the electro-
magnetic torque can be calculated by differentiation of the angular
velocity:
dωr
Tem = Tmeas + J1 (4.29)
dt
The measured torque vs. time is shown as a solid line in [Fig-
ure 4.12] and the calculated total electromagnetic torque is shown
dashed. Early in the start there is a large torque ripple, most of

25
20
15
Torque [Nm]

10
5
0
-5
-10

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45

Time [s]

Figure 4.12: Measured torque (solid) and estimated total electro-

magnetic torque (dashed) of the motor under test for nominal start-
ing conditions.

which is counteracted by the reaction torque of the rotational iner-

tia of the setup. Although the large pulsating torque has negative
peaks decelerating the rotor for part of the time the rotor is acceler-
ated by a positive average torque up to synchronous speed as shown
in [Figure 4.13]. The rotor is accelerated up to synchronous speed
in roughly 0.25 s during which the output power is much larger than
the nominal output power, consequently the input power must be
much larger as well. As the input voltage is fixed this extra power is
produced by an increased input current. The input current during
the start is around five times larger than the steady state current
66 4. Measured Performance

150
Rotational speed ωr [rad/s]

100

50

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45

Time t [s]

Figure 4.13: Measured speed vs. time. Nominal load torque and
line voltage.

as is evident in [Figure 4.14]. This current puts certain limitations

40
Phase currents [A]

20

-20

-40

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45

Time [s]

Figure 4.14: Phase currents vs. time. Nominal load torque and line
voltage.

on the mains fuse that can be used for supplying the motor. It must
have sufficiently slow characteristics and high current rating so
that it does not melt during the start. The large starting currents
cause a voltage drop at the motor terminals due to the line imped-
ance as shown in [Figure 4.15] the effect of which is however not so
severe.
Starting Capability 67

150

100
Phase voltages [A]

50

-50

-100

-150
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45
Time [s]

Figure 4.15: Phase voltages vs. time.

Worst case
To find the worst case conditions for a start, even regarding only the
rotor position and the phase of the mains at turn-on is not trivial.
The conditions used for the tests are peak voltage in phase R at
turn-on, i.e. us along the α-axis and the rotor positioned with d-axis
in β-direction. This means that the rotor initially will start turning
the wrong way, noticeable in [Figure 4.13]. While not perhaps the
worst case, this is at least not the most favorable condition.

4.3 Starting Capability

Successful start tests with load torques up to 9 − 12 Nm depending
on supply voltage were performed, at higher torques the motor fails
to synchronize. From the speed vs. phase plot in [Figure 4.16] it is
clear that the motor synchronizes in only a few revolutions even for
the higher load torques. For certain combinations of supply volt-
age, starting position, load torque and inertia the motor just barely
fails to synchronize during the slip cycle preceding the synchro-
nization. During this last slip cycle the slip frequency is very low
and thus the current can remain large for a considerable amount of
time. From a fuse point of view there are, for given load torques and
inertias, starting positions causing a much worse start compared to
the other positions. For the starts in [Figure 4.16] the starting po-
sition and load inertia are the same but for the load torque of 1 Nm
the fuse is put under more stress than for much larger load torques.
Start test with an additional virtual inertia shows that the load
torque capability is reduced when additional inertia is present as
68 4. Measured Performance

150
Rotor speed [rad/s]

TL = 1 Nm
100
TL = 11 Nm

50

0 5 10 15 20 25 30 35 40 45
Load angle δ [rad]

Figure 4.16: Phase plane for load torques TL = 0, 1...11 Nm.

noted in [14]. The combinations of load torque and inertia used for
the tests are shown in [Figure 4.17] in which also the outcome of
the tests are shown.

4.4 Magnet Braking Torque

The magnet braking torque is not only speed dependent as ex-
pressed by the quasi-static model but also time dependent which
is not included in the model. This time dependence was investi-
gated by short-circuiting the motor terminals and then using the
load machine to increase the speed linearly with time. The result-
ing braking torque for different speed ramps are shown in [Figure
4.18]. As the rotor is accelerated the magnet flux is pushed into the
air-gap by the short-circuited stator winding. This happens fairly
fast, with a time constant around

Lls
τs = (4.30)
Rs
The rotor cage keeps this flux outside the cage, i.e. in the air-gap
and teeth-top area, for a duration comparable to the rotor time con-
stant
Llr
τr = (4.31)
Rr
As the cage currents decay the magnets experience a larger reluc-
tance and the magnet flux decreases. At the same time the mag-
netic scalar potential in the rotor "pole" pieces increases and an
increased portion of the flux through the magnets do not link with
Magnet Braking Torque 69

180V
0.02

0.01

190V
0.02

0.01

0
Load inertia [kgm2]

200V
0.02

0.01

210V
0.02

0.01

220V
0.02

0.01

0
0 2 4 6 8 10 12
Load torque [Nm]

Figure 4.17: synchronization tests with supply voltages from 180 V

to 220 V. ’◦’s denote successful starts and ’×’s are failed synchro-
nizations. ’∆’s are tests that have failed due to load motor limita-
tions.

the cage as the flux passes through iron bridges and air-gaps link-
ing the magnetic poles in the rotor.
70 4. Measured Performance

20
0.07s
18
Magnet braking torque [Nm]

16

14 0.19 s

12

10 0.28 s

8
0.46 s
6

4 1.01 s

0
0 50 100 150
Rotor speed δ [rad/s]

Figure 4.18: Magnet braking torque for speed ramps with dura-
tions of 0.07, 0.19, 0.28, 0.46, 0.63, 0.82, 1.01 s. The torque calculated
by the analytical expression in 2.10 using the measured equivalent
circuit parameters is shown dashed.

The quasi-static magnet braking torque expression given in

2.10 is fairly complex but can be substantially simplified for certain
operating conditions. For low rotor speeds the induced currents are
of low frequency, hence the influence of the inductances diminishes.
The braking torque can then be simplified to

mp E02
Tb = − (1 − s) (4.32)
2ωs Rs
i.e. the slope of the initial braking torque only depends on the sta-
tor resistance and the induced EMF. For very low slips, i.e. close
to synchronization, the reactance terms are much larger than the
stator resistance terms and Equation 2.10 can be simplified to
mp Rs
lim Tb = − E0 (4.33)
s→0 2ωs Xd2

The rotor cage not only maintains an increased permanent flux

during the early start, it also maintains a decreased flux level dur-
ing the synchronization process which can affect the synchroniza-
tion negatively. When approaching synchronism the magnet flux
Starting Current and Fuse Concerns 71

linking with the stator winding will increase. The rotor cage will
however oppose this change and thus the magnet and reluctance
torque can not build up infinitely fast. This is indicated by a test
where the short-circuited stator winding is opened at synchronous
speed. The open circuit voltage, shown in [Figure 4.19] will quickly
attain a level (≈ 40 V) corresponding to the magnet flux that was
already linked with the cage during the short circuit. The open cir-
cuit voltage will then approach the steady state value (≈ 110 V)
with the time constant τr . Magnet flux leakage outside the cage is
thus good as it lowers the magnet braking torque while providing
synchronization torque almost instantly.

140

120

100
E0 [V]

80

60

40

20

0
-0.05 0 0.05 0.1 0.15 0.2 0.25 0.3
Time [s]

Figure 4.19: Magnet EMF after a short circuit.

4.5 Starting Current and Fuse Concerns

Although electronic fuses and motor protection circuits are gaining
popularity the majority of the general purpose pump motors are
protected by melt-type fuses. The current draw must thus be eval-
uated mainly from a melt-type fuse perspective. The fusing process
is highly complex [23] and although some characteristics are out-
lined in standards for low voltage fuses such as IEC 1271 and IEC

1 International Electrotechnical Commission

72 4. Measured Performance

60269, the exact characteristics can not be known without intimate

knowledge of the fuse used.
The current during starting is considerably larger than the cur-
rent consumption at steady state nominal operation. The start-
ing current is limited mainly by the leakage inductance but also
the line impedance and stator resistance affect the current magni-
tude. For the supply voltage range this maximum current is around
20 ARMS . This can be compared to the nominal current of 3−4 ARMS .
In choosing an appropriate fuse, or in designing a motor for a
certain fuse, there are mainly two concerns:

• For a normal and successful start, with a load attached that

is within specification, the fuse should not blow.
• For unsuccessful starts, for example if the rotor is locked in
position due to objects stuck in the impeller or if the motor
fails to synchronize due to too large inertia and load torque
or low supply voltage, the line fuses should blow before other
damage occurs.
For the first consideration one way to characterize a fuse is the time
integral of the current squared it takes to quickly melt the fuse,
often referred to as the i2 t-value of the fuse. This is a meaningful
measure as long as the time interval of the over current condition is
so short that negligible heat is transferred from the fuse link to the
fuse cartridge and terminals before the fuse link melts. Then the
energy dissipated in the fuse link, and thus also the temperature
and tendency to melt, is roughly proportional to this integral.
The over current condition during successful or unsuccessful
starts have durations that are comparable to or longer than the
time constant of the fuse. The time/current operation characteris-
tics are then much more complex but some values relating over cur-
rent factor to fusing time are shown in [Table 4.1]. The time/current
operation characteristics are taken from an application guide from
Socomec [21] and should correspond to those described in IEC 127.
The fusing time increases exponentially as the over current magni-
tude approaches the fuse rated current. This exponential relation
is valid down to fractions of seconds, for higher currents the i2 t is
more appropriate. As the normal starting only requires a few sec-
onds the fuse can be chosen so that the fuse will operate within
tens or hundreds of seconds at the current level present when not
synchronized. The rated fuse current most commonly available, for
example in Diazed, or ’D’-type fuses, follows a geometric series with
10 divisions per power of ten, i.e.
1.0, 1.2, 1.6, 2.0, 2.5, 3.2, 4.0, 5.0, 6.3, 8.0, 10, 12... A (4.34)
Starting Current and Fuse Concerns 73

Table 4.1: Fuse characteristics, approximately gG-characteristics

Fusing time (pre-arcing) Over-current

1000 s 1.7×
100 s 2.6×
10 s 3.1×
1s 4.3×

and so on. To have the fuse operate in 10-100s when the motor fails
to synchronize, the fuse current rating should be chosen to be ap-
proximately one third of the current draw when not synchronized.
This fuse rating should also accommodate the requirement that the
fuse should not melt during normal motor operation. Large leakage
reactance are desirable when it comes to efficiency and supply volt-
age range but a too large impedance can limit the unsynchronized
current to a level that put demands on the fuse that are hard to
meet. This problem can to some degree be alleviated by electronic
motor protection circuits and circuit breakers.
The starting current magnitude is not so dependent on load
but varies greatly with supply voltage. Thus for a certain supply
voltage the starting time alone sets the requirements on the fuse.
Starting times for various supply voltages and loads are shown in
[Figure 4.20]. As can be expected the starting time depends heav-
ily on the inertia but also on the supply voltage. [Figure 4.20] does
however not contain any information on the current magnitude
which also depends on the supply voltage and affects the fuse op-
eration. As noted earlier the fuse operation time is roughly inverse
proportional to the current squared so even though the i2 t-value is
not strictly applicable it still shows useful information about the
stress that the fuse is put under. The i2 t-values integrated over
the starting time are shown in [Figure 4.21]. Note that these i2 t-
values do not correspond to the values supplied by manufacturers
as these are a measure of the maximum energy the fuse will let
through during a short circuit. From [Figure 4.21] it is clear that
a fuse is most likely to blow when the motor is connected to a low
supply voltage and large inertia.
As the current is limited by the leakage inductance regardless
of load the i2 t values are proportional to the starting time, which is
closely tied to the total inertia as is evident from the nearly hori-
zontal bands in [Figure 4.21].
74 4. Measured Performance

180V
0.02 0.8 1

0.015 0.6
0.01
0.4
0.005
0
190V
0.02 0.8
0.6
0.015
0.01 0.4

0.005
0.2
0
Load inertia [kgm2]

200V
0.02 0.6

0.015
0.01 0.4

0.005
0.2
0
210V
0.02
0.015 0.6
0.4
0.01
0.005
0.2
0
220V
0.02
0.015
0.4
0.01
0.005 0.2

0
0 2 4 6 8 10 12
Load torque [Nm]

Figure 4.20: Starting time in seconds [s] as a function of load

torque and inertia with supply voltages from 180 V to 220 V.

4.6 Steady state performance

The focus of the test setup is on transient performance measure-
ments but it can also be used for steady state measurements. The
Steady state performance 75

180V
0.02 400
300
0.015
0.01 200
0.005
0
400 190V
0.02 300
0.015
0.01 200

0.005
0
Load inertia [kgm2]

200V
0.02 300
0.015
0.01 200

0.005
0
210V
0.02
300
0.015
200
0.01
0.005
100
0
220V
0.02
0.015
0.01 200

0.005
100
0
0 2 4 6 8 10 12
Load torque [Nm]

Figure 4.21: Starting i2 t-values, in [A2 s], for supply voltages from
180 V to 220 V.

accuracy is however limited and not as good as can be expected of a

dedicated steady state brake bench. For the steady state operation
it is mainly the efficiency and displacement power factor that are
of interest, shown in [Figure 4.22] and [Figure 4.23] respectively.
76 4. Measured Performance

The efficiency is high over a fairly wide torque range (50 − 150%

0.85 220V

0.8 180V 180V

Efficiency [ ]

0.75
0.7 220V

0.65
0.6

0.55

0.5
0 2 4 6 8 10 12
Load torque [Nm]

Figure 4.22: Efficiency as function of load torque for supply volt-

ages 180, 190, 200, 210, 220 V.

of nominal) for all the tested supply voltages. The motor is rather

0.8 180V
220V
0.6
cosφ [ ]

0.4

0.2

0
0 2 4 6 8 10 12
Load torque [Nm]

Figure 4.23: Displacement power factor as function of load torque

for supply voltages 180, 190, 200, 210, 220 V.

under-magnetized, even at 10% lower supply voltage than nomi-

nal, possibly due to the design being made for 200 V, 60 Hz. With a
50 Hz supply the magnet EMF is only 68% of the nominal supply
voltage (200 V), thus the DPF is fairly low for all load torques. The
displacement power factor suffers at high supply voltages and low
load torques. This also shows as slightly lower efficiency due to the
increased current magnitude.
Steady state performance 77

Besides the efficiency and DPF the synchronous torque as a

function of load angle is of interest. If the load angle changes
quickly there will be currents induced in the squirrel cage of the
rotor. Thus the torque is dependent on the rate of change in load
angle. The transient torque can be many times larger than the
steady state torque [4]. To ensure that little current is induced in
the cage a long sweep time is used to sweep the load angle. There
will however be currents induced anyway which will depend on the
direction of the sweep, this is shown in [Figure 4.24]. Due to finite

10

0 Generating
Torque [Nm]]

-5 Motoring

-10

-15

-20

-150 -100 -50 0 50 100 150

Load angle δ [electrical degrees]

Figure 4.24: Synchronous torque, load angle sweep in positive and

negative directions shows hysteresis due to currents induced in the
cage. Supply voltage 200 V and sweep time ≈ 13 s.

gain in the load angle control loop the transition from motoring to
generating mode and vice versa on the negative slope is fairly fast
compared to the rest of the sweep. Thus it is in this region, δ > 120◦
and δ < −120◦ , that it is evident that currents are induced in the
cage. By averaging the negative and positive sweeps the effect of
the cage torque can be minimized. The synchronous torques mea-
sured in this way are shown in [Figure 4.25]. The maximum torque
amplitude in generating mode is larger than when motoring as all
the losses serve to brake the rotor in both modes.
78 4. Measured Performance

220V
10
180V
5

0 Generating
Torque [Nm]]

-5 Motoring

-10

-15 180V

-20 220V

-150 -100 -50 0 50 100 150

Load angle δ [electrical degrees]

Figure 4.25: Synchronous torque, average of positive and negative

sweeps. Supply voltages 180, 190, 200, 210, 220 V and sweep times ≈
13 s.
Chapter 5

Calculated Performance
and Comparison to
Measurements

The behavior and performance of the LSPMSM can be calculated

using various analytical expressions. The expressions are generally
based on equivalent circuits and make use of equivalent circuit pa-
rameters, the values of which are taken from the measurements de-
scribed in Chapter 3. Compared to the performance measurements
described in Chapter 4 the expressions provide a comparably rough
picture. They, however, provide insight and means for both initial
designs and optimization.
A more accurate performance evaluation can be made by nu-
merically solving the machine equations in Section 2.3.

5.1 Starting Performance

The quasi-static model described in Section 2.3 can be used to pro-
vide a rudimentary simulation of the starting sequence. By using
the quasi-static torque expressions for cage, magnet braking and
load torque and then numerically solving the mechanical differen-
tial equation:
dω Tk (ω) + Tb (ω) + Tl (ω)
= (5.1)
dt J
results in an approximate start sequence.

Starting Torque
The expressions for the cage and magnet braking torque used in
Equation 5.1 are derived from different equivalent circuits.
80 5. Calculated Performance and Comparison to Measurements

The cage torque can be derived using the equivalent circuit

in [Figure 2.11] and making use of Thevenins theorem as shown
by Sen in [18]. This yields a fairly simple expression for the cage
torque:
2
m p Vth 1
Tk = R 0  2 (5.2)
2 ωs r R 0 2
s Rth + sr + (Xth + Xlr )

where
Xm
Vth = q Us (5.3)
2
Rs2 + (Xls + Xm )
2
Xm Rs
Rth = 2 (5.4)
Rs2 + (Xls + Xm )
2
Rs2 + Xls + Xls Xm
Xth = Xm 2 (5.5)
Rs2 + (Xls + Xm )
The effect of closed rotor slots can be treated as described by
Nee[16]. The torque expression can be derived in a similar manner
from the equivalent circuit in [Figure 2.12]. The rotor bridge volt-
age, Urb is however hard to measure as described by Nee due to the
permanent magnets. Therefore, a value calculated from the dimen-
sions of the bridge is used instead. The resulting torque expression
is significantly more complex:

m p Rr0  2 p 
Tk = F − F F 2 − 4G − 2G . (5.6)
2 ωs s

where
 2 
0 1 Xl s 1 Xl s Rs2 0
2Urb Xm + Zs2 + Xm + Zs2 + Zs4 Xrl
F =   2    
0
Rs2 Xrl Rr0
1 + Zr0 2 Zs4 + 1
Xm + Xl s
Zs2 +2 Xm + 1
Zs2 Rs s
0
+ Xls Xlr

 (5.7)
 2 
0 Rs2 1 Xl s Us2
Urb2 Zs4 + Xm + Zs2 − Zs2
G=   2    
0
Rs2 Xrl Rr0
1 + Zr0 2 Zs4 + 1
Xm + Xl s
Zs2 +2 Xm + 1
Zs2 Rs s
0
+ Xls Xlr
(5.8)
and
Zs2 = Rs2 + Xls
2
(5.9)
 0 2
0 Rr 0
2
Zr = + Xlr2 (5.10)
s
Starting Performance 81

The cage torque as a function of rotor speed calculated using equa-

tion 5.2 and 5.6 are shown in [Figure 5.1]. The expression taking
into account the saturable rotor bridges predicts a somewhat lower
cage torque and is the expression used in the following analysis.

15
Torque [Nm]

10

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Normalized rotor speed

Figure 5.1: Quasi static cage torques for the equivalent circuits in
[Figure 2.11] and [Figure 2.12]

The magnet braking torque expression can be derived from the

equivalent circuit in [Figure 2.2] as shown by Herslöf [7] yielding
equation 2.10. Both the cage and magnet braking torque as func-
tions of rotor speed are shown in [Figure 5.2] together with the
quadratic load torque. As can be seen there is a good torque mar-

14

12 Tk

10
Torque [Nm]

6
-Tb
4

2 -Tl

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Normalized rotor speed

Figure 5.2: Asynchronous quasi-static torque components. Cage,

magnet braking and load torque as functions of rotor speed.
82 5. Calculated Performance and Comparison to Measurements

gin throughout the whole speed range.

Using these torque expressions to numerically solve equation
5.1 yields the rotor speed during a start as shown in [Figure 5.3].
Normalized rotor speed

0.8

0.6 Measured

0.4
Quasi-static
0.2

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
Time [s]

Figure 5.3: Rotor speed during starting as predicted by the quasi-

static model, compared to measurement.

Starting current
The induction motor equivalent circuits can also be used to find the
current magnitude during the starting sequence. Using the equiv-
alent circuit in [Figure 2.12] as for the cage torque but solving for
the line current instead of the air-gap power the line current mag-
nitude is
|is | = |ir + im | (5.11)
where ir and im are the rotor and magnetizing branch currents
respectively:
F 1p 2
ir = − + F − 4G (5.12)
2 2
where F and G are the same as for the cage torque calculation and
 0 
R
ir sr + jXlr + jUrb
im = . (5.13)
jXm
The calculated current is somewhat lower than the measured
line current and the current ripple is of course not included. The
current peaks put the line fuses under a lot more stress than what
is indicated by the quasi-static current calculation and the fuse rat-
ing should thus not be based on the quasi-static current calcula-
tion.
Synchronization 83

35
30
Measured
25
Is [ARMS]

20
15 Quasi-static
10
5
0
0 0.05 0.1 0.15 0.2
Time [s]

Figure 5.4: Current during starting as predicted by the quasi-

static model compared to measurement

5.2 Synchronization
By adding the synchronous, load angle dependent, torque compo-
nents to the differential Equation 5.1 the quasi-static model can be
extended to include synchronization.

dω Tk (ω) + Tb (ω) + Tl (ω) + Tm (δ) + Trel (δ)

= (5.14)
dt J
where ω is the angular speed, δ is the load angle and J is the to-
tal inertia of the motor and load. The sum of the asynchronous
torque components will bring the rotor up to a certain speed where
the total average torque declines to zero. From this speed the pul-
sating torque components must accelerate the rotor and load up
to synchronous speed in less than one cycle of the load angle δ for
synchronization to take place. The simulated start is shown in [Fig-
ure 5.5]. As can be seen there is still a large discrepancy between
simulation and measurement. The model can however still pro-
vide some useful information. Looking at the different torque com-
ponents during a critical synchronization, shown in [Figure 5.6],
it is clear that the asynchronous torque is comparably large dur-
ing a large part of the synchronization process. The reluctance
torque does not directly contribute to the synchronization ability
as noted by Miller in [14] as the average reluctance torque over
180 degrees is zero. The load angle variation for a critical synchro-
nization will however not vary over the whole 180 degrees as near
180 the torque sum would be negative. This would imply that a
saliency Xd > Xq would be beneficial as the whole positive part of
84 5. Calculated Performance and Comparison to Measurements

1
Normalized rotor speed

0.8 Measured

0.6
Quasi-static + synchronous
0.4

0.2

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45

Time [s]

Figure 5.5: Rotor speed during starting as predicted by the quasi-

static model with the addition of synchronous torque, compared to
measurement.

15
Tm+Trel
10
Torque [Nm]

5
Tk
0
Tb
-5
-10 Tl
-15
-20
0 0.05 0.1 0.15 0.2
Time [s]

Figure 5.6: Simulated torque components during a critical syn-

chronization.

the reluctance torque vs. load angle is utilized but only some frac-
tion of the negative part. This is however not the case. An Xd < Xq
will allow a larger load angle variation during a critical synchro-
nization and thus the cage torque can be quite large early in the
synchronization as shown in [Figure 5.6]. This effect is dominant
for reasonable values of Xd and Xq and can result in a large in-
crease in allowable moment of inertia.
Steady state performance 85

5.3 Steady state performance

Synchronous torque
The expressions for the synchronous torques given in Chapter 2,
Equations 2.7 and 2.8, do not take into account the stator resis-
tance or leakage inductance, nor the line impedance so the mea-
sured values differ significantly from the calculated as can be seen
in [Figure 5.8]. The discrepancy is not very large for low load
torques, up to nominal load torque, as the current magnitude is low
and thus also the voltage drop over these additional impedances.
Taking into account the additional impedances can yield more ac-
curate expressions, and a way to derive the appropriate expressions
follows: Starting with the phasor diagram in [Figure 5.7] the volt-

jXmq Iq

jXmd Id
Us
jXls Is

Rs Is

δ
ϕ Is
q
d

Figure 5.7: Phasor diagram for determining the synchronous

torque. Stator and line impedances are combined and represented
by Rs and Xls

age equation along each axis yields:

      
usd Rs −Xq isd 0
= − (5.15)
usq Xd Rs isq E
86 5. Calculated Performance and Comparison to Measurements

where
Xd = Xmd + Xls (5.16)
Xq = Xmq + Xls (5.17)
and
usd = −Us sin(δ) (5.18)
usq = Us cos(δ) . (5.19)
Note that the stator leakage reactance Xls is assumed to be the
same along both the d- and q-axes. Solving for the stator currents
yields
    
isd 1 Rs Xq usd
= 2 (5.20)
isq Rs + Xld Xlq −Xd Rs usq − E

The expression used by Honsinger [8] is used to calculate the syn-

chronous torque:
mp
Te = (Ψsd isq − Ψsq isd ) (5.21)
2
and the flux linkages are, assuming that there are no rotor cur-
rents:
Ψsd = Lmd isd + Ψm (5.22)
Ψsq = Lmq isq (5.23)
which when inserted into Equation 5.21 yields
mp
Te = ((Lmd − Lmq ) isd isq + Ψm isq ) (5.24)
2
Inserting the expression for the stator currents from Equation 5.20
leads to the somewhat cumbersome final expression for the syn-
chronous torque:

mp (Rs usd + Xq (usq −E)) (Rs (usq −E) − Xd usd )
Te = (Lmd −Lmq ) 2
2 (Rs2 + Xd Xq )

(Rs (usq −E)) − Xd usd
+ Ψm (5.25)
(Rs2 + Xd Xq )

The torque calculated using this expression is shown in blue in

[Figure 5.8] and corresponds better to measurements than the sim-
pler expression of Equation 2.6. Most notably it incorporates the
asymmetric properties of the measured torque, i.e. the decelerat-
ing torque is larger than the accelerating torque. There is, how-
ever, still a large discrepancy between the calculated and mea-
sured torque but due to time limitations this has not been inves-
tigated further. The synchronous torque has a negative average
Steady state performance 87

which suggests that an additional quasi-static braking torque com-

ponent should be used for speeds close to synchronism, i.e. for an-
alyzing the synchronization process. The phenomena in the model
transition region, i.e. the transition from a quasi-static model to
a synchronous model, are however hard to model and no such at-
tempts have yet been tried.

20

15
220V
Torque reference [Nm]]

10
180V
5

0 Generating

-5 Motoring

-10 180V

-15

-20 220V

-150 -100 -50 0 50 100 150

Load angle δ [electrical degrees]

Figure 5.8: Calculated synchronous torques. The red curves cal-

culated using the expression used by Honsinger [8] and the blue
curves calculated using the extended expression taking into ac-
count the stator and line impedances. The measured torques from
[Figure 4.25] are shown in black.

Current magnitude and Displacement Power Factor

Equation 5.20 can also be used to derive the line current magnitude
and the displacement power factor. The line current magnitude
q
|is | = i2sd + i2sq (5.26)

is shown in [Figure 5.9]. The larger measured current magnitude

for load angles around 90◦ can be explained by a decreased q-axis
reactance due to saturation. The lower current magnitude at large
load angles is probably due to an appreciable increase in d-axis
88 5. Calculated Performance and Comparison to Measurements

RMS phase current [A]

15

10

0
0 20 40 60 80 100 120 140 160
Load angle [deg]

Figure 5.9: Calculated current magnitude (solid) compared to

measured current magnitudes shown as dots (·).

reactance as the magnet bridges go out of saturation due to the

negative d-current.
The displacement power factor can also be calculated from
Equation 5.20 as

cos ϕ = cos (arg(us ) − arg(is )) (5.27)

and is shown in [Figure 5.10].

0.8 180V
220V
0.6
cosφ

0.4

0.2

0
0 2 4 6 8 10 12
Load torque, [Nm]

Figure 5.10: Calculated displacement power factor in blue com-

pared to measured values in black.
Steady state performance 89

Efficiency
The efficiency of the motor is calculated taking into account only
the copper and iron losses. The copper losses are calculated using
the current magnitude from the preceding section. The iron losses
in the rotor are assumed to be negligible and the stator iron losses
are calculated separately for the stator teeth and yoke. The flux
density distribution is assumed to have the same shape as when ex-
cited by the permanent magnets alone but with an increased mag-
nitude due to the armature reaction. A sinusoidal flux distribution
is used for the stator yoke and a trapezoidal flux distribution in the
teeth [15]. The iron loss model used was of the form
 2
dB
WF e = Ch f B̂ a+bB̂
+ Ce (5.28)
dt

where WF e is in W/kg and the constants for the steel Newcore 800
0.65mm are

Ch = 0.0291 a = 1.45 b = 0.418 Ce = 28.6 · 10−6 (5.29)

0.9
220V
Efficiency [ ]

0.8 180V 180V

0.7 220V

0.6

0.5
0 2 4 6 8 10 12
Load torque [Nm]

Figure 5.11: Calculated efficiency in blue compared to measured

efficiency in black.

The results are presented in [Figure 5.11]. As can be seen, for

low torques the measured efficiency is lower because the windage
and friction losses are not included in the calculated efficiency. For
higher torques the low efficiency is mostly due to increased iron
losses.
Chapter 6

Conclusions and Future

Work

The focus of this thesis has been on measurements. Firstly on mea-

suring equivalent circuit parameters to predict motor behavior and
performance, and secondly on measuring this behavior and perfor-
mance for comparison to the model predictions. Design recommen-
dations are contained in the discussion throughout the thesis but
have not been concentrated into a distinct set or list of guidelines.
The major conclusions drawn from the work described in this thesis
are:

- It has been shown that quasi-static models are sufficient for

modeling certain figures of merit while they are inadequate
for others.

- Step-response adapted measurements can quickly and accu-

rately provide the majority of the equivalent circuit parame-
ters of interest. Furthermore, the current biasing possibilities
make it possible to evaluate the domains over which the d-
q model is adequate. Although not measured under rotating
conditions it has been shown that static measurements cor-
relate well with the rotating ones. The parameters are mea-
sured with sufficient accuracy to allow the performance of the
motor to be predicted to a satisfactory level.

- In general it is the synchronization capability that limits the

allowable load torque and inertia.

- The synchronization process is not accurately described by

the quasi-static model so the trade-off vs. steady state per-
formance is not simple.
92 6. Conclusions and Future Work

- The steady state current magnitude and displacement power

factor is accurately predicted by the equivalent circuit models.
- The steady state torque is not accurately modeled by the com-
mon expressions that do not take into account the line and
stator impedances. This is especially important for small mo-
tors, which are most common today for LSPMSM applica-
tions.

The steady state calculated performance correlates well with

measurements for nominal operating conditions. Thus the perfor-
mance of designs optimized using these models can be expected to
perform well.

Future work
In future work it would be wise to focus entirely on the synchro-
nization process and the modeling thereof. It would be interesting
to analyze the measured torque during synchronization in more de-
tail and see if it is possible to model accurately. Analyzing several
measured synchronizations with different inertia should make it
possible to extract torque components that can be accurately sep-
arated into either load angle or speed dependent components in
addition to multi-dependent torque components.
The efficiency, both measured and calculated, is high for a large
span of line voltages at a certain load torque. The factors behind
this should be investigated and if possible analytical expression of
this load torque should be derived.
The step response adapted measurement method could be eval-
uated further, most importantly measurements for combinations of
d- and q-axis currents should be made.
Appendix A

List of Equipment

Motor under test

Misc. Data 200 V 50 Hz 3.0 A 1500 rpm
200 V 60 Hz 2.6 A 1800 rpm
220 V 60 Hz 2.5 A 1800 rpm
Temp. Class E
Stator OD 135 mm
Rotor OD 84.5 mm
Air gap 0.25 mm
Stack length 80 mm
Cu mass 0.7 kg
Fe mass 6.7 kg
Magnet mass 0.24 kg
Rotor slots 32
Stator slots 36
Stator skew one stator slot pitch.

Load machine
Manufacturer Atlas Copco
Description AC Servo Motor.
Type AHR142C6-64S
Cont. Stall 11.3 Nm 15 A
Max. 6000 rpm 380 V 72 A
Resolver
Manufacturer Sagem
Type 21RX340308
DSP system
Manufacturer dSPACE Inc.
Description Real-Time Control System
R&D Controller Board
Type DS1104
94 A. List of Equipment

Torque transducer
Manufacturer KTR Kupplungstechnik GmbH
Description Optical Torque Measuring System.
Type DATAFLEX R 22/50NC
Rated Torque ±50 Nm
Inaccuracy 0.5 % FS
Limit Frequency 16 kHz
Torsional Stiffness 6383 Nm/rad
Appendix B

List of Symbols

All quantities are per-phase peak values unless otherwise noted.

x̂ peak value of time dependent quantity

ẋ time derivative
F Magnetomotive force [At]
I0 DC excitation current [A]
iphase Phase current [A]
iR Current in phase R [A]
id Stator d-axis current [A]
iq Stator q-axis current [A]
it Current through the motor terminals [A]
J Scalar moment of inertia
√ [kgm2 ]
j The imaginary unit, −1
p Number of magnetic poles
L Inductance [H]
Lls Stator leakage inductance [H]
Llsd Stator d-axis leakage inductance [H]
Llsq Stator q-axis leakage inductance [H]
L0lr Rotor leakage inductance [H]
L0lrd Rotor d-axis leakage inductance [H]
L0lrq Rotor q-axis leakage inductance [H]
Lm Magnetizing inductance [H]
Lmd d-axis magnetizing inductance [H]
Lmq q-axis magnetizing inductance [H]
Ls Synchronous inductance [H]
Lsd d-axis synchronous inductance [H]
Lsq q-axis synchronous inductance [H]
96 B. List of Symbols

m Number of phases []
R Resistance [Ω]
Rext External resistance [Ω]
Rs Stator winding resistance [Ω]
Rr0 Rotor winding resistance transferred to the stator [Ω]
s Rotor slip
s Laplace parameter
t Time [s]
tx Some specific instant
T Torque [Nm]
Tref Load motor reference torque [Nm]
Tmeas Measured torque [Nm]
TS Synchronous torque [Nm]
Trel Reluctance torque [Nm]
Tk Asynchronous cage torque [Nm]
Tm Magnet synchronous torque [Nm]
Tb Magnet braking torque [Nm]
Tl Load torque [Nm]
u Voltage [V]
Us Stator voltage [V]
usd d-axis stator voltage [V]
usq q-axis stator voltage [V]
ut Voltage at the motor terminals [V]
X Reactance, X = ωs L [Ω]
Z Impedance [Ω]
δ Load angle, angle between Us and E [rad]
φ Angle between Us and Is [rad]
ω Angular velocity [rad/s]
ωs Line voltage angular velocity [rad/s]
ωr Rotor angular velocity [rad/s]
Ψsd Stator d-axis flux linkage [Wbt]
Ψsq Stator q-axis flux linkage [Wbt]
Ψrd Rotor d-axis flux linkage [Wbt]
Ψrq Rotor q-axis flux linkage [Wbt]
Ψm Magnet flux linkage [Wbt]
θ Rotor angle [rad]
τ Time constant [s]
τr Rotor time constant [s]
τs Stator time constant [s]
Appendix C

Abbreviations

AC Alternating Current
DC Direct Current
DPF Displacement Power Factor
DSP Digital Signal Processor/Processing
EMF ElectroMotive Force
FC Friction Compensated
IGBT Insulated Gate Bipolar Transistor
IM Induction (asynchronous) Motor
LSPMSM Line Start Permanent Magnet Synchronous Motor
PMSM Permanent Magnet Synchronous Motor
RMS Root Mean Square
SM Synchronous Motor
VFD Variable Frequency Drive
Bibliography

[1] Bernard Adkins. The General Theory of Electrical Machines.

Chapman and Hall Ltd., 1959.

[2] Official Energy Statistics from the U.S. Government. En-

ergy Information Administration. International energy an-
nual 2004, international electricity consumption, total electric
power, table 6.2.

[3] Official Energy Statistics from the U.S. Government. Energy

Information Administration. International energy annual
2004, international total primary energy consumption, table
e.1.

[4] A.E. Fitzgerald, Jr. Charles Kingsley, and Stephen D. Umans.

Electric Machinery. McGraw-Hill, fifth edition, 1990.

[5] Fredrik Gustavson. Elektriska Maskiner. Royal Institute of

Technology, Stockholm, Sweden, 1996.

[6] Bedrich Heller and Václav Hamata. Harmonic Field Effects in

Induction Machines. Elsevier Scientific Publishing Company.,
1977.

[7] Urban Herslöf. Design, Analysis and Verification of a Line

Start Permanent Magnet Synchronous Motor. Licentiate the-
sis, Royal Institute of Technology, Stockholm, Sweden, 1996.

[8] V.B. Honsinger. Permanent magnet machines: Asynchronous

operation. IEEE Transactions on Power Apparatus and Sys-
tems, 99(4):137–144, July 1980.

[9] Igor J. Karassik, Joseph P. Messina, Paul Cooper, and

Charles C. Heald. Pump Handbook. McGraw-Hill, third edi-
tion, 2001.

[10] P.K. Kovacs. Transient Phenomena in Electrical Machines. El-

sevier, 1984.
100 Bibliography

[11] Gabriel Kron. Equivalent Circuits of Electric Machinery. John

Wiley and Sons, 1951.
[12] P.H. Mellor, F.B. Chaaban, and K.J. Binns. Estimation of pa-
rameters and performance of rare-earth permanent-magnet
motors avoiding measurement of load angle. Electric Power
Applications, IEE Proceedings B, 6(138), 1991.
[13] T.J.E Miller. Methods for testing permanent magnet
polyphase motors. IEEE, IAS Conf. Rec., 23(D):494–499, 1981.
[14] T.J.E Miller. Synchronization of line-start permanent-magnet
ac motors. Transactions IEEE, PAS, 103:1822–1828, July
1984.
[15] Tomas Modeer. On stator yoke rotational iron losses in per-
manent magnet machines. page 647. ICEM, September 2006.
[16] Hans-Peter Nee. On rotor slot design and harmonic phenom-
ena of inverter-fed induction motors. PhD thesis, Royal Insiti-
tute of Technology, 1996.
[17] T.M. O’Sullivan, C.M. Bingham, and N. Schofield. Enhanced
servo-control performance of dual-mass systems. IEEE Trans-
actions on Industrial Electronics, 54(3):1387–1399, June 2007.
[18] P.C. Sen. Principles of Electric Machines and Power Electron-
ics. Wiley, second edition, 1997.
[19] D.W. Shimmin, J. Wang, N. Bennett, and K.J. Binns. Mod-
elling and stability analysis of a permanent-magnet syn-
chronous machine taking into account the effect of cage bars.
Electric Power Applications, IEE Proceedings, 142:137–144,
March 1995.
[20] Gordon R. Slemon. Electric Machines and Drives. Addison-
Wesley, 1992.
[21] Socomec application guide, fuses. http://www.socomec.
fr/, 2006.
[22] Charles M. Stephens, Gerald B. Kliman, and John Boyd. A
line-start permanent magnet motor with gentle starting be-
havior. IEEE, pages 371–379, June 1998.
[23] A. Wright and P.G. Newbery. Electric Fuses. Institution of
Electrical Engineers, second edition, 1995.