Modelling and control of a

dual three-phase PMSM

With focus on performance at various failure modes
Electric Power Engineering, MSc

DAVID AHLBÄCK
RAFAEL BAUSONE
DEPARTMENT OF ELECTRICAL ENGINEERING
C HALMERS U NIVERSITY OF T ECHNOLOGY
Gothenburg, Sweden 2021
www.chalmers.se
 Master’s thesis 2021

Modelling and control of a dual three-phase

PMSM

With a focus on performance at various failure modes

DAVID AHLBÄCK
RAFAEL BAUSONE

Department of Electrical Engineering

Division of Electric Power Engineering
Chalmers University of Technology
Gothenburg, Sweden 2021
Modelling and control of a dual three-phase PMSM
With focus on performance at various failure modes
DAVID AHLBÄCK
RAFAEL BAUSONE

© DAVID AHLBÄCK, 2021.

© RAFAEL BAUSONE, 2021.

Supervisor: Johan Åström, Aros Electronics

Examiner: Torbjörn Thiringer, Department of Electrical Engineering

Master’s Thesis 2021

Department Electrical Engineering
Division of Electric Power Engineering
Chalmers University of Technology
SE-412 96 Gothenburg
Telephone +46 31 772 1000

Cover: Cross-section of the prototype machine showing the winding arrangement.

Typeset in LATEX, template by Magnus Gustaver

Printed by Chalmers Reproservice
Gothenburg, Sweden 2021

iv
Modelling and control of a dual three-phase PMSM
With focus on performance at various failure modes
DAVID AHLBÄCK
RAFAEL BAUSONE
Department of Electrical Engineering
Chalmers University of Technology

Abstract
This thesis analyses the performance and different control methods of a dual-winding
PMSM under various fault-scenarios. The machine investigated has a segregated
dual-winding layout with no neutral connections. A physical-phase variable model
is implemented utilising FEM-data in order to model the non-ideal properties of the
system and to implement faults. Simulations are compared against experimental
results on a prototype machine with satisfactory results.

The focus is on the torque capability and torque ripple, where healthy dual operation
results are compared against three-phase open and one-phase open scenarios. Three
different post-fault control approaches are investigated. Analytical calculations are
performed for short-circuit faults and the harmonic contents of the different open-
phase scenarios are analysed. The main approach is to investigate control methods
where no additional hardware is required.

It is found that one winding can comfortably carry half the rated load if the other
is disconnected. The open-phase scenarios are mainly analysed for one winding at
the time where two different post-fault control methods are applied. The so-called
pre-fire and per-phase methods. Both prove to output higher torque in post-fault
operation than if no action is taken. A resonance controller method is also investi-
gated, which proves to reduce harmonic content and torque ripple for all operating
scenarios. However, it does not function satisfactorily as the sole open-phase post-
fault controller with the winding arrangement used. The analytical calculations
indicate a very low tolerance for short-circuit faults.

At low torque loads the machine is found to function remarkably well during single
phase-open operation. Post-fault control algorithms are however essential if a higher
torque output is required.

Keywords: PMSM, fault-tolerant, Physical phase-variable modelling, dual-winding,

FEM.

v
Acknowledgements
First of all, we would like to thank Aros Electronics and CTO Magnus Wide for
giving us the opportunity and taking us in as thesis workers. We want to thank our
main supervisor at Aros, Johan Åström, for all the time spent giving us great inputs
and guidance long the way. A huge thank you goes out to Spyros Gryparis, for all
the work making it possible for us to perform the lab tests, as well as for taking
time out of his busy days for fruitful discussions. We also want to thank Daniel
Chädström at Aros for providing us with the per-phase controller and the idea of
implementing it.

Finally, we would like to thank Torbjörn Thiringer for taking us on as thesis students
and for the swift and constructive feedback throughout the project.

David Ahlbäck, Rafael Bausone, Gothenburg, May 2021

vii
 Contents

1 Introduction 1
1.1 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Theory 5
2.1 PMSM machine design . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Windings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Modelling of PMSM machine . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Inverter theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 DQ-transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4.1 Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Control theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5.1 IMC controller design . . . . . . . . . . . . . . . . . . . . . . . 14

3 Fault tolerant drive systems 17

3.1 Possible faults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.1 Winding short-circuit Faults . . . . . . . . . . . . . . . . . . . 18
3.1.2 Open phase faults . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Fault mitigation approaches . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.1 Fault tolerant machines . . . . . . . . . . . . . . . . . . . . . 21
3.2.2 Fault tolerant Inverters . . . . . . . . . . . . . . . . . . . . . . 21
3.2.3 Post-fault control methods . . . . . . . . . . . . . . . . . . . . 23

4 Test object and set-up 27

4.1 Specifications of Motor . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 Inverter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.3 Test bench set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

5 Model Implementation 31
5.1 Simulation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.1.1 Dual winding model . . . . . . . . . . . . . . . . . . . . . . . 32
5.1.2 Simulink implementation . . . . . . . . . . . . . . . . . . . . . 32
5.1.3 Dual model implementation . . . . . . . . . . . . . . . . . . . 33
5.2 Fault implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.3 Control strategies and implementation . . . . . . . . . . . . . . . . . 34
5.3.1 Three phase-open fault control strategy . . . . . . . . . . . . . 35
5.3.2 Single open-phase fault control strategies . . . . . . . . . . . . 35

ix
 Contents

5.3.3 Resonant controller . . . . . . . . . . . . . . . . . . . . . . . . 39

6 Results and analysis 43

6.1 Performed tests and simulations . . . . . . . . . . . . . . . . . . . . . 43
6.1.1 Healthy Operation of the dual winding machine . . . . . . . . 44
6.1.2 Three-Phase Open Circuit . . . . . . . . . . . . . . . . . . . . 45
6.1.3 Comparison of tests . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2 Single-Phase Open Circuit . . . . . . . . . . . . . . . . . . . . . . . . 48
6.2.1 Lab verified open phase cases . . . . . . . . . . . . . . . . . . 48
6.2.2 Simulated open phase cases . . . . . . . . . . . . . . . . . . . 51
6.2.3 Single-Phase Open summary . . . . . . . . . . . . . . . . . . . 52
6.3 Balanced Short circuit . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.4 Additional performed tests . . . . . . . . . . . . . . . . . . . . . . . . 54
6.5 Resonance controller simulations . . . . . . . . . . . . . . . . . . . . . 54
6.5.1 Healthy dual operation . . . . . . . . . . . . . . . . . . . . . . 55
6.5.2 Three-phase open operation . . . . . . . . . . . . . . . . . . . 56
6.5.3 Dual winding operation with one phase open . . . . . . . . . . 58
6.6 Sustainability Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . 59

7 Conclusion 61
7.1 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

Bibliography 65

A Appendix 1 I
A.1 Current equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I
A.2 Dual winding Simulink model . . . . . . . . . . . . . . . . . . . . . . II

B Appendix 2 III
B.1 Resonant controller derivation . . . . . . . . . . . . . . . . . . . . . . III

x
 1
Introduction

For vehicles and many other applications, permanent magnet synchronous motors
(PMSMs) are often chosen by design engineers due to their high performance. For
certain applications, high reliability of the drive systems is a desirable attribute, and
it is therefore frequently investigated. Machines which are less likely to experience
a fault, or those whose performance is not significantly affected by a fault, are at-
tractive from the point of view of productivity, in production lines for an example.
Propulsion with electric motors, steering-by-wire, and other similar automated ap-
plications require very high reliability both in the software as well as the hardware
of the drive system, especially if the safety and comfort of passengers in a vehicle is
to be maximised.

Fault tolerance is, therefore, a field of electric motors which has been expanding in
for many years and will continue to develop further in the coming years. Approaches
to improve the fault tolerance and reliability of drive systems have been developed
for the electrical and mechanical parts of the drive; a fault-tolerant design is needed
in both the motor itself as well as the power electronics to maintain operational
functions in case of a fault occurring, [1, 2, 3].

The PMSM is already a considerable improvement over the electrically excited syn-
chronous machine. It does not require a rotor circuit since the permanent magnets
contribute a reliable amount of flux for torque generation. This leaves faults in the
machine to occur only in the stator winding. Faults in the converter, on the other
hand, typically occur due to power switch failure [4].

One such improvement is in the winding arrangement, which can have a profound
impact on the capabilities of the machine in case of a fault. A dual-winding arrange-
ment, in which windings are segregated to opposite sides of the stator and have their
own neutral point, is central in this thesis. This configuration essentially splits the
machine into two smaller and identical machines and, in case one fails, torque can
be produced normally with the other one.

Modelling of PMSM is well documented in literature, as well as fault analysis in

electrical machines. The three phase model of a salient PMSM is well known [5],
and modelling in the synchronous reference frame, with rotor-fixed dq coordinates is
also common. Physical phase-variable modelling using finite element analysis (FEA)
to obtain machine parameters was used in [6] for the sake of time-effectiveness as
well as increased accuracy in the model. Literature, however, is lacking a study

1
 1. Introduction

where this method is used to simulate a dual winding PMSM under different fault
scenarios. This thesis presents such a study as well as a validation of the model with
experimental tests on a prototype motor.

This thesis is performed within the framework of an ongoing design project at Aros
Electronics. A FEM model is available for simulation purposes and a prototype is
available for bench testing.

1.1 Aim
The purpose of this thesis is to investigate and evaluate the performance and fault-
tolerance of a dual three-phase permanent magnet motor designed for high reliability
applications. The purpose can be divided in to two main aims.

The first aim is to evaluate the performance of the prototype machine during differ-
ent fault and operating conditions.

The second aim is to investigate ways to increase the post-fault performance and
mitigate the consequences of the different faults that are considered.

1.2 Scope
A simulink model is developed for the first aim and a test-rig, available at Aros-
electronics, is utilised for experimental tests. The performance of the machine
is evaluated through practical tests and simulations for single and multiple open
phase faults, and three post-fault control strategies are compared against the healthy
benchmark.

For the second aim, different control approaches are investigated and applied with
the focus of not having to add any additional components to the system. This ap-
proach is chosen because methods where fault-tolerance and redundancy is increased
through various hardware changes are quite common in literature. Non hardware-
based approaches are not as prevalent, thereby making it an interesting topic to
look in to. Not adding any components to the system also gives the opportunity of
quick after-market implementation by only updating the software, as well as keeping
down overall costs of the system.

Fault detection, which is a very important functionality of a fault-tolerant drive sys-

tem, is not specifically investigated in this thesis. Instead the faults are introduced
and presumed to be known. More specifically, it is assumed that failures occur in
the inverter switches or at the terminals, as well as that the considered faults only
affect the inverter legs that are specified. It is also assumed that a failure does
not generate another one and that it is possible to isolate any faulty converter leg.

2
 1. Introduction

Turn-to-turn faults are not specifically studied in this thesis.

The position of the rotor is assumed to be known in the simulations, and it is ob-
tained with an encoder in the bench tests. Sensorless operation is generally preferred
in reliability applications, but is excluded from this work.

Discretization, and therefore modulation, is disregarded in the simulations. The

motor is continuously controlled. Furthermore, the dynamic behaviour of the motor
is also disregarded, and focus is only on steady state behaviour.

3
1. Introduction

4
 2
Theory

PMSM design can vary greatly depending on its intended application. In the follow-
ing sections, based academic literature and for the purpose of this thesis, the most
relevant aspects of machine design will be discussed.

2.1 PMSM machine design

In permanent magnet synchronous electric machines, typically three phases are
coiled on the armature separated from each other an equal distance in space, 120°.
Once excited by a three-phase voltage source, whose phases are 120° separated in
time, a rotating magnetic field will be produced, and the speed at which it rotates
is the synchronous speed. The field in the rotor is produced by permanent magnets
and it rotates with the rotor. Since the magnetic poles are attracted to the rotating
field created in the armature, the rotor will also rotate at synchronous speed [7].

For a machine with a single pole pair, the machine will rotate once every electric
cycle. In the case of multiple pole pairs, each pole will cover 360 electrical degrees,
and therefore the machine will rotate once every Np electric cycles. From this con-
cept follows the relationship between electrical and mechanical position and speed
in radians per second, expressed as

θe = θm Np (2.1)
ωe = ωm Np (2.2)
where θ is angle, ω is speed, and the subindices m and e indicate mechanical and
electrical, respectively.

Reluctance represents the opposition to a magnetic flux, R = Fφ = Nφi , where R is

reluctance, F is the magnetomotive force, φ is the magnetic flux, N is the number
of turns in a coil and i is the current in the coil. If magnetic flux is proportional to
current in the coil, φ = Li, and current remains constant, it can be said that reluc-
tance is inversely proportional to inductance, L, which represents the ratio between
di
an induced voltage and the current causing it, V (t) = L dt .

Magnet location has significant influence on the reluctance of the path their mag-
netic field follows. The relative permeability of the magnets is similar to that of air

5
 2. Theory

and much lower than the relative permeability of iron cores. Therefore, permanent
magnets mounted on the surface of the rotor offer the lowest reluctance configura-
tion when compared with magnets mounted in the interior of the rotor. The former
configuration is considered “non-salient” and the variation of inductance with rotor
position is very small compared with the latter, known as “salient”[8].

Assuming magnetic linearity by disregarding saturation, the magnetic coenergy in

the motor can be expressed by
1
Wco = Li2 + iΨ, (2.3)
2
where the first term will be equal to the energy stored in the coupling field provided
by the inductance of the machine and the stator currents, and the second term rep-
resents the energy provided by the permanent magnets in the rotor. Differentiating
the previous expression with respect to the mechanical rotor angle and considering
(2.1) yields a general expression for the electromagnetic torque produced by the
motor !
1 dL(θe ) dψpm (θe )
Te = np i ·i+ . (2.4)
2 dθe dθe

Ideally, the back emf induced in the coils by the permanent magnets will have a si-
nusoidal distribution in space. However, magnet shape and their location ultimately
determine the shape of this waveform. Additionally, the magnet location leads to
a tendency of the rotor to align with the stator in directions where the reluctance
is minimized. This tendency is known as cogging torque and contributes to torque
ripple in the output of the machine. Careful consideration of magnet location, such
as skewing of the rotor considering the specific number of poles and slots in the
machine, will reduce such cogging torque and make the back emf waveform more
sinusoidal [9].

2.1.1 Windings
Motors with more than one 3-phase sets, called multi-stator machines, typically
adopt arrangements of such sets in two different ways: Interlaced subsystems and
segregated subsystems. The differences between arrangements will be explained be-
low, inspired from [10].

Interlaced windings
In this configuration of multi-stator machines, coils of one subsystem are next to
coils of another subsystem, as shown in figure 2.2. With this arrangement, coils of
one winding set are more coupled to other sets in the stator than with the segregated
arrangement. Currents in one set will have a stronger effect on currents in other

6
 2. Theory

sets. Additionally, phases of each set are evenly distributed in space, which provides
smoother operation of the machine under set failure, provided an appropriate con-
trol strategy is in place.

Segregated windings
Segregated windings are characterized by individual winding sets that are geomet-
rically separated from other sets in the stator. This provides reduced magnetic
coupling between winding sets when compared with interlaced windings, and hence
better magnetic isolation. Furthermore, coupling between phases within a winding
will not be symmetrical due to their physical position in the stator. The windings
configuration can be complicated further by introducing two different winding meth-
ods.

Figure 2.1: Non-

Figure 2.2: Interlaced
overlapping segregated
dual winding
winding

Figure 2.3: Distributed

winding

Distributed windings
In this winding arrangement, each coil is wound through slots that are not adjacent
to each other. If all slots are occupied, coils will overlap with each other and surround

7
 2. Theory

multiple teeth in the stator, as shown in figure 2.3. Depending on if the winding sets
are interlaced or segregated, sections of the stator can provide a stronger magnetic
coupling between coils of different phases or different subsystems when compared
with the non-overlapping arrangement.

Non-Overlapping concentrated winding

Non overlapping concentrated winding arrangements are those that wind each coil
in the stator around a single tooth. The advantage of this method is a low mutual
inductance between coils. Furthermore, coils in this arrangement that share slots
are known as double-layer windings and those that don’t share slots are referred to
as single-layer windings. The latter further reduces the magnetic coupling between
coils.

2.2 Modelling of PMSM machine

Modelling of the machine is based on [11], where computation results from a FEM
model are used as inputs into the equation based phase variable model instead of
using analytical calculations of these values. This approach was used to obtain ac-
curate values of the self and mutual inductance between phases and winding sets,
and also for the reproduction of the EMF generated by the permanent magnets.

The phase variable model of a PMSM proposed in [11, 12] is presented below com-
posed of three phase differential equations for voltage, flux linkage, and the torque
and motion equations:

d
Vabc = Riabc + ψabc (iabc , θ), (2.5)
dt

ψabc = ψsabc (iabc , θ) + ψrabc (θ), (2.6)

ψsabc (iabc , θ) = Labc (θ)iabc (2.7)

where Vabc are the terminal voltage vectors [Va Vb Vc ]T , Iabc are the current vectors
[ia ib ic ]T , ψabc (iabc , θ) the stator flux linkage, ψsabc (iabc , θ) the flux linkage produced
by stator currents, ψrabc (θ) the flux linkage produced by the rotor permanent mag-
nets [ψra (θ) ψrb (θ) ψrc (θ)]T , Rabc the resistance matrix, [ra rb rc ], and the inductance
matrix

 
Laa (θ) Lab (θ) Lac (θ)
 Lab (θ) Lbb (θ) Lbc (θ) 
Labc (θ) =  (2.8)


Lac (θ) Lcb (θ) Lcc (θ)

8
 2. Theory

The electromagnetic torque produced is

!
1 dLabc (θe ) dψrabc (θe )
Tm = np iTabc · iabc + , (2.9)
2 dθe dθe

and the motion equations are given as

dω
J = Tm − Bω − TL (2.10)
dt
dθ
=ω (2.11)
dt
where J is the inertia of the machine ω is the angular speed and equal with the
position derivative, B is the viscous damping coefficient, and TL is the external load
torque.

This model is further expanded to include two winding sets, resulting in a double
system coupled by an inductance matrix which includes coupling between phases
and winding sets:

 
La1a1 (θ) La1b1 (θ) La1c1 (θ) La1a2 (θ) La1b2 (θ) La1c2 (θ)
 Lb1a1 (θ) Lb1b1 (θ) Lb1c1 (θ) Lb1a2 (θ) Lb1b2 (θ) Lb1c2 (θ) 


 Lc1a1 (θ) Lc1b1 (θ) Lc1c1 (θ) Lc1a2 (θ) Lc1b2 (θ) Lc1c2 (θ) 
 
Labc1abc2 (θ) =   (2.12)
 a2a1 (θ) La2b1 (θ) La2c1 (θ) La2a2 (θ) La2b2 (θ) La2c2 (θ)
L

 Lb2a1 (θ) Lb2b1 (θ) Lb2c1 (θ) Lb2a2 (θ) Lb2b2 (θ) Lb2c2 (θ) 
 

Lc2a1 (θ) Lc2b1 (θ) Lc2c1 (θ) Lc2a2 (θ) Lc2b2 (θ) Lc2c2 (θ)

The torque calculation for the dual winding implementation results as

!
1 dLabc1abc2 (θe ) dψrabc1abc2 (θe )
Tm = np iTabc1abc2 · · iabc1abc2 + , (2.13)
2 dθe dθe

where iabc1abc2 is the current vector [ia1 ib1 ic1 ia2 ib2 ic2 ]T , and ψrabc1abc2 (θ) is the
rotor flux linkage vector.

2.3 Inverter theory

Voltage Source Converters are used to transform DC into AC and vice versa. A
single-phase full bridge converter is known to have 4 switches whilst a half-bridge,
also known as a half-wave converter, only requires two [13]. When the conversion
is DC to AC they are referred to as inverters and they are conveniently used to
provide power to electric motors since they are fully controllable voltage sources.
In simple terms, an inverter amplifies digital signals to create an AC power source

9
 2. Theory

from a DC source. There are different topologies used for this purpose, the most
common of which is the three-phase inverter. It employs one half-bridge converter
per leg, shown in figure 2.4.

Figure 2.4: Voltage-source inverter and AC machine.

Other topologies may include 4 switches (full bridge) per leg, and a neutral wire
connection with or without active switching. The output of the inverter is limited
to the available combinations of its switches while avoiding to close both switches in
a leg simultaneously. For the inverter shown in figure 2.4, 8 possible combinations
exist. Each of these will correspond to a voltage vector in the αβ coordinate system,
as shown in figure 2.5. These combinations can then be chosen using one of several
modulation techniques to produce the desired output.

Figure 2.5: Space-vector diagram.

10
 2. Theory

2.4 DQ-transformation
The electrical quantities describing either voltage or current in phase a, b or c, are
usually described by three cosine functions phase shifted by 120°. In a three phase
system, with a floating neutral point, the generalised quantities can be summed up
to zero as

xa + xb + xc = 0 (2.14)
if no zero sequence is assumed, which is the case if no neutral connection is present.
These three quantities can be expressed as the two equivalent α and β quantities by
applying the Clarke transformation
 
"2 xa
− − 13
1
" # #
xα 3 3
=K  xb  (2.15)
 
xβ 0 √13 − √13
xc
obtaining the two perpendicular axes shown in figure 2.6. These quantities are often
referred to as the real and imaginary axes in the complex plane. The normalised
result of transforming the abc quantities to αβ frame is shown in figure 2.7. The
K in (2.15) is the scaling constant which can be chosen as to obtain an Amplitude-
invariant, Power-invariant or RMS-invariant transformation depending on what is
more convenient in the specific case.

As figure 2.6 shows, the αβ frame as well as the standard abc frame, are stationary
reference frames. In order to obtain the DC quantities shown in the last section of
figure 2.7, the Park transformation is applied as
" # " #" #
xd cos θ sin θ xα
= (2.16)
xq − sin θ cos θ xβ
thereby moving from the stationary αβ frame to the rotating dq frame, where the
transformation angle θ is the electrical angle. The d and q notation stands for direct
and quadrature directions, where the d-axis is defined as the direction of rotor flux
and q-axis as leading the d-axis by 90°, as can be seen in figure 2.7. The Clark and
Park transformations are extensively used and thoroughly described in literature
such as in [14] and [15].

11
 2. Theory

Figure 2.6: Relationship between abc,

αβ and dq reference frames. Figure 2.7: Signals in different
reference frames using amplitude
invariant transformation.

2.4.1 Harmonics
In an ideal three phase system, the three sinusoidal waveforms has no distortions
and sum up perfectly at all times. This is however not the case in physical systems.
The harmonic distortions of the supply voltage can originate from the mains for grid
connected machines or from the inverters and other power electronic devices. The
machine itself also represents a nonlinear load caused by such things as air-gap field
distortion, asymmetric inductances and distorted back emf [14, 16, 17].

In a symmetric three phase system it can be shown through fourier series expansion
that the even-order harmonics disappear due to the half-wave symmetry. With the
three phases shifted by 120°, the odd numbered harmonics are all present where the
first is the fundamental component. Harmonics that are multiples of three sum to
zero due to symmetry and are called the zero sequence components. It can therefore
be concluded that the prevailing harmonics are of order 5, 7, 11, 13, 17, 19 and so
on, with a decrease in amplitude as the harmonic order increases [14].

When transforming the signals in to the dq frame using the Park and Clark trans-
forms the harmonic order also changes. As shown in [17], the 5th and 7th harmonic
in the abc frame is represented by the 6th harmonic in the dq frame, while the 9th
and 11th abc frame harmonics are the equivalent of the 12th dq harmonic. The
prevailing dq current harmonics continue as multiples of the 6th as, 6, 12, 18, 24
and higher. As is the case for the abc domain harmonics the higher orders can how-
ever usually be neglected, especially when the motor is operation in the high speed
region, as they then are filtered out by the mechanical system [18, 19]. The 5th and
7th order harmonics in abc frame have a big impact on the torque ripple but does

12
 2. Theory

not effect the average torque significantly [20].

If the system is unbalanced, as in the case of one phase not being connected, the
asymmetric inductance leads to a 2nd order harmonic being present in the dq axis
currents [21]. The presence of the 2nd and higher even-numbered harmonics in the
dq frame leads to unwanted harmonics in the abc frame [22]. Inaccurate estimation
of motor parameters, discussed in the next section, can also lead to 2nd order har-
monics in the dq frame [23].

2.5 Control theory

The transformation from the abc to dq quantities allow for a PI controller to be
implemented, as the three AC quantities are simplified to two DC quantities. Field
oriented control (FOC), also known as vector control, is used to control the current,
flux and voltage vectors. The control technique decomposes the stator currents as
two orthogonal vectors, one representing the magnetic flux and the other the torque
production of the motor. As described in section 2.4, the flux is oriented in the
d-current direction and the torque in the q-current direction [14].

A speed controller is used which compares the measured speed with the speed ref-
erence given. The error signal is used to calculate a torque reference limited by the
machines maximum current magnitude. The limited reference torque is then used to
calculate the d and q axis reference currents. Since the motor is salient, meaning that
the d-axis current also produces torque, a maximum torque per ampere (MTPA)
calculation is preformed producing a current angle input to the current reference
calculation. The reference currents are fed to the current controllers, which com-
pares them with the measured winding currents, and outputs the needed voltage
signal in order to control the speed of the motor [14].

Constant motor parameters are assummed, and the condition to achieve MPTA
operation is

∂Te 3
= np Is [ψm cos β + (Ld − Lq )Is cos 2β] = 0 (2.17)
∂β 2
where β is the current vector angle. The reference value of the angle can therefore
be obtained from (2.17) as
 q 
−ψm + 2 + 8(L − L )2 I 2
ψm d q s
β = cos−1   (2.18)
4(Ld − Lq )Is

If variations of motor parameters are considered, MTPA control is much more diffi-
cult to achieve, as explained in [24]. These would be Ld and Lq considering satura-
tion, and ψm considering temperature variation.

13
 2. Theory

2.5.1 IMC controller design

The PMSM stator voltages can be modelled in dq-frame as
disd
usd = Rs isd + Lsd − ωr Lsq isq (2.19)
dt
disq
usq = Rs isq + Lsq + ωr Lsd isd + ωr Ψpm (2.20)
dt
where Rs is the stator resistance, Ls the stator inductance, ωr the electrical rotor
speed, Ψpm the rotor permanent flux linkage and subscript d and q denotes respec-
tive axis. The ωr Lsq isq and ωr Lsd isd terms represent the cross coupling between the
d and q axis and ωr Ψpm the back emf [14].

By solving (2.19) and (2.20) for isd and isq and applying Laplace transformation the
electrical transfer functions of the PMSM are obtained. By applying the closed loop
current control design strategy, according to internal model control (IMC) described
in [14], a PI controller can be derived.

In figure 2.8 the simplest form of the closed loop system block diagram Gcl is shown
containing the controller Fc , and the machine model Gc . Here subscript c is used
for current controller, but the same method is used for the speed controller, then
denoted by subscript s.

Figure 2.8: Closed loop system block diagram, Gcl

By making the Gcl a first order low pass filter as

Fc Gc αc
Gcl = = , (2.21)
1 + Fc Gc αc + s

where αc is the bandwidth of the system, the proportional and the integrating
parameters of the controller are obtained. The relationship between the rise time
and the bandwith of the system is

ln(9)
τ= (2.22)
αc

where the rise time is defined as the time from 10% to 90% of the step response time.

14
 2. Theory

In order to improve the disturbance rejection of the controller active damping, anti-
windup and feed-forward of the disturbance is applied. The PI controller parameters
then becomes

kp = αc L̂sx , ki = αc kp , Rax = αc L̂sx − R̂s (2.23)

where Ra is the active damping term and L̂sx represents L̂sd or L̂sq giving the gains
for both of the respective current controllers. The hat signs over the variables in-
dicate estimated values used by the controller. Perfect estimation and disturbance
cancellation is presumed during the control design.

The plant model Gc is for the control design purpose considered ideal, later however,
the actual plant model it controls is derived in section 5.1 and 5.1.1 and implemented
i Simulink in section 5.1.2.

The speed controller is cascaded outside the inner current control loop, which is
considered infinitely fast for derivation purposes. Ideally the speed and current
control bandwidths should have a relation of

αc · 10 > αs (2.24)
in order for the speed controller not to be affected by the current controller. The
speed controller is derived using the same steps as for the current controller and
active damping and anti-windup is implemented in the same way.

15
2. Theory

16
 3
Fault tolerant drive systems

This section presents how fault tolerant machines are defined in available literature.
A review of the different possible fault that can occur in a PMSM and its accom-
panying power electronic components is performed. Different suggested methods of
fault-tolerant designs and post-fault control strategies are also presented.

Fault tolerant drive systems are defined in [25] as a system fulfilling the following
criteria:
• Partitioning and redundancy
• Isolation between units
• Fault detection and reporting
• Continued operation until the next service opportunity

In [26] it is elaborated on how the first two points can be achieved either by increas-
ing the number of machine units, increasing the number of phases, or implementing
a dual-winding. It is also emphasised that, depending on the various faults the ma-
chine is required to tolerate, different machine configurations can be utilised.

The meaning of fault-tolerance can further be divided, as in [4], into either a system
that is designed to reduce the number of occurring faults, or a system with post-fault
performance capabilities. In the case of post-fault capabilities two approaches can
be chosen. Either having redundant components where in case of fault, a healthy
component is available for the same task, or a configuration where the remaining
healthy components can be controlled to still be operational.

Depending on application requirements, fault tolerant machines can be designed

to have a continuous post-fault operation capability, or a more limited short-term
operation capability. Various combinations of controllable post-fault systems and
systems with redundant components can be combined to achieve desired level of
post-fault performance. The terminology to distinguish between a fault-tolerant
and a redundant systems is not standardised in literature and the approaches often
overlap [4]. The reliability of the drive is improved through redundant design and
fault-tolerant control in [3] and [27, 28].

17
 3. Fault tolerant drive systems

3.1 Possible faults

Post-fault functionality of an electric drive system depends on the type of fault that
has occurred and its severity, which in turn depends on the design of the drive
itself. This section describes possible faults in the machine and converter, and their
possible consequences, disregarding possibilities of a DC-bus failure.

3.1.1 Winding short-circuit Faults

Conductors inside a machine are wound several times around each stator tooth and,
depending on winding design, the wires from different phases can also be in con-
tact with each other and with the machine stator. Winding insulation failure can
therefore cause a short between different parts of the machine, as mentioned in [29].
When the insulation on these wires breaks down, a path for current to flow will be
created and its effects will depend on the location of the breakdown. Reasons for
winding insulation failure, such as thermal and electric stress, are also described in
[29].

One common location for the insulation to fail is between one or more turns around
a tooth short, creating what is known as a turn-to-turn short circuit fault. This type
of short can cause a current higher than rated to be induced inside the damaged
coil and excessive heat produced by this current will eventually cause degradation of
the insulation in the neighbouring turns. Even though this single turn-to-turn fault
is a small part of the winding it will eventually lead to more turns being shorted,
resulting in larger currents and heat in the stator. In [29], the author mentions this
as one of the most severe types of faults since they are difficult to detect when they
occur and impossible to remedy without replacing the winding since the currents
will be induced in the loop whenever the machine is rotating.

Other possible short circuit faults in the stator circuit are those between phases or
between a phase and ground. The former can only occur if coils or wires of different
phases are in contact, and the latter may cause damage to the mechanical structure
of the stator. Unbalanced faults are those that short a current path between two of
the phases, and a balanced fault shorts all three phases together. In both balanced
and unbalanced faults, currents are induced while the machine is rotating and are
accompanied by a breaking torque.

Balanced Short Circuit Fault

From [30], the induced short-circuit currents for a balanced fault is expressed as a
function of speed as

" # " #
Idsc 1 ω 2 Lq ψpm
=− 2 (3.1)
Iqsc Rs + ω 2 Ld Lq ωψpm Rs

18
 3. Fault tolerant drive systems

and the current magnitude as

q
(ω 2 Lq ψpm )2 + (ωψpm Rs )2
|Isc | = (3.2)
Rs2 + ω 2 Ld Lq

The breaking torque is expressed as

3 R2 + ω 2 L2
Tbrk = − npRs ψ 2 ω 2 s 2 q 2 (3.3)
2 (Rs + ω Ld Lq )
From these equations it is notable that the short circuit current increases with speed
and approaches ψpm /Ld , as indicated in [27, 30], while the maximum breaking torque
and the speed at which it occurs are a function of the saliency ratio [30]. This is
illustrated in figure 3.1

Figure 3.1: Behaviour of current and torque during a three phase short circuit.

In the case of a dual winding machine, operation under a three phase fault is the-
oretically possible if the machine is appropriately designed. This would be with a
large Ld in proportion to ψm . Furthermore, in a dual machine, a short between
windings is also possible and its likelihood depends on the winding arrangement.
Although some of the faults mentioned in this section might be tolerable, they can-
not be remedied once they occur unless the winding or faulty inverter leg is replaced.

3.1.2 Open phase faults

Open circuit faults are typically a result of physical damage to the machine, or to
the conductors between the machine and the converter. In the case of a single con-
ductor being disconnected, the remaining phases may still retain functionality if an
adequate control strategy is implemented. This fault can also occur if one or several
of the inverter phase legs is purposely isolated.

19
 3. Fault tolerant drive systems

In the case of a single open phase, phase A for example, there is only one remain-
ing path for the current and the only possible voltage is between terminals B and
C. This is under the assumption that there is no neutral point connection, which
means that the phase voltages loose any practical use. Vbc therefore can only change
in magnitude and direction but cannot rotate. As mentioned in [31], this configura-
tion has consequences on the possible torque production of the machine, since the
voltage vector is perpendicular to the flux at only two instants.

In the case of a three-phase disconnection, voltages dependent on speed will be in-

duced on the terminals of the winding and pose a risk if they are exposed. A drive
system where the machine and inverter are enclosed in the same case makes this
type of fault unlikely. If a three-phase disconnection or three-phase inverter failure
occurs in a dual winding machine, torque can still be produced from the healthy
winding.

20
 3. Fault tolerant drive systems

3.2 Fault mitigation approaches

3.2.1 Fault tolerant machines
The machine that is the focus of this report is a two segment-three phase winding
PMSM, correlating to the segregated winding described in section 2.1.1. This makes
it inherently fault-tolerant due to its double winding layout where one winding-set
can output torque if the other fails. Thereby three phase open-fault operation is
possible by design, albeit at a reduced torque output if rated current is applied.

In case of phase-open fault, the main method described in literature is to increase

the fault-tolerant abilities by utilising a neutral connection [22, 32]. This gives
an increased possibility
√ to control the remaining phases, and a maintained torque
output reduced by 1/ 3, when operated at rated current for the affected winding.
The√neutral connection needs to be dimensioned for a current magnitude increased
by 3 of nominal in this case [33]. In order to maintain √ the rated torque with
this configuration, the currents needs to be increased by 3 and the neutral by a
factor of 3, thereby requiring the hardware affected to be redimensioned accordingly.

Copper losses in the windings naturally increase when currents are increased. The
calculation for these losses, per phase, is
2
Pcu = Irms Rph (3.4)

where Irms and Rph are the effective phase current and phase resistance, respectively.

Stator winding turn-to-turn short-circuit faults are an especially severe type of fault
in permanent magnet machines due to the rotors continued excitation of the sta-
tor during maintained operation. Therefore winding arrangements that reduce the
probability of these faults to occur, as discussed in [34], and fast detection methods
mentioned in [35], are the common measures employed. Also by designing the ma-
chine with this type of faults in mind, configuring the rotor magnet flux linkage and
the inductance saliency ratio, it is possible to minimize the effects of these faults [27].

Finally, the machine can be designed to tolerate a 3-phase short circuit if the design
parameters of the rotor are taken into account. The maximum current in this
scenario approaches ψpm /Ld , as discussed in section 3.1.1, so a machine designed
with this taken into consideration can therefore limit the current to a desired value.

3.2.2 Fault tolerant Inverters

The most common approach to increase the fault tolerance of inverters are by adding
a redundant leg in the converter [36, 37, 38]. If a neutral connection is available the
two most common topologies are shown in figure 3.2 and 3.3. In the case when the
neutral is connected to the fourth leg, as in figure 3.2a, a rotating magnetic field can
be maintained by controlling the neutral point voltage. If the neutral is connected
to the dc mid point of the converter, as shown in 3.2b, the capacitors will have to

21
 3. Fault tolerant drive systems

be sized for the large neutral current. The excitation of remaining two phases can
be done by controlling the switches in the two remaining converter legs [33, 37].

Figure 3.2: Fault-tolerant converter topologies with neutral connection.

In [38] a fault tolerant topology without neutral connection is also discussed shown
in figure 3.3. In this case isolating thyristors are implemented to isolate the faulty
converter leg which then is replaced by the redundant leg in the converter. In [28]
and [39] similar topologies are analysed as in [38] but instead fuses are used to isolate
the faulty inverter leg.

Figure 3.3: Fault-tolerant converter inverter with isolating thyristors and fuses.

If a very high level of fault tolerance and post fault capability is needed, every phase

22
 3. Fault tolerant drive systems

can be connected to its own H-bridge, as discussed in [40]. The system described is
a 6-phase modular permanent magnet machine where every H-bridge is connected
to its own power source. If one of the phases experiences a fault, it is simply discon-
nected and the machine keeps operating as a 5-phase machine. Different types of
converters for dual-winding motors are covered in [41, 42, 43]. These however em-
ploy a design where one main board supplies both winding sets. It is also possible
to run the machine with two 6-phase converters, but since this report focuses on a
layout with 3-phase converters, this is not further elaborated on.

It is important to keep in mind the added cost of any fault-tolerant topology, as

it may add to the over all system cost. As it is concluded in the cost survey in [44],
a certain approach and topology might be beneficial for one application but not add
much value to another depending on the required post-fault capability.

3.2.3 Post-fault control methods

A wide range of post-fault control strategies are available in literature depending
on the faults analysed and topologies used. Two distinguishable sub-groups can
be identified. One that focuses on maintaining the same performance capabilities
as during healthy operation. Usually reliant on additional hardware being imple-
mented, such as including neutral connections, redundant converter legs, fuses or
isolation switches, or independent H-bridge converters for every leg, as covered in
section 3.2.1 and section 3.2.2. Designs with full mechanical redundancy, where dual
windings are dimensioned to be able to deliver the full rated torque independently,
also falls under this category. The other sub-group focuses on remedial post-fault
strategies that mainly tries to mitigate the effects of the fault. Usually by imple-
menting control strategies that allow for a continued operation of the machine at a
reduced performance. Dependent on type of fault, detection and isolation measures
are usually also needed.

Maintained performance strategies

When adding a neutral connection, as mentioned in section 3.2.1, it is possible to
maintain the torque output if the current magnitudes are increased for open-phase
faults. This also requires the additional change to the control algorithms of chang-
ing the reference voltages by 30°. A field oriented controlled machine with space
vector modulation is purposed in [45] showing good performance where a neutral
is the only additional hardware. In [22] a similar method is stated to preform well
under open-loop and V/Hz control but have a poor performance during closed loop
control. Therefore a PI controller with an adaptive resonance part is proposed that
suppresses the 2nd and 4th harmonics. This increases the tracking capabilities of
the PI controller and thereby reducing torque and speed ripple.

Paper [12] also investigates open-phase faults. It utilises a neutral wire, combined
with a fourth inverter leg, and introduces a novel transformation matrix which al-
lows for the same PI controller to be used during both healthy and faulty operation.

23
 3. Fault tolerant drive systems

This allows the two remaining phases to output the same dq-axis control signals and
maintain the same torque output in post-fault as in healthy operation.

For dual-winding machines the same PI controllers are widely used as for con-
ventional 3-phase machines. Different methods of handling the mutual influence
between the two sets of windings and reducing torque and speed ripple are found
in literature. In [46] a sliding mode control combined with a normal PI controller
is analyzed. This method increases the dynamic response and makes the controller
insensitive to parameter variations and external loads. Fault-tolerance is however
not addressed specifically.

The two most common control strategies for dual-winding machines are in [47]
and [21] presented as Vector space decomposition(VSD) and two-individual cur-
rent control. VSD treats the machine as one 6-phase machine while the individual
current control treats it like two separate 3-phase machines. [47] concludes that
the individual current control has practical benefits compared to the more complex
VSD-control, but that it might have instability issues. However it concludes that if
the mutual coupling between the two sets of winding is not too strong, the perfor-
mance is equivalent of that of the VSD-control.

Remedial strategies
The simplest way to continue operation in case of a fault for a dual winding layout,
either in the winding or converter, is simply to switch one converter off and continue
operation at 50% with the healthy one. However, more performance is of course de-
sirable.

In paper [48] a dual winding machine with isolated neutral points is supplied by
two current source inverters. By using field oriented control, and in case of a open-
phase fault, a control scheme where the healthy winding generates the opposite
torque oscillation as the faulty one, the torque output can be kept constant.

The operation under single- and two-phase open-circuit faults of a dual-winding ma-
chine is investigated in [21]. The two Y-connected winding-sets neutral connections
are connected in the proposed topology. A VSD-control method is developed com-
bined with a resonant controller to suppress specific dq-current harmonics, arising
from the unbalanced operation. A feed forward method is also compared to the res-
onant controller when used to handle the post-fault 2nd order harmonics in the dq
frame. It is concluded that the resonant controller is the more robust and adaptable
solution and works well as a post-fault controller for the machine.

A method of fault severity estimation, to make it possible to continue operation

after a turn-to-turn fault, is developed in [49]. The focus is to minimize torque
ripple, and maximize the post-fault torque output, by re-configuring a FOC with
the proposed fault-estimation control.

24
 3. Fault tolerant drive systems

In [50] and [31] similar methods are developed for conventional 3-phase PMSMs.
Both look in to the possibility of continued operation after an open-phase fault with
the goal of not having any additional hardware added to the system. A Direct
Torque control is used in [31] while [50] focuses on FOC, but also compares the per-
formance of the developed post-fault algorithm when DTC or PTC(predictive torque
control) is applied. The approach is to utilize the four remaining switches in the
healthy inverter legs to control the two active and two zero voltage vectors available
in this operation mode. To compensate for lag in the stator current commutation, a
"pre-firing" method of the switches is employed to counteract breaking torque that is
the result of the abrupt switching action. Both papers report an average post-fault
torque output of over 75% of reference value. Two drawbacks of this method are the
inability to start the motor again once stopped as well as inability to operate in the
field-weakening region. The obvious advantages being that the only change to the
system is the modification of the control algorithm, making it very cost efficient.

25
3. Fault tolerant drive systems

26
 4
Test object and set-up

This chapter presents the prototype motor available, the inverters used and the lab
set-up used for verification of the model and evaluation of the machine.

4.1 Specifications of Motor

The motor prototype this thesis is based on is a two-segment three phase PMSM.
The stator windings are divided into two independent Y-connected sets where each
set has two parallel coils per phase. There are no neutral wires and the two neutral
points have no physical connection. The winding arrangement is shown in figure
4.1 where the first set is represented by the parallel branches A1//A2, B1//B2 and
C1//C2, while the second is represented by A3//A4, B3//B4 and C3//C4. The
corresponding wire connections for each coil are shown in figure 4.2.By connecting
A, B and C terminals together the machine can be operated as a conventional three
phase machine with corresponding motor parameters presented in table 4.2.

Figure 4.1: Cross section of ma-

chine. Figure 4.2: Winding connections.

The maximum continuous RMS current allowed per phase is 121.8A, when the ma-
chine is connected as a three phase machine, with four parallel coils per phase. In
the dual winding configuration, this value changes to 61.9A per phase, since there
is now only two parallel coils in each phase.

27
 4. Test object and set-up

From figure 4.1 it is clear that the two winding sets occupy one half of the stator
each, making it a segmented winding arrangement. The same figure shows the
12 slot stator design with a diameter of 145 millimetres and a rotor length of 45
millimetres. The rotor is skewed in order to reduce the torque ripple due to cogging,
as well as the ripple during loaded operation.

Parameter Measurement
Parameters of the machine are measured to verify FEA calculations and to use more
accurate values in the simulations. The measured values for the 3-phase connection
and dual 3-phase connection, which are the values used in the model, are presented
in table 4.1. Theoretical motor parameters from FEM simulations of the 3-phase
connected machine are presented in table 4.2, along with other relevant design pa-
rameters.

Phase resistance is measured from phase A to phases BC connected in parallel. The

measured value is equal to 3/2 of the actual, per phase value, and shows a small
difference when compared with the FEA values. The difference is expected since
the FEA does not include the connection wires.

Additional wire extensions are used on each phase to facilitate disconnection of one
of the phases from the inverter, and to reduce the total number of wires through
the current sensor on each inverter leg. The additional resistance is 0.45mΩ and is
included in the values shown in table 4.1.

Table 4.1: Measured motor parameters for dual winding connected machine

Motor Parameters
Parameters Symbol Value Value Unit
(1x3ph) (2x3ph)
Phase resistance rph 3.23 5.94 mΩ
D-axis inductance Ld 14.86 32.53 µH
Q-axis inductance Lq 27.53 56.83 µH

Inductance is measured in the same way as resistance, and the rotor is turned to
find the minimum and maximum values of inductance, which are 3/2 of the Ld and
Lq respectively. To obtain the permanent magnet flux linkage ψm , the induced volt-
age is measured at the terminals of the machine with the oscilloscope mentioned on
table 4.4. In this measurement, the machine is spun at a certain speed and the PM
flux linkage results as ψm = ωVe , where ωe is in radians per second. The obtained
value is the same for both machine connection configurations.

28
 4. Test object and set-up

Table 4.2: Theoretical FEM motor parameters for three phase connected machine

Motor Parameters
Parameters Symbol Value Unit
Rated mechanical speed Ωr 2800 rpm
Rated power Pnom 2.785 kW
Number of pole pairs np 4 -
Phase resistance rph 1.6 mΩ
D-axis inductance Ld 14 µH
Q-axis inductance Lq 28 µH
Permanent magnet flux linkage Ψm 0.0096 Wb
Rated voltage VLLpeak 24 V
Rated current Irmsnom 122 A
Max. current for 3 minutes Irmsmax 269 A
Rated torque Tnom 9.5 Nm
Max. torque for 3 minutes Tmax 21 Nm
Torque constant Kt 0.078 Nm/A
Inertia J 0.002 kgm2

4.2 Inverter
The inverters used are shown in figure 4.3 where two inverter are stacked on top of
each other in order to supply one winding each. The DC voltage input is rated at
24 volts and the current is limited by the inverters.

Figure 4.3: Dual inverter used in lab setup.

The encoder used is also visible in figure 4.3, as are some of the measurement
instruments.

4.3 Test bench set-up

The motor is mounted on a constant speed test setup to perform the experiments.
An additional machine is mechanically coupled to the test object and is controlled
by a separate converter. The additional machine can keep the speed constant in-

29
 4. Test object and set-up

dependently of the torque produced by the test object, since it is rated for higher
speed and power than the prototype. The ratings are shown in table 4.3.

Figure 4.4: Test-bench lab setup.

Devices used to measure voltages, currents and torque are all listed in table 4.4. In
figure 4.2 the wiring for the temperature sensors, placed in the end windings, are
also visible. The test object is liquid cooled and the connections can be seen at the
top of the motor in the same figure.

Table 4.3: Parameters of the additional motor.

Rating
Power 10.2 [kW]
Speed 4500 [RPM]
Torque 34.5 [Nm]

Table 4.4: Devices used to perform parameter measurements.

Measurement equipment
Oscilloscope Tektronix DPO3014
Current clamp AC Tektronix TCP404XL
Current clamp DC Tektronix TCP 0150
Torque transducer HBM T20WN
LCR Meter GW Instek LCR-6100

30
 5
Model Implementation

The derivation of the mathematical model and Simulink implementation of machine

model and controllers are described in this chapter.

5.1 Simulation model

The widely used dq-machine model, mentioned in section 2.5.1, is not suited for im-
plementing and simulating faults. It also does not take the non-ideal inductance and
back-emf in to account. The finite element-model models these effects accurately
but is instead time consuming. Therefore the physical three phase machine model,
suggested in [6] and [11], based on the equations introduced in section 2.2 is im-
plemented. By rewriting (2.5) on line-to-line form and applying Kirchhoff Current
Law

ia + ib + ic = 0 (5.1)
at the Wye-point connection, the line-to-line voltage equations are obtained as
" # " #" # " #
Vab r −rb ia d Ψa − Ψb
= a + (5.2)
Vbc rc rb + rc ib dt Ψb − Ψc

Then (2.6) and (2.7) are substituted in to (5.2) and the obtained dtd (Lab (θ)iab ) term
is expanded with the chain rule in order to solve for the state equations didta and
dib
dt
. For the purpose of utilising look-up tables in the simulation, the obtained flux
derivative term is rewritten as
dΨr dΨr dθ dΨr
= · = ω (5.3)
dt dθ dt dθ
to make it speed dependent instead of time dependent. On matrix form the obtained
equation, when solved for the state equations, can be expressed as
d d d
I = L−1 (V − RI − ω LI − ω ψ) (5.4)
dt dθ dθ
where the inductance matrix on line to line form becomes
" #
(Laa − Lac − Lba + Lbc ) (Lab − Lac − Lbb + Lbc )
L= (5.5)
(Lba − Lbc − Lca + Lcc ) (Lbb − Lbc − Lcb + Lcc )

31
 5. Model Implementation

Equation (5.4) combined with (2.11), expressing the electrical speed and position,
constitutes all quantities needed to express the electrical part of the machine model.
The state variable of ic is calculated from (5.1) as it is needed in the torque calcu-
lation in (2.9).

Finally (2.10), combined with (2.9), is used to express the mechanical part of the
machine model and calculate the electromagnetic torque.

5.1.1 Dual winding model

In order to model the dual winding layout of the motor in question, the same method
as in section 5.1 is used. By applying the inductance table introduced in (2.8) and
doing the same steps as in the previous section, but designating two versions of
(2.5) as Vabc1 and Vabc2 to account for both windings sets, a new version of (5.2) is
obtained as

Vab1 ra −rb 0 0 ia1 Ψa1 − Ψb1

      
V 
 bc1 
r r + r
 c b c 0 0  i 
  b1  d Ψ − Ψ 
 b1 c1 
=   + (5.6)
Vab2  0 0 ra2 −rb2  ia2  dt Ψa2 − Ψb2 
  

Vbc2 0 0 rc2 rb2 + rc2 ib2 Ψb2 − Ψc2

Continuing the steps with the expanded set of equations gives a new version of (5.4),
di
expressing the current state equations for the dual winding machine, now as dta1 ,
dib1 dia2 di
dt
, dt and dtb2 . The ic1 and ic2 currents are then calculated as in section 5.1.

The torque calculation is also done in the same way as in the previous section, but
with the vectors and matrices in (2.9) expanded to account for all phases from both
sets of windings, resulting in (2.13).

The expanded equations become quite lengthy but an example of the current state
equation for the dual winding machine model is shown in appendix A.1. The line-
to-line dual-winding inductance matrix is also included.

5.1.2 Simulink implementation

A three phase model is initially implemented, using the parameters from table 4.1,
in order to simulate and analyse control methods on a three phase connected ma-
chine as well as for verification purposes.

The electrical part of the three phase connected PMSM implemented in Simulink
is shown in figure 5.1. The inputs to the block are the line-to-line voltages and the

32
 5. Model Implementation

electrical speed and position. Since an encoder is used in the lab set-up the position
is presumed to be known and no position estimation is preformed. The outputs of
the block are the currents ia , ib and ic .

Subsystem 1 in figure 5.1 represents the inverted inductance matrix, dependent on

the electrical angle, in (5.4) and subsystem 2 the resistance matrix. Subsystem 3
represents the electric-speed dependent inductance matrix term, and finally subsys-
tem 4 the speed dependent magnetic flux term, representing the back-emf produced
in the machine. Subsystem 1, 3 and 4 utilises look-up tables acquired from the
FEM-model of the system. However, skewing of the rotor is not accounted for in
the FEM-model, therefore the back-emf look up table output is reduced by 10% to
account for this. This makes it possible to model the non-ideal behaviours that a
dq-model does not take in to account, but still allows for a fast simulation time. In
the right hand side of figure 5.1 it can be seen how the current state variables are
integrated and the ic current is calculated according to (5.1).

Figure 5.1: Line-to-line three phase electrical machine model.

The torque calculation is implemented in a corresponding fashion as in figure 5.1

based on (2.9). Lookup tables based on the FEM-model are used also in this case
in order to account for the non-idealities when calculating the inductance and flux
terms.

5.1.3 Dual model implementation

Implementing the dual winding machine is performed as in section 5.1.2 but with
the expanded equations from section 5.1.1. The Simulink layout is the same as
in figure 5.1 but expanded to account for the additional equations. The expanded

33
 5. Model Implementation

equations allow for simulations to include all the mutual couplings between coils in
(2.12). These couplings are visualised for one cycle in figure 5.2.

Figure 5.2: Self and mutual inductances within each winding and between each
other.

The dual winding parameters from table 4.1 are initially used and the expanded
(2.13) is used to calculate the torque. An example of the dual-winding Simulink
layout is shown in appendix A.2.

5.2 Fault implementation

To the right, just before the outputs in figure 5.1, three step functions and mul-
tiplications blocks are shown. They are used to implement the open phase faults
by multiplying the outputs with zero. In the case of one phase open the faults
are introduced at the currents zero crossing in order to reduce transient behaviour.
Additional step functions are implemented on the input voltage signals in order to
force relevant voltage terms to zero when simulating open-phase faults in the in-
verter. The other relevant terms in the current equations are forced to zero when
the phase current goes to zero, which can be seen in equation A.1 in the appendix.
To simulate the case of a three phase open fault, that is if a whole inverter looses
power, all three step functions are applied.

5.3 Control strategies and implementation

For the healthy machine operation, FOC with PI controllers designed according to
IMC, described in section 2.5 are used. For the dual winding connected machine,
the two-individual current control approach, discussed in section 3.2.3, is applied.

A speed controller, as described and designed in 2.5, is used in all cases and, as men-
tioned in section 5.1.2, the position is presumed to be known in the simulations as
an encoder is used in the lab set-up. The position of the rotor is therefore obtained

34
 5. Model Implementation

directly from integrating the electric speed and used as controller input.

To switch between the standard controller for healthy operation, and the different
post-fault control algorithms, the controllers are run in parallel in Simulink and step
functions are implemented, similarly to the fault implementation, in order to choose
which control signal to use. In other words, the healthy control is switched to the
post-fault control at the time of the fault, and therefore assuming perfect fault de-
tection.

5.3.1 Three phase-open fault control strategy

If an entire converter is lost, or if the best way to mitigate a fault is to shut one
converter completely off, the other keeps operating with its existing controller. No
additional control algorithms are used. By utilising the inherent redundancy of the
dual winding layout, 50% torque output can then theoretically be maintained by the
healthy winding at nominal current. The alternative is if the machine should keep
the same torque output as in the healthy case, a current of twice the magnitude is
expected, thus an oversized inverter is needed.

5.3.2 Single open-phase fault control strategies

Open-phase faults are the single most common type of faults in machine drive sys-
tems according to [12], constituting 38% of all fault occurrences. Therefore this
fault scenario is the main scenario investigated. Two methods of post-fault control
is implemented and evaluated in the case of a one-phase open fault.

Method 1 -Prefire
Under remedial strategies in section 3.2.3, the modified FOC algorithm proposed in
[50] is introduced. Since this method adheres to the aim of not having to add any
extra hardware, it is implemented and evaluated. The main difference between the
machine in [50] and the one used in this thesis being the dual-winding layout.

Assuming phase A is lost, only two active voltage vectors are now available ori-
ented in 90°and 270°. The two remaining phases are therefore the equivalent of one
remaining phase where the current can be either positive or negative. The only
quantities that can be controlled are then the sign and the magnitude. [50] suggest
the new torque relationship
√
Telec = is 3np ψpm cos (θelec ) (5.7)
to calculate the new current reference, that is implemented in the αβ frame of ref-
erence.

35
 5. Model Implementation

When solving for the current in (5.7) the cosine function causes the current to be
infinite when the angle is 90° or 270°. Therefore the new current reference has to
be limited to the magnitude of which ever component that has the lowest current
rating. The current reference then becomes as shown in figure 5.3.

Figure 5.3: Method 1 open-phase control method current reference.

The current limit, shaping the current at the maximum and minimum values in fig-
ure 5.3, causes the machine to generate less torque than its reference value. Hence,
as it is stated in [50], at low loads when the current magnitude is less limited, the
average electromagnetic torque matches the reference well. But at higher loads the
output is getting de-rated due to the increased current saturation.

Ideally the current should change polarity instantly, as it can be seen from the
reference in figure 5.3. However, due to the stator inductance, this is not the case.
Instead the current commutating gradually, causing the actual polarity change to
happen with a few electrical degrees delay. This delay in commutation causes a
negative torque production, reducing the performance of the already faulty machine.
In order to mitigate this problem the so called prefire of the switches is implemented.
This alters the theoretical switching angles of 90° and 270° to an "earlier" value to
allow for the current to commutate according to the reference. The required time
in which an earlier commutation has to be implemented is calculated as

2Ls is
tpre = (5.8)
VDC

from article [51] by the same authors as [50], both in which the method is thoroughly
explained.

36
 5. Model Implementation

Method 2 -Per phase controller

The second control method implemented to handle open-phase faults, is an in-house
developed control algorithm at Aros Electronics. The method decomposes every
phase into two quantities, representing the magnitude and phase of the signal re-
spectively, then controlling both quantities individually.

A sinusoidal current reference is required for the error signal calculation. The current
magnitude is calculated from the average torque over one period, expressed as

1 Z 2π
Tavg = is ψpm sin2 (θ)dθ (5.9)
2π 0
which is then multiplied with a sin-function of the rotor angle as

2Tavg
iref = sin (θelec + φ) (5.10)
ψpm np

where φ is the angle added in order to orient the vector in the corresponding direc-
tion of the voltage vector of which ever fault that has occurred. For phase A, φ is
for example set to 90° in order to create the VBC vector in figure 5.4.

Figure 5.4: Voltage vectors.

Analogous to how the electrical angle θ is used as transformation angle in the dq

transformation, θ is here used to transform the current error into DC values in order
to be able to implement PI controllers, as shown in figure 5.5. The Sine trans-
formation is chosen so that the D component controls the phase, and thereby the
currents zero crossing. Much to the same effect as in control method 1, resulting in
a reduction of the negative torque production. The Cosine function is then chosen
to transform the Q component controlling the amplitude.

37
 5. Model Implementation

Figure 5.5: Per-phase controller Simulink implementation.

After the PI controllers the same trigonometric functions with a 90° phase shift
are used to transform the quantities back into to AC values, where they are then
summed together to one signal. Finally the voltage signal is then transformed into
αβ frame quantities as shown in figure 5.6.

Figure 5.6: Per-phase controller transformation.

The PI controllers use the same gains as the controller for the healthy operation
and a current limiter is implemented after the current reference calculation. The
inputs to the integrator blocks in the healthy operation controller are used as initial
conditions for the integrator blocks in the fault-mode controllers to reduce transient
behaviours, as can be seen in figure 5.5.

38
 5. Model Implementation

5.3.3 Resonant controller

The resonant controllers in [22] and [21] aims to reduce post-fault torque and speed
ripple, thus increasing the post-fault performance. They also have the ability to
reduce possible harmonics during healthy conditions. Therefore the resonant con-
troller proposed in [22] and [21] is implemented in order to investigate its impact on
the performance both pre- and post-fault.

The ideal resonant controllers has the transfer function

Kr s
Gcres = (5.11)
s2 + (nωe )2

where Kr is the controller gain, ωe the fundamental electric frequency and n the har-
monic order of interest [21]. Due to its narrow bandwidth and infinite gain, imple-
menting the ideal resonant controller can lead to stability problems [52]. Therefore
a quasi-resonant controller is used with the transfer function

Kr s
Gcres = (5.12)
s2 + 2ωc s + (nωe )2

where the ωc is the cut off frequency of the controller. Including the ωc gives the
possibility to increase the bandwidth of the controller, thus increasing the controllers
frequency robustness [21].

Figure 5.7 and 5.8 show the bode plot transfer function for the quasi-resonant con-
troller at 1000rpm and n chosen to be the 6th harmonic. Kr is fixed to 200 in 5.7
while ωc is varied. In 5.8 the opposite is done where ωc is set to 10 and Kr is varied.
The figure shows that peak gain increases with larger Kr but at a fixed bandwidth
while ωc both affects peak gain and bandwidth. The high open loop gain at at
specified frequencies makes it possible for the controller to attenuate the selected
harmonics.

It is evident that Kr and ωc are important parameters for the performance of the
controller. By basing the parameter values on what is used in [21] and adjusting
them at different speeds and loads by trail and error, the values of Kr = 200 and
ωc = 10 are used. As [53] states, a straight forward tuning approach for resonant
controllers is barely available and a trial and error approach is often used. Since
tuning them thoroughly requires a thorough analysis of the dynamic behaviour of the
machine, it falls somewhat outside of the scope of this thesis as steady state values
are the main focus. Instead the parameter gains have been set rather restrictively,
and tested over a wide speed and load range of the motor, not to jeopardise the
stability of the system.

39
 5. Model Implementation

Figure 5.7: ωc = 1, 10, 100, Kr = 200.

Figure 5.8: Kr = 10, 200, 500, ωc = 10.

As is suggested in [34], the fundamental electric frequency, ωc is fed through a low-

pass filter and into the resonant controller in order to make it adaptable to its entire
speed range. As seen in figure 5.9, several resonance controllers can be put in par-
allel with the standard PI-controller in the dq frame.

Figure 5.10 shows the implementation of (5.12) in Simulink with all its parame-
ters. The n value, specifying what harmonic order the controller is operating on, is
squared in the Matlab code in order to satisfy (5.12). The derivation of the expres-
sion for the Simulink implementation is provided in appendix B.

40
 5. Model Implementation

Figure 5.10: Resonance controller Simulink

implementation.
Figure 5.9: PI controller with adap-
tive resonance controller.

The resulting bode plot of of the three parallel connected controllers is show in figure
5.11 where the gain at the three resonance frequencies 6, 12 and 18 can be seen. As
mentioned, more parallel elements could be added if desired.

Figure 5.11: Bode plot transfer function for resonance controller with 6th, 12th
and 18th harmonics chosen, Kr =200, ωc =10.

41
5. Model Implementation

42
 6
Results and analysis

This chapter describes the different practical tests performed in the lab and the
simulated cases. Measured values from the lab are compared against the simulated
results in order to verify the model behaviour.

6.1 Performed tests and simulations

The studied machine is rated to produce 9.5Nm at 2800rpm, when it is connected
as a single three phase machine. This rating is maintained when the winding is
connected as two 3-phase sets with a maximum allowed RMS current per phase of
61A, as described in chapter 4.

The bench set-up described in section 4.3 is used for measurements. The load motor
visible in figure 4.4 is set to the desired speed and the id and iq currents are applied
to the prototype machine with the same angle as calculated for MTPA in the sim-
ulations. Values for RMS current and average torque are measured alongside DC
voltage and current.

No changes to the controller are made when testing the healthy operation or the
three phase open cases. The position of the rotor is obtained with an encoder and
the information is shared with both inverters through opto-isolated communication.

The bandwidth of the speed and current controllers are in the simulation model
initially set to 200 and 2000rads/s respectively. All values presented are measured
or simulated at steady-state conditions unless otherwise stated. For all cases where
both lab-tests and simulations are performed the lowest and highest comparable
operating points are compared. In order to have a comparable mid range operating
point as well, the 40A rms point is arbitrarily chosen.

43
 6. Results and analysis

6.1.1 Healthy Operation of the dual winding machine

The healthy dual connected machine is tested at 500, 1500 and 2500rpm. The dq-
current set points, obtained from simulations when sweeping the torque from 1Nm
to the nominal torque output of 9.5Nm, are used as inputs to the test-bench. The
simulation results are shown with the experimental results in figure 6.1 together with
the torque-constant, Kt = Tnom /Irated representing the theoretical performance of
the machine at rated current.

(a) 500 rpm (b) 1500 rpm

(c) 2500 rpm (d) Minimum measurement error

Figure 6.1: Simulations and Lab-test result for the healthy dual winding machine.

Overall the simulated values align well with the theoretical design values of the
torque constant. The simulated values corresponds closely with the experimental
results at low range while the deviation increases as the load and speed becomes
higher at higher operation points.

One source of discrepancy between the simulation and the experimental values can
be attributed to the measurement uncertainty originating from the characteristic
tolerance of the transducer used to measure the torque. The device, mentioned in
table 4.4, can have an error of ±1Nm at zero torque and ±0.1Nm at its rated torque,

44
 6. Results and analysis

50Nm. The minimum measurement error is therefore 0.1Nm and is shown as the
thick line in figure 6.1d.

It can be noted that for the 500rpm case the model underestimates the torque output
slightly for operating points below 20A rms, while overestimating it when above 20A
rms. This however falls within the measurement error margin discussed in previous
paragraph. The discrepancy in percentage between the measured and simulated val-
ues are shown in table 6.1 where the tendency of larger errors with both increased
loads and speeds are visible.

The lower T /Irms -ratio of the measurements at higher loads and speeds can be ex-
plained by magnetic saturation in the motor which the simulation model does not
account for. At higher currents, for example, Lq is slightly decreased due to sat-
uration. If this is taken into consideration for the MTPA angle calculation, the
theoretically determined high power machine performance might behave more close
to its measured characteristic.

Table 6.1: RMS current and difference in torque output

Healthy dual winding operation

low range —————– mid range—————– high range————
Irms %dif f Irms %dif f Irms %dif f
500rpm 7 3.08 40 4.54 55 10.23
1500rpm 7 2.55 40 9.82 55 13.07
2500rpm 10 5.42 40 11.29 48 14.58

6.1.2 Three-Phase Open Circuit

It is possible that an inverter fails in such a way that the best way to continue
operation is to completely disconnect it from the motor. A similar case could occur
if the inverters are fed from separate DC-sources and one source fails. Similar tests
are therefore performed as the previous test cases at 1500rpm with only one inverter
supplying one of the windings. The same scenario is simulated in Simulink by
introducing the faults as explained in section 5.2. The results are shown in figure
6.2. In this case, Kt is calculated with Tnom = 4.75Nm.

45
 6. Results and analysis

Figure 6.2: 1500rpm, 3 phase open.

As figure 6.2 shows, when operating the machine on only one winding, 4.4Nm of
torque are produced at the nominal rms current of 61A. As the Kt line shows, the
theoretical value is half of that of nominal load at 4.75Nm. The simulation slightly
overshoots the Kt value but the high range discrepancy between simulation and
measurements are of the same order as for the healthy case, as shown in table 6.2.

Table 6.2: RMS current and difference in torque output

Three phase open operation

low range —————– mid range—————– high range————
Irms %dif f Irms %dif f Irms %dif f
1500rpm 8 14.56 40 9.34 61 11.56

The high error in the low range operation point in table 6.2 is likely due to mea-
surement error as discussed earlier. Due to the slightly different gradients of the
simulated and measured values that can be seen in figure 6.2, the lines overlap at
14A rms. Consequently, the error is 0% at this point, with a relationship between
the measured and simulated values analogous to earlier tests in section 6.1.1, at
higher loads.

A simulation is run at the 1.5Nm load since the T /Irms -ratio closely matches between
simulated and measured values in this range. A slight unbalance in the abc-currents
can been seen in figure 6.3, which can be attributed to the non ideal inductances
and back emf, which is in line with the findings in [16] and [21].

46
 6. Results and analysis

Figure 6.3: 1500rpm, 3-phase open phase currents of winding B at 1.5Nm load.

6.1.3 Comparison of tests

Figure 6.4 shows the measured results for the dual winding healthy operation (DH)
and the three phase open circuit (3PO) case at 1500rpm for one winding, on the
same plot. The behaviour is as expected at the for the 4Nm operating point, where
the DH-case requires 28A rms per phase while 3PO requires almost twice the amount
(55.6A). However, for higher operating points, the DH machine produces 7.75Nm
at 55.6A, while the 3PO produced 4Nm at the same current. The expectation is
that the DH machine could produce twice the torque as the 3PO case, if both cases
have the same current per phase. The lower result could be due to the simultane-
ous operation of the windings, where an increased saturation of the stator is likely
since close to rated current is flowing in all phases. Coupling between the windings
could also be a factor, since the separate control implementation does compensate it.

It is worthy to note that joule losses for the 3PO case will be twice as large as DH
case for the same torque, assuming exact doubling of the phase current after the
fault. The relationship is only 2x and not 4x since there are twice as many phases
in the DH case.

The torque-per-ampere ratio in the DH case is close to exactly double the ratio in
the 3PO case, meaning that the “effectiveness” of the current to produce torque is
reduced by half, as expected. However, since there are more losses, this configura-
tion is also less efficient.

Figure 6.5 shows the losses for the DH and 3PO case, at 1500rpm, calculated from
the difference between the measured input power and the mechanical power at the
shaft. The plot shows higher losses for the 3PO case than the DH case, except at
low torque. This can be due to increased measurement errors at low power, both
for the input power and torque.

47
 6. Results and analysis

Figure 6.4: Measured results for DH Figure 6.5: Losses for DH and 3PO
and 3PO cases at 1500rpm. cases.

6.2 Single-Phase Open Circuit

In this scenario, the torque production capabilities of a single winding is tested

while it undergoes a single phase open fault. Although magnetic coupling is limited
between the windings, the other winding is disconnected in this case to avoid any
coupling interaction and more accurately quantify the single winding’s capability
and torque contribution.

6.2.1 Lab verified open phase cases

Two versions of this case are tested in the lab, both conducted at 1500rpm. First,
a test with one phase-open with no post-fault control implemented is performed, to
compare against the post-fault control strategy. The inputs are, as in the healthy
case, the id and iq currents calculated for different loads. Secondly, the per-phase
controller introduced in section 5.3.2 is implemented and tested. In this case, the
input is the current magnitude for the post-fault controller as calculated by (5.10).
The same cases are again recreated in Simulink with the faults implemented as
described in section 5.2. The measured values are shown in figure 6.6 and 6.7, along
with the simulated values.

48
 6. Results and analysis

Figure 6.6: One winding with one phase Figure 6.7: One winding with one phase
open, dq-controller -lab vs sim. open, per-ph controller-lab vs sim.

As can be seen in figure 6.6, the torque output has in this case decreased to 2.64Nm
at rated rms current. Which can be compared with the 4.4Nm observed in the previ-
ous test shown in figure 6.2, giving a ∼41% reduction of output torque when loosing
one phase. A very similar result can be seen for the per-phase controller in figure 6.7.

It should be noted that, for the case shown in figure 6.6, contrary to the earlier cases
the model now underestimates the torque output. The characteristics of the error
is also reversed compared to previous cases with a larger discrepancy at low range
operation points. However, for the operation point at rated current, the simulation
is very close to the measured value, as shown in table 6.3. The deviations can, as
for the earlier cases, possibly originate from measurement errors. That the slope of
the measured values in figure 6.7 do not extrapolate downwards to the origin, as in
all other cases, might indicate that the measured values are slightly too high for low
range operation points.

The results from the per-phase controller test and simulation, seen in figure 6.7,
shows very similar behaviour to earlier tests. The non-linear behaviour of the mea-
sured values around 30A rms is most probably due to measurement error. The
differences between measured and simulated values, shown in table 6.3, are for this
test the smallest of all performed tests for both low, mid and high range operation
points. The percentage value at the 40A rms point is however slightly distorted due
to the measured value at this point. If the difference instead is calculated at 43A
rms, it becomes 6.4%.

Table 6.3: RMS current and difference in torque output at 1500rpm

Open phase operation, no post-fault controller and per-phase controller

low range —————– mid range—————– high range————
Irms %dif f Irms %dif f Irms %dif f
dq-controll 12.3 31.9 40 7.7 61 1.3
Per-phase 13 2 40 4.1 61 8.8

49
 6. Results and analysis

Figure 6.8 shows the measured and simulated torque at 1500rpm when a load step of
2Nm is applied. The average of both output signals are 2Nm. The measured signal
is the mechanical torque and is thereby filtered through the mechanical system while
the simulated signal is the electromagnetic torque. The discrepancy in amplitude is
therefore expected. The machine performs remarkably well during one-phase open
conditions when no post-fault controller is implemented. There is only a slight
negative electromagnetic torque production of less than 0.1Nm in the simulations.
When the per-phase controller is applied this is however reduced to zero after the
controller has settled at steady state conditions, as can be seen in figure 6.9.

Figure 6.8: Open-phase dq-controller, Figure 6.9: Per-ph controller, con-

measured and simulated torque ripple at trolling away the negative torque at
1500rpm, 2Nm. 1500rpm, 2Nm.

Figure 6.10 shows a comparison between the 3PO case and the single phase open
circuit cases. As seen, the torque-per-ampere is further diminished; more current is
needed to produce the same torque. Figure 6.11 is similar to 6.5 and shows losses
for the single phase open fault case, with the normal dq controller, are larger for a
given output torque, indicating reduced efficiency.

Figure 6.10: Results for 3PO and 1ph-

Figure 6.11: Losses of DH, 3PO and
Open case, with per-phase control and
1ph-dq cases.
with dq control.

50
 6. Results and analysis

6.2.2 Simulated open phase cases

The post-fault pre-fire method introduced in section 5.3.2 is not validated through
lab-tests, however it is implemented in the model and simulated at 1500rpm for
comparisons sake.

The simulated results from said method are very dependant on current-limits and
bandwidth of the current controller. As discussed in section 5.3.2, the current limit
is the deciding factor for the maximum current amplitude. Hence, if the current
limit is chosen to a magnitude of 61A, at high loads the current will theoretically
become completely square and result in a 61A rms current. The theoretically instant
current commutation also requires the controller to be very fast in order for the out-
put current to follow the reference current. At lower current controller bandwidths
the waveform become more and more distorted, therefore altering the rms-value.

To show the functionality and capabilities of the control method, a simulation where
αc = 10000rads/s at 1.8Nm load is shown in figure 6.12. At the end of the first Iref
period it can be seen how the Iout starts to commutate when the reference switches
polarity. The time it takes for the current to drop to zero causes a slight negative
torque production. The pre-fire algorithm is therefore engaged at 0.3s and the
new current reference is shown as the dashed line. At the two following instances
of commutation, the earlier switching action results in that the current changes
polarity at the angles 90° and 270° instead of lagging behind and the negative
torque is removed. A simulation at higher load is shown in figure 6.13 in order to
exaggerate the effect which is not clearly visible in figure 6.12. The two first cycles
shows the negative torque, and the following two cycles shows how it disappears
when the pre-firing is activated.

Figure 6.12: Current reference with Figure 6.13: Negative torque with and
and without prefire, torque and commu- without pre-fire during post-fault opera-
tating current. tion.

Since the maximum torque production is desired close to 90° and 270°, a current
limit of 61A is not optimal. To allow for the 1/cos(θ) shape of the current from (5.7),
in order to maximize the current and consequently the torque production closer to

51
 6. Results and analysis

the switching instances, a higher current limit performs better. The 61A current
limit is therefor multiplied by two for the nominal current case shown in table 6.4
and multiplied by three for the maximum operation point.

Ls in (5.8), used to calculate the pre-fire angle, is approximated to Lq and is is set to

the current limit. In order to perform a fair comparison with the other control meth-
ods, is should be continuously updated and the current limit should be optimised
to reduce the rms value for every load level. Since the result is not experimentally
validated this is however not done. In table 6.4 the pre-fire operation points are for
comparison sake simulated at 2000rad/s giving it a more favourable rms current.
The T /Irms -ratio is however nonetheless worse than for the two other cases.

Although the pre-fire method performs less favourable when comparing T /Irms -
ratios, it can push the performance up to the max rms current and produces the
highest torque output at 4.9Nm as seen in table 6.4. The drawbacks of not having
the possibility of starting once stopped, as noted in [50], is for the dual layout
analysed not an issue since the healthy winding can be used to perfome this task.
The max operation point shown table 6.4 is also simulated with current controller
bandwidth of 10000rads/s with a similar result.

6.2.3 Single-Phase Open summary

Table 6.4: Performance comparison for the different one- phase open cases

Performance of post-fault control strategies

Nominal current ———— Max operating point———————
Case Tload Irmsnom Tloadmax Irmsmax
No controller 2.65 Nm 61.7 A 3 Nm 78.2 A
Per phase controller 2.69 Nm 61.8 A 4.2 Nm 101 A
Pre-fire controller 2.3 Nm 61.9 A 4.9 Nm 134.6 A

As the comparison in table 6.4 and the plotted values in figures 6.6 and 6.7 sug-
gest, performing no action and implementing the per-phase controller in post-fault
conditions have marginal effects for low range operating points. However, when
loading the machine more, the per-phase controller performs better. The max op-
eration point in table 6.4 for the no-controller case is the last load-step before the
simulation can not keep the speed reference anymore. The per-phase controller can
actually keep the speed reference at loads above 5Nm, but with a very long settling
time and high transient currents. Therefore a current limit of 3 times 61A was cho-
sen for this operating point to keep the transient behaviour within reasonable values.

It can be concluded that the pre-fire controller needs a very fast current controller
and has a poor T /Irms -ratio at low operating points. It however have a very high load
operating capability under faulty conditions and very stable transient behaviour. If
high torque capabilities under intermittent load cycles is the requirement, it can be

52
 6. Results and analysis

a good choice.

The per-phase controller has a longer settling time when a fault occurs and the tran-
sient behaviour can be an issue. When comparing the T /Irms -ratio, the method how-
ever outperforms both other approaches over a reasonably wide operational range
in post-fault conditions. It can therefore be a good choice if a continuous operation
is the most important requirement even after a fault occurs.

6.3 Balanced Short circuit

Calculations from (3.1) and (3.3) are performed with parameters from table 4.1 for
the dual-winding connected machine, and the value of ψpm from table 4.2 was re-
duced by 10%, as mentioned in previous sections, due to the skewing of the rotor.
The results are plotted in figure 6.14 along with measurements of a balanced short
circuit on one of the windings, while the other is disconnected. The figure shows
that the maximum amplitude of the short circuit current at rated speed approaches
270A, as indicated by I = ψ/Ld , discussed in section 3.1.1, which would damage
the healthy winding.

Measurements coincide well with calculation of the current, and show that if a three
phase short circuit were to occur, the amplitude of the current induced in the af-
fected winding would become 85A at 145rpm. Furthermore, measurements differ
from the calculations for the breaking torque for an unknown reason. Nevertheless,
the maximum measured breaking torque becomes -6Nm at 400rpm and therefore the
unaffected winding would experience a severe increase of load, exceeding its rated
capability. From this analysis, it is clear that operation under a three phase short
circuit is very limited and can result in severe damage to the windings. The balanced
short circuit fault could not be simulated accurately with the model built in Matlab.

Figure 6.14: Behaviour of current and torque during a three phase short circuit.

53
 6. Results and analysis

6.4 Additional performed tests

In order to verify the reliability of the measured results, some of the experiments
where repeated. However, these additional measurements did not reproduce the
same results. The reason, at the moment of writing this thesis, is unknown. A
repetition of all measurements is proposed, in order to increase the confidence of the
results and hence prove or disprove repeatibility of the experiments.

(a) 500 rpm (b) 1500 rpm

(c) 2500 rpm (d) 1-ph controller

Figure 6.15: First and second round of experimental tests.

6.5 Resonance controller simulations

The resonance controller introduced in section 5.3.3 is implemented and its effect is
first investigated on the healthy dual machine performance, then the various faulty
cases. There are two main differences between the cases investigated in [21] and [22],
where resonant controllers are used, and the machine analysed in this thesis. The
first case is a dual-winding layout with a neutral wire present between the Y-points
and the second case is a standard three-phase machine with a neutral connection to
the inverter.

54
 6. Results and analysis

6.5.1 Healthy dual operation

The healthy dual machine is simulated at 500, 1500 and 2500rpm with and without
the resonance controller engaged at a 3Nm load to investigate its performance at
different speeds. 3Nm load is chosen since at this load the model closely resembles
the measured values, as figure 6.1 shows, where the load is shared equally between
the winding sets. A fast fourier transform(FFT) analysis is performed with the
Powergui tool in Simulink. The results are shown in table 6.5.

Table 6.5: Healthy dual operation at 3Nm load

Total harmonic distortion(THD) at different speeds

No resonant controller—– With resonant controller——
rpm THD THD
500 11.4 % 6.3 %
1500 12.6 % 2.4 %
2500 12.8 % 1.9 %

The resonance controllers are tuned as described in section 5.3.3 with, for this case,
the 6th, 12th and 18th harmonics are chosen to be suppressed. This results in
the suppression of the 5th and 7th abc current harmonics as shown in figure 6.16
and 6.17, where it can be noted that the 11th and 13th harmonics have totally
disappeared in the latter case.

Figure 6.16: abc-current harmonic Figure 6.17: abc-current harmonic con-

content with no resonant controller, tent with resonant controller engaged,
1500rpm, 3Nm. 1500rpm, 3Nm.

The system is in this case balanced, as can be seen from the absence of 3rd order
harmonics in 6.16. The results in table 6.5 and figures 6.16-6.18 are equal for all
currents of both windings.

The effect on the current over four periods can be seen in figure 6.18.

55
 6. Results and analysis

Figure 6.18: Stator current at 1500rpm, 3Nm load, with and without resonant
controller.

The torque ripple is, as expected, reduced while the average torque remains the
same when the abc-current harmonic content is reduced. Figure 6.19 shows how the
ripple decrease from ±0.7Nm to ±0.5Nm at this specific load and speed when the
resonance controllers are engaged.

Figure 6.19: Torque at 1500rpm, 3Nm load, with and without resonant controller.

6.5.2 Three-phase open operation

The same simulation as in figure 6.3 is performed again and a FFT analyse is done
on the dq-currents. As figure 6.20 shows, 2nd and 4th order harmonics are now
present in the q-current with a THD of 10.6%. The harmonics are even higher in
the d-current, at a THD of 86%, but as the main torque producer the q-current is
the one shown. Consequently, an extra parallel resonance controller is implemented
in the current controller and the 2nd, 4th, 6th and 12th harmonics are selected to be
suppressed. The resulting q-current harmonics are shown in figure 6.21, now with a
THD of 1.7%. The d-current THD is reduced to 6.5%.

56
 6. Results and analysis

Figure 6.20: q-current harmonic Figure 6.21: q-current harmonic con-

content with no resonant controller, tent with resonant controller, 1500rpm,
1500rpm, 1.5Nm, 3-phase open. 1.5Nm, 3-phase open.

The corresponding abc-currents are shown in figure 6.22 where the resonant con-
trollers are engaged at 0.35s. The THD of the currents prior to 0.35s is 11.7-14.3%
and is then reduced to 1.3-1.7%. The torque ripple is reduced from ±0.45Nm to
±0.26Nm.

Figure 6.22: 1500rpm, 3-phase open, abc-currents of winding B at 1.5Nm load

with resonant controllers applied at 0.35s.

As [22] states, the presence of the 2nd order harmonics in the system results in poor
reference tracking for the PI-controller which operates on DC-values. The improved
performance of the controller can be seen from observing the dq-current controller
error-signal shown in figure 6.23.

Figure 6.23: dq-current controller error signals, resonant controllers engaged at

0.35s.

57
 6. Results and analysis

6.5.3 Dual winding operation with one phase open

To investigate how an open-phase fault in winding B effects the operation when

winding A is operating normally, simulations at 5Nm was performed with both
windings sharing the load equally. At this low load, running the healthy winding at
nominal operation and reducing the output of the faulty to as low as possible should
be the most beneficial operating strategy, since this reduces the cross-coupling ef-
fects between the windings. The simulations are however performed this way in
order to investigate the harmonic content in winding A when winding B suffers an
open-phase fault. The 5Nm total load is chosen since at 2.5Nm load per winding,
the model matches well with measured results, as seen in figure 6.2 and 6.6.

The FFT analyse in figure 6.24 shows a very prominent 2nd order harmonic with a
THD of 26.7%. The same harmonics as in section 6.5.2 are chosen to be suppressed
by the resonance controllers.

Figure 6.24: q-current harmonic

Figure 6.25: q-current harmonic con-
content with no resonant controller,
tent with resonant controller, 1500rpm.
1500rpm.

As figure 6.25 shows, the resonance controllers can however not suppress the 2nd
order harmonic induced from the faulty winding in this case. Only the higher order
harmonics are suppressed which results in a THD of 25.9% and a slight torque ripple
reduction.

The unbalanced currents in winding A, caused by the faulty B winding, is shown in

figure 6.26

58
 6. Results and analysis

Figure 6.26: abc-current winding A and B, Open-phase in winding B at 0.2s,

resonant controllers in winding A engaged at 0.23s.

The PI-controller current-error is in this case not able to be reduced because of the
large disturbance from the faulty winding. It can therefore be concluded that the
neutral connections used in [21] and [22] are needed in order to use the resonant
controller as the actual post-fault controller.

6.6 Sustainability Aspects

The study of fault tolerance is especially relevant in the context of sustainability.
There are two main aspects to consider when considering a comparison of a machine
with extended reliability and one without special consideration to fault tolerance.
The first of these is the potential to reduce the amount of material that is scrapped
when a winding becomes damaged, and the second, is the extended productive time
of the machine.

Regarding the first mentioned aspect, a modular approach to the winding can be
valuable. This approach means that only the damaged part of the machine must
be replaced, instead of the entire winding. This also implies less time spent in
the re-manufacturing process, and hence less resources. Furthermore, a concen-
trated winding approach, where coils are wound around a single tooth, reduces the
likelihood of a fault to occur between to coils, or phases, when compared with a
distributed winding approach. With this approach the motor is less likely to need
a winding replacement, thereby reducing wasted material and remanufacturing re-
sources.

The second aspect is also benefited from the winding arrangements what where just
mentioned. The modular winding approach implies that the healthy windings can
still provide torque normally, provided there is no damage to their inverters, hence
prolonging the time before maintenance must be performed. With a concentrated
winding, the reduced fault probability will also prolong the productive time of the

59
 6. Results and analysis

machine.

The benefits of this approach must be weighted against the increase in cost and
additional materials used to produce the fault tolerant machine and drive. For
this reason, this thesis is concerned with methods to retain post-fault capability
of the drive with minimal investment. Applications where very robust reliability
is needed might call for considerable investment and additional materials, however,
this drawback can be balanced by the increased reliability itself. One example of this
is in aircraft fuel pumps, as discussed in [25]. A failure of a motor in such situation
would be catastrophic, and the costs of the failure will dramatically exceed any cost
related to improvement of the drive.

60
 7
Conclusion

Based on the findings from the previous chapter, the prototype machine described
in chapter 4 has a reasonable inherent fault tolerance to open circuit faults. In case
one of the windings or one of the inverters fails, the other can comfortably continue
operation up to 50% nominal torque without any modification whatsoever. The
remaining winding can also start the machine from zero speed, provided the initial
rotor position is known.

Furthermore, the usefulness of the machine in a 3PO state is dependent on the

application that it should be used for. In case of high load operation, an intermit-
tent drive cycle is more favourable. The reason for this is that the winding can be
operated above its nominal torque for brief periods of time, and therefore the only
drawback is the increased losses in the system.

Section 6.2 presents the capabilities of one of the windings in case of a single-phase
open circuit. If no change to the controller is implemented, the winding perform
surprisingly well at low torque loads. At loads above 3Nm, a post-fault strategy is
however essential to maintain a controlled output.

Two control strategies are evaluated and compared with the case where no post-fault
control is implemented. The pre-fire method and the per-phase controller. The per-
phase controller results in a very similar T /Irms -characteristic to when no post-fault
controller is implemented. It can however output much higher torque, with the
limits being how high transient currents that are acceptable and how long settling
time is allowed. The pre-fire method exhibits a lower T /Irms -ratio but a very sta-
ble transient performance and high torque output capability. The very fast current
controller needed might however prove problematic when implementing the method
in practice. The findings, summarised in section 6.2.3, suggests the per-phase con-
troller can be an option for continuous post-fault operation at lower torque loads,
and the pre-fire method an option for intermittent high load cycles.

Both of these cases, 3PO and single phase open circuit, are less effective at producing
torque within the machine. This is clear from the torque-per-ampere ratio, which
indicates higher current magnitudes are necessary to produce the same torque, since
less coils are available to produce said torque. This leads to a reduction in efficiency,
since higher currents cause higher joule losses.

The implementation of a resonance controller is investigated in section 6.5. It is

61
 7. Conclusion

found that harmonic content and torque ripple is effectively reduced for healthy-
dual as well as three phase-open operation. If the resonance-controller is to be used
as post-fault controller in the case of open-phase faults, it is however found that a
neutral connection between the windings Y-points is necessary.

Short circuit tolerance is practically nonexistent. In the case a short circuit oc-
curs, this particular machine would not be able to produce torque with the healthy
winding, due to the excessive breaking torque. If operation was forced, the currents
induced in the shorted winding would be more than 4 times the rated continuous
current.

The practical implementation of the above findings necessitates quick identification

of the fault and the implementation of the post-fault control algorithm. Further-
more, as mentioned in the scope of this thesis, the affected leg, or legs, would need
to be quickly isolated to prevent further damage to the inverter. In addition, in the
case of a single-phase open circuit, a master controller should be implemented which
could share the load between windings, taking into account their new, degraded, ca-
pability. An example of this would have the healthy winding operate the machine
entirely on its own for loads lower than 50%, and have the other winding comple-
ment the torque production only if the load surpasses the 50% level for continuous
operation. This would also reduce the torque ripple as well as the harmonic content
in the healthy winding.

7.1 Future work

The results show modeling of the motor in this manner performed in this thesis
provides results that match reality with reasonable accuracy. The method, how-
ever, always utilizes an inductance matrix obtained from only one operating point.
Inductance, however, is dependent on the currents in the stator; large currents satu-
rate the core. Therefore, the model can be improved by implementing an additional
dimension to the lookup table used to obtain the inductance data where their depen-
dence on current is accounted for. Furthermore, it could be interesting and useful
to improve the model such that short circuit faults can be accurately simulated,
including the breaking torque.

Regarding the control techniques, a method for a multi winding machine where the
a decoupling of the windings is discussed in [54, 55]. This method was ignored due
to the relatively low magnetic coupling between the windings in the segregated con-
figuration. However, it is evident from these reports that oscillations in torque and
speed can be minimized if this is accounted for.

The winding arrangement in the machine is such that connections to the individual
coils are available. This indicates the possibility to connect each of the windings
to a 6-leg inverter. This method would half the current rating for each of the in-
verter legs and could provide more reliable operation in case of an open phase fault,

62
 7. Conclusion

leveraging the benefits of multi-phase machines [56], since the winding could then
be operated with 5 phases, instead of 2. Finally, the need of a position sensor has
proved as a drawback for reliability. Therefore, the performance of the machine with
the employment of an angle estimator, such as in [57], should be evaluated.

To improve the capabilities of this motor, the most obvious approach is to have a
neutral connection to the inverter on standby, along with active switching. This
configuration, as mentioned in chapter 3, can output close to nominal torque with
minimal hardware expenses. If wire ratings are increased, this configuration can
provide rated torque.

A more expensive but robust configuration is to have a phase leg on standby, ready to
replace the faulty leg itself. This method does not require resizing of the conductors,
or a change in control algorithm, but is more expensive due to the SCR needed to
isolate the faulty phase and connect the redundant one.

63
7. Conclusion

64
 Bibliography

[1] M. Ruba and L. Szabo. “Fault Tolerant Electrical Machines. State of the Art
and Future Directions”. In: Journal of Computer Science and Control Systems
(2008).
[2] E. A. Bhuiyan et al. “A survey on Fault Diagnosis and Fault Tolerant Method-
ologies for Permanent Magnet Synchronous Machines”. In: Inernational Jour-
nal of Automation and Computing 17 (2020), pp. 763–787. doi: https://doi.
org/10.1007/s11633-020-1250-3.
[3] B. A. Welchko et al. “Fault Tolerant Three-Phase AC Motor Drive Topologies:
A Comparison of Features, Cost, and Limitations”. In: IEEE Transactions on
Power Electronics 19.4 (2004), pp. 1108–1116. doi: 10 . 1109 / TPEL . 2004 .
830074.
[4] Christian Du-Bar. “Design and analysis of a fault-tolerant fractional slot PMSM
for a vehicle application”. In: (2016). doi: ISBN978-91-7597-467-5.
[5] Stephen Umans. Electric machinery. McGraw-Hill, 2002. isbn: 9780070530393.
[6] J. Åström and T. Thiringer. “Time efficient FEM modeling of a PMSM by
iteration between field calculations and state”. In: (2016), pp. 2112–2118. doi:
10.1109/ICELMACH.2016.7732814.
[7] Prabha Kundur. Power System Stability and Control. McGraw-Hill, Inc., 1994.
Chap. 3.1.
[8] Oskar Wallmark. “Control of Permanent-Magnet Synchronous Machines in
Automotive Applications”. In: (2006). doi: ISBN91-7291-846-2.
[9] R. Islam and A Fardoun. “Permanent-Magnet Synchronous Motor Magnet
Designs With Skewing for Torque Ripple and Cogging Torque Reduction”. In:
IEEE Transactions on Industry Applications 45.1 (2009). doi: 10.1109/TIA.
2008.2009653.
[10] M. Kozovsky and P. Blaha. “Double three-phase PMSM structures for fail
operational control”. In: (2019).
[11] O. A. Mohammed, S. Liu, and Z. Liu. “Physical modeling of PM synchronous
motors for integrated coupling with Machine drives”. In: IEEE Transactions
on Magnetics 41.5 (2005), pp. 1628–1631. doi: 10.1109/TMAG.2005.846122.
[12] O. A. Mohammed, S. Liu, and Z. Liu. “A Phase Variable PM Machine Model
for Integrated Motor Drive Systems”. In: 2004 35rk Annual IEEE Power Ele-
crronics Specialisrs Conference (2004).

65
 Bibliography

[13] Muhammad H. Rashid. Power Electronics Handbook. Elsevier, 2007. isbn: 0-

12-088479-8.
[14] L. Harnefors. Control of Variable-speed Drives. Applied Signal Processing and
Control, Department of Electronics, Mälardalen University, 2002. url: https:
//books.google.se/books?id=wyXXjwEACAAJ.
[15] Colm J. O’Rourke et al. “A Geometric Interpretation of Reference Frames and
Transformations: dq0, Clarke, and Park”. In: IEEE Transactions on Energy
Conversion 34.4 (2019), pp. 2070–2083. doi: 10.1109/TEC.2019.2941175.
[16] Myung Joon Nam et al. “Torque ripple reduction of an interior PM syn-
chronous motor for electronic power steering”. In: 2015 9th International
Conference on Power Electronics and ECCE Asia (ICPE-ECCE Asia). 2015,
pp. 2445–2450. doi: 10.1109/ICPE.2015.7168115.
[17] Zhenbao Pan et al. “Combined Resonant Controller and Two-Degree-of-Freedom
PID Controller for PMSLM Current Harmonics Suppression”. In: IEEE Trans-
actions on Industrial Electronics 65.9 (2018), pp. 7558–7568. doi: 10.1109/
TIE.2018.2793232.
[18] Suk-Hee Lee et al. “A novel control method for reducing torque ripple in
PMSM applied for Electric Power Steering”. In: 2008 International Conference
on Electrical Machines and Systems. 2008, pp. 3142–3145.
[19] Heonyoung Kim and Subhashish Bhattacharya. “A Novel Current Control
Strategy Based on Harmonic Voltage Injection for Power Losses Reduction
of PMSMs with Non-Sinusoidal Back-EMF”. In: 2019 IEEE Industry Appli-
cations Society Annual Meeting. 2019, pp. 1–6. doi: 10 . 1109 / IAS . 2019 .
8912372.
[20] Xiaoyuan Wang. “Analysis of Harmonic Current in Permanent Magnet Syn-
chronous Motor and Its Effect on Motor Torque”. In: Journal of Electromag-
netic Analysis and Application 4 (Jan. 2012), pp. 15–20. doi: 10.4236/jemaa.
2012.41003.
[21] Zhuohang Li et al. “Single- and Two-Phase Open-Circuit Fault Tolerant Con-
trol for Dual Three-Phase PM Motor Without Phase Shifting”. In: IEEE Ac-
cess 8 (2020), pp. 171945–171955. doi: 10.1109/ACCESS.2020.3024736.
[22] Chandana Jayampathi Gajanayake et al. “Fault tolerant control method to
improve the torque and speed response in PMSM drive with winding faults”.
In: 2011 IEEE Ninth International Conference on Power Electronics and Drive
Systems. 2011, pp. 956–961. doi: 10.1109/PEDS.2011.6147371.
[23] Yuliang Guo et al. “An Adaptive Torque Ripple Suppression Method of Three-
Phase PMSM During Single-Phase Open-Circuit Fault-Tolerant Operation”.
In: 2019 22nd International Conference on Electrical Machines and Systems
(ICEMS). 2019, pp. 1–5. doi: 10.1109/ICEMS.2019.8921740.
[24] Gaolin Wang, Guoqiang Zhang, and Dianguo Xu. Position Sensorless Control
Techniques for Permanent Magnet Synchronous Machine Drives. Springer,
2020. doi: https://doi.org/10.1007/978-981-15-0050-3.

66
 Bibliography

[25] J. A. Haylock et al. “Operation of a fault tolerant PM drive for an aerospace

fuel pump application”. In: (1997), pp. 133–137. doi: 10.1049/cp:19971053.
[26] M. O. E. Aboelhassan et al. “A fault-tolerant control scheme for a dual flux-
switching permanent magnet motor drive”. In: (2011), pp. 1–6. doi: 10.1109/
ICEMS.2011.6073481.
[27] B.A. Welchko et al. “IPM synchronous machine drive response to symmetri-
cal and asymmetrical short circuit faults”. In: IEEE Transactions on Energy
Conversion 18.2 (2003), pp. 291–298. doi: 10.1109/TEC.2003.811746.
[28] S. Bolognani, M. Zordan, and M. Zigliotto. “Experimental fault-tolerant con-
trol of a PMSM drive”. In: IEEE Transactions on Industrial Electronics 47.5
(2000), pp. 1134–1141. doi: 10.1109/41.873223.
[29] Joachim Härsjö. “Modeling and Analysis of PMSM with Turn-To-Turn Fault”.
In: (2016).
[30] N. Bianchi, M. Dai Pré, and S. Bolognani. “Design of a Fault-Tolerant IPM
Motor for Electric Power Steering”. In: IEEE TRANSACTIONS ON VEHIC-
ULAR TECHNOLOGY 55.4 (2006), pp. 1102–1111. doi: 10.1109/TVT.2006.
877716.
[31] et al. A. Kontarcek. “Operation of a PMSM after Single Open-Phase Fault”.
In: International Conference-Workshop Compatibility And Power Electronics
(2013), pp. 177–181. doi: 10.1109/CPE.2013.6601150.
[32] Amr Saleh, Ghada Ahmed Abdel Nada Sayed, and Mona Nagieb Eskander.
“Fault-Tolerant Control of Permanent Magnet Synchronous Motor Drive un-
der Open-Phase Fault”. In: Advances in Science, Technology and Engineering
Systems Journal 6.5 (2020), pp. 455–463. doi: 10.25046/aj050654.
[33] N. Bianchi et al. “Innovative remedial strategies for inverter faults in IPM
synchronous motor drives”. In: IEEE Transactions on Energy Conversion 18.2
(2003), pp. 306–314. doi: 10.1109/TEC.2002.808334.
[34] P. Giangrande et al. “Braking Torque Compensation Strategy and Thermal
Behavior of a Dual Three-Phase Winding PMSM During Short-Circuit Fault”.
In: 2019 IEEE International Electric Machines Drives Conference (IEMDC).
2019, pp. 2245–2250. doi: 10.1109/IEMDC.2019.8785164.
[35] Arun Gandhi, Timothy Corrigan, and Leila Parsa. “Recent Advances in Mod-
eling and Online Detection of Stator Interturn Faults in Electrical Motors”.
In: IEEE Transactions on Industrial Electronics 58.5 (2011), pp. 1564–1575.
doi: 10.1109/TIE.2010.2089937.
[36] Xinxiu Zhou et al. “PMSM Open-Phase Fault-Tolerant Control Strategy Based
on Four-Leg Inverter”. In: IEEE Transactions on Power Electronics 35.3 (2020),
pp. 2799–2808. doi: 10.1109/TPEL.2019.2925823.
[37] Giuseppe Scarcella et al. “Fault-Tolerant Capability of Deadbeat-Direct Torque
and Flux Control for Three-Phase PMSM Drives”. In: IEEE Transactions on
Industry Applications 53.6 (2017), pp. 5496–5508. doi: 10.1109/TIA.2017.
2743070.

67
 Bibliography

[38] Mongi Moujahed et al. “Fault detection and fault-tolerant control of power
converter fed PMSM”. In: Electrical Engineering 98.1 (201), pp. 121–131. doi:
10.1007/s00202-015-0350-5.
[39] R.L. de Araujo Ribeiro et al. “Fault-tolerant voltage-fed PWM inverter AC
motor drive systems”. In: IEEE Transactions on Industrial Electronics 51.2
(2004), pp. 439–446. doi: 10.1109/TIE.2004.825284.
[40] Xuefeng Jiang et al. “Short-Circuit Fault-Tolerant Operation of Dual-Winding
Permanent-Magnet Motor Under the Four-Quadrant Condition”. In: IEEE
Transactions on Industrial Electronics 66.9 (2019), pp. 6789–6798. doi: 10.
1109/TIE.2018.2878131.
[41] Guodong Feng et al. “Open-Phase Fault Modeling and Optimized Fault-Tolerant
Control of Dual Three-Phase Permanent Magnet Synchronous Machines”. In:
IEEE Transactions on Power Electronics 34.11 (2019), pp. 11116–11127. doi:
10.1109/TPEL.2019.2900599.
[42] Y. Zhao and T.A. Lipo. “Space vector PWM control of dual three phase in-
duction machine using vector space decomposition”. In: Proceedings of 1994
IEEE Industry Applications Society Annual Meeting. Vol. 1. 1994, 742–749
vol.1. doi: 10.1109/IAS.1994.345429.
[43] Xiaolin KUANG et al. “Research on a six-phase permanent magnet syn-
chronous motor system at dual-redundant and fault tolerant modes in avi-
ation application”. In: Chinese Journal of Aeronautics 30.4 (2017), pp. 1548–
1560. issn: 1000-9361. doi: https : / / doi . org / 10 . 1016 / j . cja . 2017 .
05.001. url: https://www.sciencedirect.com/science/article/pii/
S1000936117301048.
[44] B.A. Welchko et al. “Fault tolerant three-phase AC motor drive topologies; a
comparison of features, cost, and limitations”. In: IEEE International Electric
Machines and Drives Conference, 2003. IEMDC’03. Vol. 1. 2003, 539–546
vol.1. doi: 10.1109/IEMDC.2003.1211315.
[45] Yonghui Liao. “Analysis of fault conditions in permanent magnet in-wheel
motors”. In: Master Thesis KTH (2011). doi: diva2 : 470634. url: http :
//kth.diva- portal.org/smash/record.jsf?pid=diva2%5C%3A470634&
dswid=9967.
[46] Ge Qi, Zhen Wang, and Bing-ying Zhao. “Sliding mode control strategy for
dual three-phase 180° electrical degrees permanent magnet synchronous mo-
tor”. In: 2017 IEEE 2nd Information Technology, Networking, Electronic and
Automation Control Conference (ITNEC). 2017, pp. 818–822. doi: 10.1109/
ITNEC.2017.8284848.
[47] Yashan Hu, Z. Q. Zhu, and Milijana Odavic. “Comparison of Two-Individual
Current Control and Vector Space Decomposition Control for Dual Three-
Phase PMSM”. In: IEEE Transactions on Industry Applications 53.5 (2017),
pp. 4483–4492. doi: 10.1109/TIA.2017.2703682.

68
 Bibliography

[48] Pengcheng Liu et al. “Remedial Strategies of Cascaded CSIs-fed Dual Three-
phase PMSM Drives under One-phase Open-circuit Faults”. In: 2019 IEEE
Energy Conversion Congress and Exposition (ECCE). 2019, pp. 6134–6139.
doi: 10.1109/ECCE.2019.8912699.
[49] Jiayuan Zhang, Wei Zhan, and Mehrdad Ehsani. “Fault-Tolerant Control of
PMSM With Inter-Turn Short-Circuit Fault”. In: IEEE Transactions on En-
ergy Conversion 34.4 (2019), pp. 2267–2275. doi: 10 . 1109 / TEC . 2019 .
2936225.
[50] Andraž Kontarček et al. “Cost-Effective Three-Phase PMSM Drive Tolerant
to Open-Phase Fault”. In: IEEE Transactions on Industrial Electronics 62.11
(2015), pp. 6708–6718. doi: 10.1109/TIE.2015.2437357.
[51] Andraž Kontarček et al. “Single Open-phase Fault Detection with Fault-Tolerant
Control of an Inverter-fed Permanent Magnet Synchronous Machine”. In: Au-
tomatika 55.4 (2014), pp. 474–486. doi: 10.7305/automatika.2014.12.624.
eprint: https://doi.org/10.7305/automatika.2014.12.624. url: https:
//doi.org/10.7305/automatika.2014.12.624.
[52] M. Liserre R. Teodorescu F. Blaabjerg and P.C. Loh. “Proportional-resonant
controllers and filters for grid-connected voltage-source converters”. In: IEE
Proceedings 153.5 (2006), pp. 750–762. doi: 10.1049/ip-epa:20060008. url:
https://digital-library.theiet.org/content/journals/10.1049/ip-
epa_20060008.
[53] Charles Lorenzini et al. “A model-free tuning method for proportional-multi-
resonant controllers”. In: arXiv e-prints, arXiv:2007.01942 (July 2020), arXiv:2007.01942.
arXiv: 2007.01942 [eess.SY].
[54] Samuli Kallio et al. “Decoupled d-q Model of Double-Star Interior-Permanent-
Magnet Synchronous Machines”. In: IEEE Transactions on Industrial Elec-
tronics 60.6 (2013), pp. 2486–2494. doi: 10.1109/TIE.2012.2216241.
[55] Jussi Karttunen et al. “Decoupled Vector Control Scheme for Dual Three-
Phase Permanent Magnet Synchronous Machines”. In: IEEE Transactions on
Industrial Electronics 61.5 (2014), pp. 2185–2196. doi: 10.1109/TIE.2013.
2270219.
[56] Mario J. Duran, Emil Levi, and Federico Barrero. “Multiphase Electric Drives:
Introduction”. In: Wiley Encyclopedia of Electrical and Electronics Engineer-
ing. American Cancer Society, pp. 1–26. isbn: 9780471346081. doi: https:
//doi.org/10.1002/047134608X.W8364.
[57] Oskar Wallmark, Lennart Harnefors, and Ola Carlson. “Control Algorithms
for a Fault-Tolerant PMSM Drive”. In: IEEE Transactions on Industrial Elec-
tronics 54.4 (2007), pp. 1973–1980. doi: 10.1109/TIE.2007.895076.

69
Bibliography

70
 A
Appendix 1

Appendix 1 concerns the derivations and implementation of the dual-winding sim-

ulation model.

A.1 Current equation

Expanded current state equation derived in section 5.1 and 5.1.1. Showing the
expression for current of phase a in winding 1. Where subscript 1 and 2 indicate
winding set one and two and double digit subscripts denotes the element in the
inductance matrix. Single letter subscripts a, b and c denotes phase quantities and
subscript ab and bc line-to-line quantities.
d
ia =
dt 1
L−1 d d d d
44 (Vab1 − Ra1 ia1 + Rb1 ib1 − ωr ( dθ L11 ia1 + dθ L12 ib1 + dθ L13 ia2 + dθ L14 ib2 )−
d
ωr dθ (ψma1 − ψmb1 ))+
−1 d d d d
L34 (Vbc1 − Rc1 ia1 − (Rb1 + Rc1 )ib1 − ωr ( dθ L21 ia1 + dθ L22 ib1 + dθ L23 ia2 + dθ L24 ib2 )−
d
ωr dθ (ψmb1 − ψmc1 ))+
L−1 d d d d
24 (Vab2 − Ra2 ia2 + Rb2 ib2 − ωr ( dθ L31 ia1 + dθ L32 ib1 + dθ L33 ia2 + dθ L44 ib2 )−
d
ωr dθ (ψma2 − ψmb2 ))−
L−1 d d d d
14 (Vbc2 − Rc2 ia2 − (Rb2 + Rc2 )ib2 − ωr ( dθ L41 ia1 + dθ L42 ib1 + dθ L43 ia2 + dθ L44 ib2 )−
d
ωr dθ (ψmb2 − ψmc2 ))(A.1)

Line to line inductance matrix of the healthy dual winding machine is expressed as

(La1a1 − La1c1 − Lb1a1 + Lb1c1 ) (La1b1 − La1c1 − Lb1b1 + Lb1c1 )


 (L
 a1b1 − Lb1c1 − Lc1a1 + Lc1c1 ) (Lb1b1 − Lb1c1 − Lc1b1 + Lc1c1 )
(La1a2 − La1b2 − La2c1 + Lb2c1 ) (La2b1 − Lb1b2 − La2c1 + Lb2c1 )


(La1b2 − Lb2c1 − Lc2a1 + Lc1c2 ) (Lb1b2 − Lb2c1 − Lc2b1 + Lc1c2 )

L=
(La1a2 − La2b1 − La1c2 + Lb1c2 )
(La1b2 − La1c2 − Lb1b2 + Lb1c2 )


(La2b1 − La2c1 − Lb1c2 + Lc1c2 )

(Lb1b2 − Lb2c1 − Lb1c2 + Lc1c2 ) 

(La2a2 − La2b2 − La2c2 + Lb2c2 )
(La2b2 − Lb2b2 − La2c2 + Lb2c2 )


(La2b2 − La2c2 − Lb2c2 + Lc2c2 )

(Lb2b2 − Lb2c2 − Lb2c2 + Lc2c2 )
(A.2)
where 1 and 2 in the subscripts denotes the two windings and a,b and c the respective
phases.

I
 A. Appendix 1

A.2 Dual winding Simulink model

Implementation of the electrical equations for the dual winding machine as discussed
in section 5.1.3.

Figure A.1: Line-to-line dual winding electrical machine model

II
 B
Appendix 2

B.1 Resonant controller derivation

Derivation of (5.12) for Simulink implementation as described in section 5.3.3.

out Kr s
Gc = = 2 (B.1)
in s + 2ωc s + (nωe )2

out(s2 + 2ωc s + (nωe )2 ) = inKr s (B.2)

out 2
(s + 2ωc s + (nωe )2 ) = in (B.3)
Kr s
out out out
s+ 2ωc + (nωe )2 = in (B.4)
Kr Kr Kr s
out out out
s = in − 2ωc − (nωe )2 (B.5)
Kr Kr Kr s
Kr out out
out = (in − 2ωc − (nωe )2 ) (B.6)
s Kr Kr s

III
DEPARTMENT OF ELECTRICAL ENGINEERING
CHALMERS UNIVERSITY OF TECHNOLOGY
Gothenburg, Sweden
www.chalmers.se