12

CHAPTER 2

LITERATURE REVIEW

2.1 INTRODUCTION

In this chapter, the technology status of switched reluctance

machines has been reviewed. Also, in-depth and exhaustive control aspects of
SRM are presented with particular emphasis on sensorless control.

The advancement in power electronics, control systems and the

computer aided design systems has revived the once dormant SRM and made
a viable alternative to conventional variable speed electric drives (Chan and
Chau 1997, Ali Emadi 2005, Chris Edrington and Babak Fahimi 2003). The
contributing factors for the development of high performance electric drives
are (i) the selection of power semiconductor devices, (ii) power electronic
converters, (iii) control and switching strategies and (iv) the system
integration through the efficient packaging of individual units. Equally
important factors are the optimization and the complexity of control
algorithms, switching and blocking losses that occur in power semiconductor
device with switching stresses (Chan and Lo 1987).

The attributes of robustness, simplified construction, fault tolerant

capability, high efficiency and high torque density have made SRM a
candidate to be deployed in electric drives.
 13

2.2 MODELING, DESIGN AND ANALYSIS OF SRM

Though the SRM was invented by Davidson in 1835, the

development in the power electronics and computing technology spurred a
number of researchers to renew the interest in the once dormant SRM
(Krishnan, 2001).

The reasons that justified the choice of SRM as a cost effective

machine for electric drive applications (Mahesh Krishnamurthy et al 2006)
are:

1. High instant power and power density;

2. High torque at low speeds for starting and climbing, as well as

high power at high speed for cruising;

3. Very wide speed range;

4. Fast torque response;

5. High efficiency over wide range of speed and torque;

6. High efficiency for regenerative braking;

7. High reliability and robustness.

The design objectives of SRM for effective electric drive system

are high torque, high power density and high efficiency. The design
methodologies for SRM applications can be categorized as

1. Analytical method;

2. Finite element method and

3. Optimization method.
 14

2.2.1 Analytical Method

The linear analytical method is based on examining the magnetic

circuit with current as independent parameters. In the case of nonlinear
method, the current dependent parameters are taken into account, (Miller and
McGilp 1990). Maximization of torque density, output power, efficiency,
(Yoichi Hayashi and Timothy Miller 1995), speed range and minimization of
torque ripple, temperature rise, acoustic noise and overall cost are some of the
design objectives and critical issues that must be addressed during the design
process (Anwar et al 2001, Chris Edrington et al 2007).

2.2.2 Finite Element Method

In finite element method, the design process starts with the manual
calculation of several SRM geometries with varying pole numbers and pole
dimensions. A two dimensional Finite Element Analysis (FEA) is then carried
out to obtain the nonlinear field solutions within the motor. The FEA field
solutions are used in the development of a nonlinear SRM model to search the
optimal control parameters that extends the constant power range of each
motor with maximum torque per ampere. Both the steady state and the
dynamic performance of each designed SRM are investigated for these
optimal control parameters and an iterative process for each design aids in the
determination of an appropriate SRM geometry for electric drive applications
(Chai et al 2006).

2.2.3 Optimization Method

The design optimization of SRM has been attempted commonly

under the following categories:

1. Shape modification through choice of geometrical parameters;

 15

2. Deterministic methods and

3. Soft computing methods.

1. Shape Modification through Choice of Geometrical Parameters

The design optimization through shape modification involves a

judicious choice of geometrical parameters and is the approach followed by
many researchers. Lovatt (2005) has significantly improved average torque
profile by incorporating special pole shapes and asymmetry of stator pole
with average torque as figure of merit. He has demonstrated that maximum
torque can be obtained by varying geometrical parameters such as stator pole-
arc per pole pitch, rotor pole-arc per pole pitch and air-gap length.

2. Deterministic Methods

Deterministic methods employed numerical optimization procedure

to achieve parameter optimization. Brisset and Brochet (1998) have employed
deterministic methods for the shape optimization of SRM with the average
torque as the parameter to be optimized. Faiz and Finch (1993 and 1997) have
concentrated on optimizing the various objective functions such as torque,
torque per ohmic loss, efficiency, stator back core width through the judicious
choice of tooth width/tooth pitch ratio, split ratio (rotor diameter/stator outer
core diameter), tooth pitch/air-gap ratio. The developed computer package has
been utilized in the initial sizing of the basic parameters and in the optimal
performance prediction.

3. Soft Computing Methods

Soft computing methods make use of AI techniques like fuzzy

logic, neural networks and evolutionary algorithms and their combination, to
arrive at an optimal design. In (Sahin et al 2000), the main objective is the
 16

torque ripple optimization, which used neural network as the main

computation module to formulate the objective function. The constrained
optimization problem is converted into an unconstrained form to be solved by
the methods of Lagrange multipliers and the penalty function method. In
(Mirzaeian et al 2002), the Genetic-Fuzzy Algorithm (GFA) (Wiak and
Pelikant 1989) has been used for optimal design of SRM with two objective
functions: high efficiency and low torque ripple, and GFA has showed that it
can be successfully applied to multi-objective optimization problem when all
the design variables are considered as discrete variables.

The comparison of the design methods can be summarised as

follows:

Analytical Method: As the phase current of SRM has widely varying

shapes, analytical approaches are restricted in their application as these
methods are more of qualitative than of quantitative.

Finite Element Method: With great strides in computer technology and with
the introduction of highly efficient, high speed processors, improvement in
computer memory technology, the finite element method based software suite
allows for the accurate and quick analysis of switched reluctance machines in
three dimensions.

Optimization Method: The optimization methods though render accurate

results are computationally intensive for rapid implementation.

The magnetic properties of the iron, the number of phases and the
number of poles per phase have a nonlinear effect on SRM's performance and
the dependence of flux linkages on the rotor position and current, as well as
the single source of excitation makes the SRM simulation more complex.
Krishnan (1988, 2001) describes that the drive system simulation requires the
 17

validation of drive system performance, which may include thermal and

vibration behaviour.

2.3 POWER CONVERTER TOPOLOGIES

For SRM drives, the torque is independent of excitation current

polarity and, it requires only one switch per phase. During commutation, the
stored energy has to be dissipated, so as to protect the semiconductor switches
from excess voltages across the windings (Debiprasad Panda et al 2007). The
management of this energy leads to a number of converter topologies like Np,
(Np+1), 1.5 Np and 2 Np, where Np is number of phases (Krishnan 2001,
Sayeed Mir 2000). These topologies have the advantages of lesser number of
switches and diodes that result in compactness, lower overall cost and higher
reliability of the drive (Syed Hossain et al 2003).

The C-dump with asymmetric bridge converter has a benefit of low

cost and simple structure. The R-dump converter consists of one transistor
and one diode per phase. The bifilar type is one with single transistor and
diode per phase but regenerating the stored magnetic energy to the source.
The asymmetric bridge converter consists of two switches on either side of
the one phase of winding and two diodes, one in anti-parallel with one switch
and winding in either side is employed in this research work, for the reasons
of its control flexibility, magnetic independence and fault tolerant capability
thus enhancing the system reliability (Sayeed Mir 2000).

2.4 CONTROL STRATEGIES

The drive control system design for an SRM can be classified based
on the performance requirement such as torque ripples and speed response.
This control can be carried out by current or torque or speed control or both
 18

(Kjaer et al 1996, Robert Heath et al 2000, Robert Inderka et al 2002).

Depending on the application, they may be sensor or sensorless control.

The SRM control requires an inner current loop and an outer speed
feedback control loop. The speed controller’s input is the speed error and its
output is the unmodified torque command (Krishnan 2001). Conventional PID
control cannot satisfy the requirements of high-performance applications.
Many modern control strategies such as Model-Referencing Adaptive Control
(MRAC), Self-Tuning Control (STC), Variable Structure Control (VSC),
fuzzy control, and Neural Network Control (NNC) have been proposed (Chan
et al 1987).

Microelectronic devices which include microprocessors,

microcontrollers, and Digital Signal Processors (DSPs) are essential to
implement the modern control strategies. The generally used microcontrollers
are 8096, 80196, and 80960. DSPs such as the TMS32030, TMS32040, and
i860 possess the capability of high-speed floating-point computation to
implement sophisticated control algorithms for high-performance electric
drive systems. For very high speed applications Field Programmable Gate
Array (FPGA), Application specific Integrated Circuit (ASIC) can be used
(Subramanian Vijayan et al 2007).

The control strategies of a SRM can be broadly classified as:

Current control: The current controllers designed for SRM are based on
(i) linearised model for reasonable performance (Giuseppe Buja et al 1991);
(ii) nonlinear current controller with feedback linearization and decoupling
without considering mutual inductance and (iii) nonlinear current controller
with feedback linearization and decoupling considering mutual inductance. In
the hysteresis switching based current controller, the current error is
 19

computed from the switching generated depending on its relationship to the

hysteresis current window (Paramasivam et al 2003).

Delta modulation control: Pulse Width Modulation (PWM) involves

modulation of duty cycle, to control the amount of power sent to a load. Using
delta modulation known as bang-bang control, the output signal is compared
to a reference signal. When the output signal reaches the limit, PWM signal
changes state (Chan and Lo 1987).

Speed control: For a closed loop speed control, speed can be processed
through a PI controller and a limiter to yield the torque command (Muniraj
and Chandrasekar 2007). An advanced digital speed control scheme based on
discrete P, PI and PID control algorithm, implemented using DSP which
yielded accurate speed control over a wide range of speeds and can perform at
different operating conditions had been proposed (Paramasivam et al 2004).

2.5 SRM SENSORLESS CONTROL TECHNIQUES

Rotor position sensing is an integral part of SRM control because of

the nature of reluctance torque production. In fact, excitation of the SRM
phases needs to be properly synchronized with the rotor position for effective
control of speed, torque and torque pulsation. A shaft position sensor is
usually employed to determine the rotor position. However, these discrete
position sensors not only add complexity and cost to the system but also tend
to reduce the reliability of the drive system when the SRM drive is to be
operated in extreme environmental conditions, such as high pressure,
temperature and humidity. Also, there are certain applications where the
ambient conditions (high vibrations) do not allow the usage of external
position sensors.
 20

The recent interest to eliminate the mechanical sensors for SRM is

mainly due to the following two reasons:

Reduction of cost: The mechanical sensor may be a significant part of the

overall system cost in cost-sensitive applications for fractional kW (or) hp
drives. The size of the mechanical sensor and the required number of leads
also play an important role in this kind of application, where the resolution
required in the rotor position is usually low.

Operation in a harsh environment: In applications where the motor is

operated in extreme environmental conditions, such as high pressure,
temperature, humidity and extremely high speeds, the use of a mechanical
sensor may lead to reliability problems. Usually, high resolution in rotor
position and robustness are necessary for this kind of application.

Other factors which motivate the elimination of mechanical sensors

are the reduction in the overall physical envelope and the weight of the motor
drive. Furthermore, the requirement for mechanical mounting of the rotor
position transducer complicates the drive design, increases the requirement
for maintenance and is a possible source of failure (Mehrad Ehsani and Babak
Fahimi 2002).

2.5.1 Open Loop Methods

When the machine is operated in the open loop, the motor

frequency and the dwell angle are controlled, but load torque fixes the torque
angle. Miller and Bass (1986) showed that the maximum pull-out torque is
proportional to the dwell angle i.e. at maximum dwell angle there is
maximum pull-out torque. Also it is shown that the efficiency is inversely
proportional to the dwell angle. Furthermore the two maxima (pull-out torque
and efficiency) are around the same torque angle. Hence, it was proposed that
 21

the dwell angle is adjusted to maintain a constant torque angle which

maximizes the efficiency at different load torque. When there is a load
transient the dwell angle is increased to compensate the load torque
maintaining the torque angle.

The feedback signal is the average DC link current which is used to

adjust the dwell angle i.e. with an increase in the DC link current, the dwell
angle must be increased in order to maintain the torque angle. To improve the
stability still further, the frequency oscillator is also adjusted when there is a
transient in the load torque. Oldenkamp (1995) made some improvements to
this method by adding a controller circuit which allows change in speed and
direction of rotation.

Vukosavic et al (1990) proposed a similar method. The main

difference is that the dwell angle is fixed and the commutation angle is
advanced. The feedback signal is the average energy returned to the power
supply, i.e. the current that flows through the power converter diodes.

The main advantages of these methods are: maximized efficiency

with improved stability around the open-loop control and low-cost
implementation. The main disadvantage is that it is poor in dynamic
performance because of the nature of the feedback signal; therefore it is not
applicable for variable-speed drives. It appears that it is applicable to motors
working at constant speed with approximately constant load. Zero speed is not
possible and the method does not provide a direct rotor position signal since it
assumes synchronism of the firing angles with the rotor position.

2.5.2 Unenergized Phase Methods

Most of the methods of this group measure the phase inductance in

an unenergised phase, which is used to estimate the rotor position. Generally,
 22

these methods use the injection of a low level chopping current waveform in
unexcited phases. The advantages of these methods are that they minimize the
effect of magnetic saturation (problems found in many sensorless methods
that use the active phase), because the position is estimated in the linear
region of the machine. The current rise or fall time of the probing pulses may
be measured in an unexcited phase for rotor position estimation. The use of a
low level regenerative current in an unexcited phase, and the use of mutual
induced voltage in an unexcited phase for indication of rotor position are
methods that belong to this group.

Active Probing

Acarnley et al (1985) suggested, using a low chopping current

waveform in a non-torque productive phase. The probing pulses are injected
at high frequency from the main power circuit. The current peak of the
probing pulses is relatively small (usually 10% of the rated current) so that
any negative produced torque can be neglected. The pulses are normally of
fixed duration at frequencies in the range of 4 to 20 kHz, depending on the
phase inductance. The phase commutation is found simply by comparing ipeak
with a threshold value. The commutation angle can be advanced or retarded
by reducing or increasing such a threshold value respectively. An alternative
method is to measure the inductance profile continuously for use as an index
in a look-up table. Alternatively, Myungi et al (1990) claim that the
flux-linkage at constant current has better sensitivity than the current at
constant flux-linkage. Therefore, it is proposed to use probing pulses of
constant peak current (variable frequency) in an unenergised phase. In this
case, the rotor position is estimated by observing the magnitude of the flux
linkage of the probing pulses.

MacMinn et al (1990) suggested estimating the rotor position in the

two unenergised phase simultaneously to avoid error in the estimated rotor
 23

position due to switching noises or mutual coupling. If two phases are

unenergized during the sampling period then it is suggested that the
extrapolated rotor position should be used instead of the instantaneous
estimated value. They also suggested monitoring the variation of the supply
voltage for more accurate rotor position estimation.

Van Sistine et al (1996) suggested comparing the peak current of

the probing pulses in two Unenergized phases. When both peak currents are
equal the next phase in the sequence is energized. Hedlund and Lundberg
(1989) proposed improvements in inductance measurements by introducing a
compensation factor which takes into account the mutual coupling effect.

McCann (1997) proposed to use probing pulses in the unenergised

coil phase to estimate the rotor position. McCann suggests having two sets of
coil per pole (four per phase) and therefore two asymmetric bridges per phase.
One set of coils is energized to drive the motor while the probing pulses are
injected in the other set of coils to estimate the rotor position. The main
advantage of this method is that the time for probing pulses is not limited at
high speeds because the sets of coils are independent. On the other hand, the
disadvantage is that, two asymmetric bridges per phase are required. An
effective and reliable means of generating commutation signals of an SR
motor based on inductance profile of its stator windings determined using
active probing technique is presented (Myungi 2007).

The main advantages are: four quadrant operation is possible, the

probing pulses are injected by the main converter itself and no extra circuit is
required, it is applicable as standstill, unique rotor position information can be
obtained if two phases are probed, the effect of back emf is minimized, and
the saturation effect is avoided due to the low current level of the probing
pulses.
 24

These methods possess the following disadvantages: (i) it is usually

difficult to implement them at high speed because current flows in a phase for
almost the whole electrical cycle, hence there is little time for probing pulses.
For this reason, the speed range is limited to medium values. (ii) The current
in the active phase induces voltage in the unenergised phases which strongly
distorts the probing pulses; therefore these methods are very sensitive to
mutual coupling. There is the necessity of knowing the specific phase
inductance as a function of rotor position. It is difficult to estimate the rotor
position close to the unaligned position due to the flatness of the phase
inductance in these regions. The injection of the probing pulses deteriorates
the overall drive efficiency.

Modulated Signal Injection

Ehsani et al (1992 and 1994) proposed to measure the phase

inductance in an unenergised phase by modulation techniques similar to those
used in communication system (frequency, amplitude and phase modulation).
Also, he suggested an alternative for the frequency injection method. A low
level sinusoidal voltage with fixed frequency and amplitude is applied to an
unexcited phase from an external oscillator via a resistance. The phase
displacement and the amplitude between voltage and current vary as a
function of the phase inductance. Therefore the inductance can be measured
by detecting the change in either phase displacement or the amplitude. Ehsani
claims that the phase and amplitude modulation schemes are better for
position estimation at low and high inductance respectively. Therefore, a
combination of both is suggested in order to obtain better rotor position
estimation and robustness over the whole electrical cycle.

Pasquesoone et al (2011), presents a method of accurate indirect

position estimation for switched reluctance machines suitable for starting and
continuous operation. The position estimation and control of the SRM is
 25

accomplished in terms of a few sectors within an electrical cycle. The sector

of machine operation is obtained by comparing the injected current amplitude
in an idle phase with two thresholds. Each sector is unique and identifies
which phase has to be used for sensing and which phase has to be used for
powering the SRM.

The main disadvantages of the modulation techniques are: (i) a

multiplexer is required for the external modulator/demodulator to the probing
phase, (ii) the sensing circuit needs to be isolated from the power converter,
(ii) the test signal is susceptible to error due to mutual coupling effects, (iv)
the specific phase inductance should be known, (v) an external
modulator/demodulator circuit is required, hence increasing the analogue
circuitry and (vi) the speed range is limited up to minimum speed where there
is sufficient zero current interval.

Mutually Induced Systems

Husain and Ehsani (1994) suggested splitting the induced voltage in

two values, one for the on time and the other for the off time of the PWM of
the active phase. That is, the sampling and hold of the induced voltage has to
be synchronized with the PWM switching frequency. The commutation
instants are obtained by comparing the induced voltage with a threshold.

Horst (1997) claims that the current flowing in the closed

unenergised phase path has a pronounced indentation, which is representative
of rotor position, due to parameter variation in the machine from phase to
phase such as air gap, phase inductance and phase resistance. A one-to-one
correspondence between magnetic status of the SRM and rotor position
removes the need for an explicit access to the rotor position (Babak Fahimi
et al 2005) The main disadvantage of this method is that at least an extra pair
of leads and one extra current sensor per phase are necessary.
 26

Jun Cai and Zhiquan Deng (2012) implemented this method in

which, First, based on the pulse injection technique, a new method called
phase current slope difference is proposed for identifying the full-cycle phase
inductance of each phase. Second, the phase inductances are considered as
vectors with fixed phase difference and the composed vector is rotated with
the varied rotor position. Combining the composed vector orthogonal
decomposition method and a new inductance sub-regional method, the rotor
position can be estimated.

The advantage of this method is that this method estimates the rotor
position by the direct measurement of an internal signal, which is available
without the injection of any diagnostic pulses.

The main disadvantage of this method seems to be limited to

systems using constant current regulation because the mutual voltage induced
depends on the level of the current in the excited phase and therefore the
current should be constant over the conduction period of the active phase (this
means that current profiling is not allowed). Furthermore the speed range is
limited up to base speed, where there is enough zero current period to observe
the induced voltage.

2.5.3 Energised Phase Methods

The methods that belong to this group are those which make use of
variables from the phase that is generating torque. The first publications
concerning indirect detection of rotor position were published by Acarnley
et al (1985).

Chopping Waveform

Acarnley et al (1985) proposed this method in which the motor is

controlled by hysteresis current regulation. The current is maintained
 27

approximately constant by a chopper in a hysteresis band. The current swings

around the required level at a rate dictated by incremental inductance. As the
incremental inductance is rotor-position dependent, the instantaneous rotor
position may be detected indirectly from the chopping characteristics. It can
be observed that both rise time and fall time are functions of inductance and
therefore the position can be detected assuming that the current swings in a
small band relative to the mean current level. However, there are other
variables, for instance the voltage drop across the phase resistance and the
back-EMF which vary significantly as function of speed and rotor position.
The uncertainty in the phase resistance and back-EMF complicates rotor
position detection using this principle. The use of rise time for detecting rotor
position is preferable instead of fall time (Panda and Amaratunga 1993), the
reason is that rise time depends on voltage unlike fall time and therefore the
back-EMF and voltage drop can be neglected at low speed. A further study of
this method was done by Panda and Amaratunga and the results showed the
difficulty of estimating the position with this method.

The main advantage of these methods is that they can be

implemented with simple electronics. The major problems of this method are:
(i) the back-EMF should be known which is difficult to measure while the
motor is running, (ii) the incremental inductance also depends upon the
current amplitude and introduces ambiguity at high current levels and (iii) the
assumption of current chopping limit these methods to low speeds.

Regenerative Current

This method, proposed by Reichard and Weber (1989), senses the

regeneration current in the energized phase. The phase is de-energized in
response to such current. In this method, the drive is controlled with the
 28

current control, when the current is above the current reference the phase is
turned off for a constant period of time (i.e. no hysteresis control).

Ibrahim Al-Bahadly (2008) method is based on estimating a

particular rotor position on a phase-by-phase basis and measuring the flux
linkage and current when the estimated position is reached. By comparing the
measured flux linkage with the prestored flux linkage corresponding to the
particular position for the measured current, the angular difference between
the estimated and particular positions can be calculated. Lei Shen et al (2011),
proposes a method which seeks to obtain the rotor position at standstill by
analyzing the time required for the charging current reaching its peak value in
the bootstrap circuit.

The disadvantage of this method is that (i) the energized phase is

normally turned off just after the aligned position, where considerable
negative torque may be generated and this may result in poor efficiency of the
drive and (ii) the turn-off angle cannot be advanced and the speed range is
limited to speed below the base speed where current regulation is possible.

Flux-Linkage

Hedlund and Lundberg (1991) proposed to identify a particular

inductance value per stroke which corresponds to a rotor position in the rising
inductance region. The signal which identifies the specific rotor position is
created from the current and the voltage across the phase winding (without the
motor drive being affected) is obtained directly by the measuring process.
According to this method, a known position is reached when the inductance L
of the phase winding reaches a predetermined value Lref. The condition L>Lref
is calculated by comparing the integrated value in the phase winding to the
product of the value of the actual phase current and the predetermined
inductance Lref. The method uses two extra compensation factors
 29

implemented in the microcontroller system. The current dependant factor

corrects the errors caused by the phase inductance which varies as a function
of the phase current. The second factor introduces a delay for phase
commutation as a function of speed.

The method of Lyons et al (1991) makes use of a set of

magnetization curves stored in a multi-dimensional table. Flux-linkage and
phase current measurements are made during motor operation in
predetermined sensing regions defined over an electrical cycle. For a given
flux-linkage and current, the rotor position of the energized phase is estimated
from the magnetization curves. In order to minimize the amount of stored
data, Lyons and MacMinn (1992) proposed to store only one magnetization
curve which represents a reference position between the unaligned and the
aligned positions instead of a multi-dimensional table of a range of positions.
The flux-linkage and current of the active phase is measured continuously
from the beginning of the phase commutation. The comparison between the
measured flux-linkage and the reference flux-linkage is done continuously
until the reference value is reached.

DiRenzo and Khan (1997) proposed to use an intermediate

magnetization curve for rotor position estimation. The difference is that the
intermediate magnetization curve is approximated based on the magnetization
curve at the aligned position. The reference position is reached when the
estimated flux linkage is larger than the one calculated.

Alternatively, Lyons et al (1992) proposed a lumped parameter

reluctance network model of the motor that includes a lumped network of
stator, rotor and air gap reluctance as a function of rotor angle taking into
account multi-phase saturation, leakage and mutual coupling effects. Flux-
linkage and current for each phase are measured simultaneously. The
reluctance terms of the model are determined from these measurements and
 30

are used to estimate the rotor position. It is important to note that this method
tends to include all non-linear effects of the machine but requires a large
amount of stored data and computation time.

Lyons and Preston (1996) also suggested using the flux-linkage

method to cover all the speed range. The flux-linkage and current of probing
pulses are measured in an unenergised phase at low speed, and their values in
the energized phase are used at high speed instead.

An example of a fuzzy logic algorithm applied to sensorless control

is presented in Cheok and Ertugul (1996). They claim high performance of
the method; however, it seems that it is computationally heavy.
A modification of this method to reduce the amount of computation time is
done by Xu and Bu (1997). Mese and Torrey (1997) applied neural networks
for rotor position estimation, and high performance of the drive is claimed;
however, it needs long off-line time for training the neural network. Both
fuzzy logic algorithm and neural network are based on magnetic
characteristics of the machine, (Christopher Hudson et al 2008, Estanislao
Echenique et al 2009, Zhong et al 2012).

The advantages offered by these methods are: (i) immunity to the

effect of saturation, (ii) applicable in a wide speed range, (iii) the mutual
effect can be taken into account, (iv) good accuracy and (v) four-quadrant
operation of the drive is possible. It seems that these methods are more
suitable for medium and high-speed applications due to the difficulty of
calculating flux-linkage at low speeds.

The main disadvantage of the method included in this group is the

classification of flux-linkage by integration of the phase voltage, because the
phase resistance varies strongly with temperature. Its effect is worse at low
speeds where integration errors can be large due to long integration periods.
 31

Another drawback is the necessity of a significant amount of pre-stored data

of the magnetization curves, which increases the amount of operations to be
computed. Also, the drive may be limited by the maximum speed of the DSP
at high speeds.

State Observers

This method was proposed by Lumsdaine and Lang (1990).

A mathematical model of the complete system, which includes the
mechanical load, is run simultaneously with the real system. In such a model,
the voltage is considered to be the input and current is considered to be the
output. The state variables are flux-linkage, speed and rotor position. The
estimated current is compared with the measured current and the error is used
as an input to the model to adjust the gains (matrix gain and vector gain).

A sliding mode observer for position estimation has been proposed

by Husain et al (1994). Further study of this method is made by Blaabjerg
et al (1996), who implemented the result in real time McCann and Husain
(1997) presented this method but based on flux-linkage observer rather than
current observer. It is concluded that the sliding mode observer is robust to
flux observer errors. McCann (1997) also proposed a hybrid position
estimation method using the sliding mode observer and a low resolution
position sensor (i.e. one position per energy conversion). The low resolution
position sensor is used to correct error in position estimation, particularly at
high speed where there is not enough time for observer to converge (Yang
and Kim 2000). Ahmed Khalil et al (2007), uses the technique which
combines two different methods to deliver high-resolution position
information over the wide speed range. At zero and low speeds, a voltage
pulse injection method is used to estimate the rotor position in all four
quadrants. For higher speeds, a sliding-mode-observer- based algorithm is
used and combined to work with the low-speed algorithm.
 32

The advantage of the sliding mode observer are: (i) high resolution
in detecting rotor position, (ii) high accuracy in estimating the rotor position,
(iii) applicability to whole speed range, and (iv) good performance in load
torque transients

The disadvantages of this method are: (i) real time implementation

of complex algorithm which require a high speed DSP, (ii) a significant
amount of stored data and (iii) additional extra circuitry which greatly
increases the cost.

Current Waveform

Sood et al (1995) proposed a method which compensates the

commutation angle as a function of the rate of change of the current
waveform in an energized phase. The current profile is sampled twice, one at
the middle of the commutation angle and the second just before the phase is
turned off. From this information di/dt is calculated and compared with the
desired value. The error indicates if the commutation angle is in phase and is
adjusted accordingly. The main disadvantages of this method are: the current
waveform information of the specific motor is needed for a number of speeds
before driving the motor, no specific rotor position is estimated but the motor
is stabilized by matching the desired current waveform. Conversely, the
advantages are: reasonable stability for speed transients, easy implementation
and four-quadrant operation.

In the method proposed by Lim (1996), the phase current waveform

is differentiated directly from each phase. The resulting signal di/dt is
amplified and compared to the reference value and the output of the
comparator is used the commutation signal. As an alternating option, the
signal di/dt is passed through a low-pass filter in order to introduce a delay.
This signal is amplified and compared with a second reference value and the
 33

output of the second comparator is “ORed” with the signal of the first
comparator to obtain a commutation signal. The main disadvantages of this
method are: it is not applicable at standstill or at low speed, it is inflexible in
advancing the commutation angle and the dwell angle is fixed. It seems that it
is applicable to constant speed operation and limited to two-quadrant
operation. The advantages of this method are its low cost and simple
implementation.

Current gradient and voltage magnitude sensorless methods were

proposed by Kjaer et al (1994). Both methods detect the specific rotor
position where the stator and rotor pole start to overlap. The difference is that
the current gradient method uses the change in di/dt when the motor is
operated in the PWM-voltage control or single pulse and the voltage
magnitude method uses the change in the average phase voltage when the
motor is operated in constant current regulation.

Laurent et al (1995) proposed the use of eddy current variation to

estimate the rotor position when the motor is operated in the PWM-voltage
control. The main disadvantage of this method is that the motor cannot be
operated in single pulse mode. This means that it is only suitable for low and
medium speed. Alternatively, Watkins (1998) proposed to use the rate of
change of the phase current at particular rotor position where the phase
current is forced to free-wheel in a zero voltage loop (i.e. only one switch is
ON). The advantage is that it is easy to implement and it works for the whole
speed range; however, the de-energized period of the phase has to be
extended, which may result in a decrease of the motor efficiency. The
disadvantages of this method are that stored data are required and it is not
applicable at standstill.
 34

2.6 FLUX LINKAGE – STATE OBSERVER BASED

ENERGISED PHASE METHOD

With the advent of high speed, efficient processors researchers have

attempted to combine the flux linkage method with the state observer to
exploit the advantages in these methods to realise a versatile sensorless SRM
drive. The research carried out in this avenue has been described briefly in
this section.

The rotor position estimation method using a current observer was

initially proposed by Acarnley et al (1995), which employed flux- linkage or
current to correct errors in rotor position estimation. This sensorless method
made full use of nonlinear magnetic characteristics through correlation of
flux-linkage, current and rotor position. Further the development of the
algorithm was made by Gallegos-Lopez et al (1999), who proposed a simpler
variation of the algorithm, but with non-loss in accuracy, leading to a
reduction in real-time computations. It was demonstrated that both flux-
linkage observer and current observer are equally sensitive to errors in the
predicted flux-linkage for position estimation.

De Araujo Porto Henriques et al (2011) presented a neurofuzzy

learning system used as a “virtual” speed sensor (inferential position sensor)
which utilised the voltage from each conducting phase and the reference
current signal as inputs to estimate the rotor speed. The main advantage of
this method is that instantaneous rotor position can be estimated with
minimum real time computation, because the algorithm makes use of a single
phase (best phase) for position estimation based on a simple model of
machine. On the other hand, the main disadvantage of the method is a
prediction of flux linkage which is calculated by integration of phase voltage,
and thus, the method is limited at high speed by the speed of the processor to
cope up with the mathematical operation and the requirement of pre-stored
 35

accurate data of magnetization curves. The estimation accuracy can be

enhanced if the current and voltage of probing pulses are used at low speed
with the phase current and the phase voltage at high speed.

2.7 INFERENCE FROM LITERATURE REVIEW

The magnetic characteristics of a switched reluctance machine is

usually represented by its magnetization curves, which show that the
relationship between current, flux-linkage and rotor position is unique but
strongly non-linear which complicates the rotor position estimation. While
most of the sensorless methods are based on the magnetic characteristics of
the Switched Reluctance Motor, there are many ways in which they can be
classified. A straightforward classification method is to consider whether the
method is based on (i) variables of the energized phases, (ii) variables of the
unenergised phases or (iii) use of other variables. Using this criterion, the
sensorless methods mainly fall into three major groups.

1. Open loop methods

x Dwell angle compensation.

x Commutation angle compensation.

2. Unenergized phase methods

x Active probing.

x Modulated signal injection.

x Mutually induced systems.

3. Energized phase methods

x Chopping waveform.

x Regenerative current.

x Flux-linkage.
 36

x State observers.

x Irregularities in stator/rotor poles.

x Current waveform.

The first group, also called stability torque control, does not
provide a direct position indication. The motor works in an open-loop as a
stepper motor from a variable frequency oscillator in a traditional
synchronous manner and adjustment is made to the dwell angle or
commutation angle by observing the dc link current in order to improve
stability in pull-out torque with maximum efficiency.

In the second group, different kinds of test signals are introduced or

measured during the time when a phase is normally unenergised (i.e. the
phases that are not generating torque), usually during the negative slope of the
phase inductance when the machine is motoring. The test signal needs to be of
low amplitude for the following reasons:

x To minimize the negative torque production;

x To avoid saturation effects;

x To minimize back-emf effects;

x To minimize the power rating of additional circuitry wherever

necessary.

The principle of these methods is to detect the phase inductance or

flux variation from the injected signal. These methods suffer at high speed
where there may not be enough time to inject the probing signal between the
extinction and turn-on of the phase, because the phase current waveform is
non-zero over the majority of the electrical cycle. Furthermore, the probing
 37

signal is sensitive to mutual interference from excitation currents in other

phases.

The third group uses the excitation current, voltage waveform or

derivatives (i.e. variables from the phase that is generating torque) to detect
the rotor position indirectly. An early method senses di/dt, which is a function
of the rotor position, when the motor is in hysteresis current regulation. Other
methods that belong to this group make use of the regenerative phase current
when the rotor crosses the aligned position. These methods mainly are for low
speed where the current can be regulated (i.e. below the motor base speed).
Alternately, there are methods in this group involving the detection of
flux-linkage or phase inductance when the rotor passes through a particular
threshold value which represents a specific rotor position. These methods may
provide more resolution if the number of reference values is increased instead
of having only one threshold value. An example of these methods is the
method based on flux-linkage observer or current observer, which gives
instantaneous rotor position estimation. Similarly, a complex lumped
parameter reluctance network model of the motor has been proposed, which
also estimates instantaneous rotor position. This group also includes state
observer methods, which are based on mathematical motor model that is run
in real time. The methods that use flux-linkage suffer at low speed, where it is
difficult to calculate rotor position accurately because of the variation in the
phase resistance and noise in the system. Conversely, methods that include a
full model of the motor may be limited at high speed because of the large
computational burden. That is, the speed of the DSP may limit the use of
these sensorless methods for high speed applications.

Other methods include irregularities in the stator poles, rotor poles

or both. In this way, the inductance profile is modified so that when the phase
is excited, it is easy to detect the irregularity in the phase current which
 38

indicates a specific rotor position. The last subgroup of this category uses the
shape of the current waveform to estimate the position. A method has been
proposed that identifies if the commutation angle is leading or lagging based
on the shape of the current waveform. The commutation angle is compensated
in order to have the desired current shape. However, this method does not
estimate any specific rotor position. The current gradient and voltage
magnitude sensorless methods are based on the current and voltage waveform
respectively. These methods identify one specific rotor position per stroke and
are simple to implement.

In general, in all the sensorless control technique works reported

hitherto, it can be noted, that, the latency in flux linkage error estimation of
windings is high due to the phase resistance variation of the windings.
Attempts by researchers to circumvent the above problems by incorporating
ANN / ANFIS techniques were partially successful.

Based on the literature survey of the sensorless control aspects of

SRM, it is envisaged in this research work to (i) simulate and study the flux
linkage combined state observer (energised phase) method of sensorless
control for 8/6 SRM prototype drive using FPAA processor and (ii) to
implement an ANN based flexible tuning algorithm in a FPAA processor for
sensorless control of the 8/6 SRM. The emphasis will be on the realisation of
a high performance, robust, less complex, fast acting but low cost sensorless
SRM drive.

2.8 SUMMARY

The literature related to sensorless control of SRM drive was

reviewed. From this review, the proposed research work to be implemented is
arrived at. This proposed implementation of sensorless control of SRM will
be described in subsequent chapters.