sensors

Article
Digital Twin Model of Electric Drives Empowered by EKF
Mohsen Ebadpour 1, * , Mohammad (Behdad) Jamshidi 2 , Jakub Talla 1 , Hamed Hashemi-Dezaki 1, *
and Zdeněk Peroutka 1

1 Research and Innovation Center for Electrical Engineering (RICE), Faculty of Electrical Engineering,
University of West Bohemia (UWB), 30100 Pilsen, Czech Republic
2 Faculty of Electrical Engineering, University of West Bohemia (UWB), 30100 Pilsen, Czech Republic
* Correspondence: [email protected] (M.E.); [email protected] (H.H.-D.)

Abstract: Digital twins, a product of new-generation information technology development, allows

the physical world to be transformed into a virtual digital space and provide technical support for
creating a Metaverse. A key factor in the success of Industry 4.0, the fourth industrial revolution, is
the integration of cyber–physical systems into machinery to enable connectivity. The digital twin
is a promising solution for addressing the challenges of digitally implementing models and smart
manufacturing, as it has been successfully applied for many different infrastructures. Using a digital
twin for future electric drive applications can help analyze the interaction and effects between the
fast-switching inverter and the electric machine, as well as the system’s overall behavior. In this
respect, this paper proposes using an Extended Kalman Filter (EKF) digital twin model to accurately
estimate the states of a speed sensorless rotor field-oriented controlled induction motor (IM) drive.
The accuracy of the state estimation using the EKF depends heavily on the input voltages, which are
typically supplied by the inverter. In contrast to previous research that used a low-precision ideal
inverter model, this study employs a high-performance EKF observer based on a practical model of
the inverter that takes into account the dead-time effects and voltage drops of switching devices. To
demonstrate the effectiveness of the EKF digital twinning on the IM drive system, simulations were
run using the MATLAB/Simulink software (R2022a), and results are compared with a set of actual
data coming from a 4 kW three-phase IM as a physical entity.

Keywords: sensorless control; Extended Kalman Filter (EKF); digital twin; induction motor (IM);
Citation: Ebadpour, M.; Jamshidi, M.; state estimation; Metaverse
Talla, J.; Hashemi-Dezaki, H.; Peroutka,
Z. Digital Twin Model of Electric
Drives Empowered by EKF. Sensors
2023, 23, 2006. https://doi.org/ 1. Introduction
10.3390/s23042006 Digital manufacturing has been beneficial for modern industries because of the ad-
Academic Editor: Aleksei Tepljakov vances in communication and information technology. This has been made possible through
the use of digital twin (DT) models and process development simulations [1–3]. The digital
Received: 14 January 2023 twin model, which is a system that can simulate the physical twin and is made up of
Revised: 4 February 2023
multiple probabilistic phases, has been developed since around 2000. It is a formal concept
Accepted: 8 February 2023
that allows for the creation of a feasible model [4]. A digital twin needs a platform to create
Published: 10 February 2023
a digital simulation of a real-world process, as well as to perform simulations of electric
drive systems. The Metaverse is an excellent virtual platform for digital twin models
because it combines digital and physical environments through the use of communication
Copyright: © 2023 by the authors.
technologies, web infrastructures, and the Internet of Things. Many researchers have been
Licensee MDPI, Basel, Switzerland. interested in various platforms, but the Metaverse, with its ability to allow users to work in
This article is an open access article a virtual world through avatars, has shown the importance of conducting research [5,6]. In
distributed under the terms and a study [7], the instability in flexible fixturing systems caused by incomplete constraints and
conditions of the Creative Commons weak rigidity has been addressed. This way, a digital twin-based approach to generate re-
Attribution (CC BY) license (https:// configurable fixturing schemes, using a hidden Markov model to account for time-varying
creativecommons.org/licenses/by/ system stiffness and coupling effects, was proposed. The optimization process was guided
4.0/). by local information from front-running simulations. The approach was demonstrated

Sensors 2023, 23, 2006. https://doi.org/10.3390/s23042006 https://www.mdpi.com/journal/sensors

Sensors 2023, 23, 2006 2 of 21

through trimming experiments on a large compliant workpiece using a reconfigurable

fixturing system developed by the authors.
The practical implementation of electric drives has been difficult for manufacturers
and researchers due to the high cost of installation and the time-consuming nature of the
process. This has made digital twinning a useful application in these situations. Three-
phase induction motors (IMs) have been widely used in a variety of industrial projects, such
as electric vehicle propulsion drives, railway traction, and aerospace industries, due to their
simple and robust construction, low maintenance, and cost-effectiveness [8]. Nevertheless,
controlling Ims can be complex due to their highly nonlinear model, which has an unknown
load profile and time-varying parameters. This requires a thorough understanding of rotor
flux and position/speed to be effective. In addition, speed sensorless IM drives have several
advantages, including low cost, reduced complexity, and improved reliability [9,10]. Digital
twins for IM drive systems are not commonly used for drive performance monitoring,
health diagnostics, state optimization, and risk evaluation at different levels, including the
entire system, subsystems, specific components, and others [11].
There have been several approaches to improving the control of Ims, including model-
based predictive control (MPC), field-oriented control (FOC), and direct torque control
(DTC). There have been numerous reports that have combined the benefits of IM drives
with sensorless control techniques to create low-cost, high-performance actuators [12,13].
In the literature, there have been a variety of approaches proposed for sensorless control
of IM drives, but many challenges and uncertainties remain. To address these issues,
several model-based estimation techniques have been suggested in the literature, including
full-order estimators [14], sliding-mode observers [8], nonlinear Kalman filters, extended
Luenberger observers (ELO), and model reference adaptive systems (MRAS) [15].
Nonlinear Kalman filter-based observers, the most commonly used algorithm, provide
a probabilistic view of state estimation problems because they rely on the measurement
and system noise covariance in the estimation process [16]. In recent years, various non-
linear Kalman filter algorithms such as the Extended Kalman filter (EKF) [17], unscented
Kalman filter (UKF) [18], and cubature Kalman filter (CKF) [19] have been applied to the
state/parameter estimation of Ims. The EKF method, which is a linearized version of the
traditional Kalman filter, serves as an optimal state estimator for actual values of Ims such
as rotor flux, stator currents, speed, and parameters based on the system Jacobian matrix.
Additionally, the computational requirements of the EKF, which are largely influenced by
the size of the motor states, are lower than those of the other mentioned nonlinear Kalman
filters. These beneficial features make the EKF estimator more general and universal
compared to most other observers.
The integration of the Kalman filters with a virtual asset of the real system, called
DT, can offer real-time process evaluation. Considering of such data in actual system
operation can enhance control performance with improved productivity. Several examples
in the literature present use of the EKF algorithm for DT purposes in different aspects and
applications. DT of the backer’s yeast batch cultivation system is analyzed in [20] based on
the linear discrete Kalman filter, and its nonlinear variants such as EKF and UKF are used
to predict the bioprocess variables employed for optimization and control of the real entity.
Various aspects of Li-ion battery prognostics and lifespan monitoring are reviewed in [21]
using the DT technology based on the EKF algorithm for estimation of state-of-charge,
capacity, current/voltage, and remaining-useful-life. It concludes that the combination of
DT with an EKF observer can prevent the battery from having severe failure, optimize its
maintenance schedules, and conduct health monitoring with a focus on fault identification
and correction. In study [22], the sparse detection of nonlinear dynamics challenges is
integrated with the EKF algorithm to automatically recognize the time-variant DT models
for online system monitoring. The robustness of the algorithm is validated through a
simulation model based on Lorenz process and an industrial diesel hydrotreating case. A
research work in [23] presented technical features of merging machinery with a physics-
based DTs based on EKF methods for providing successful implementation information.
Sensors 2023, 23, 2006 3 of 21

In this regard, two industrial processes were analyzed to demonstrate the methodology;
the first is a mobile log crane, which is applicable for the heavy mobile machinery, and the
second process concerned a rotating electric machine in case of unexpected failures.
To obtain a reliable state estimate of an IM using the EKF, accurate information on
stator phase voltages is essential. Many previous studies have used an ideal inverter
model in the IM drive with low-precision stator voltage knowledge as input to the EKF
algorithm. However, the switching devices of an inverter are not ideal in practice, as they
have small dead times, turn-on and -off times, and voltage drops in each switching interval
to prevent shoot-through on the DC link. Dead time creates a nonlinear converter feature
that distorts the stator voltage. In other words, it produces a set of low-order harmonics
in the stator voltage of the drive inverter, leading to additional losses, particularly at high
switching frequencies [24]. To address this distortion and make the model more feasible,
the effect of the dead time and voltage drop of inverter switching devices must be taken
into account through practical modeling of the inverter. While much research has been
conducted on sensorless IM drives based on the EKF using an inaccurate ideal model of
the inverter [24–26], there is relatively little literature on practical state estimation.
This paper presents a practical digital twin EKF-based sensorless rotor FOC technique
for IM drives. Unlike previous studies, the effect of switching devices’ time delays and
voltage drops is mathematically analyzed and considered in the estimation process. To
evaluate online state estimation, the measurement and system noise covariance matrices are
included in the drive model. Accurate values of stator current and rotor flux are obtained in
stationary reference frames through a recursive updating process. The rotor position/speed
and load torque are also estimated using a precise EKF algorithm. This study is conducted
on a 4 kW three-phase IM using MATLAB/Simulink in two operating scenarios, one using
an inaccurate ideal model of the inverter and one based on an accurate feasible model. The
organization of this manuscript is as follows. In Section 2, the concepts of digital twins
and the Metaverse in field of electrical drives are effectively explained. In Section 3, the
mathematical model of the drive system, including the IM model, EKF algorithm, and DT
model of the three-phase inverter, is presented. Section 4 describes the control strategy
based on the rotor FOC method. In Section 5, the simulation results are presented with
more numerical analyses, and the final sections contain the discussion, conclusion, and
future work.

2. Concepts of Digital Twins and the Metaverse in Electrical Drives

2.1. Digital Twins and Electrical Drives
Over the past twelve years, significant research results have emerged, with the con-
cepts of parallel intelligence published in 2004 and digital twins coined in 2010 leading
to important implications for the study of Cyber Physical Systems (CPS) and the latter to
Cyber Physical Social Systems (CPSS) [27–29]. These concepts have provided innovative
insights for addressing challenges related to the construction of digital intelligent societies,
such as the integration and mining of large amounts of data from Internet of Things de-
vices, the separation between artificial systems and actual systems, and the coordination
of diverse, multisource resources. The key difference between digital twins and parallel
intelligence lies in their research infrastructure, with the former being related to CPS and
CPSS. In addition, digital twins did not initially consider human factors, while parallel
intelligence was primarily focused on them [27].
A new method for controlling deformation of thin-walled parts during milling using a
digital twin called milling process digital twin was presented in a study [30]. Accordingly, a
new reference framework for online optimization and control of milling deformation, which
improved the quality of the final product, was introduced. The method was enabled by
three key technologies: multidimensional high-fidelity modeling, knowledge-driven low-
latency milling deformation simulation, and online optimal control of milling deformation.
The authors also presented a prototype system of the proposed method that demonstrated
its feasibility in an industrial setting.
Sensors 2023, 23, 2006 4 of 21

In addition, another study [31] has a different perspective and mentioned that in the
past, digital modeling techniques such as BIM and data acquisition tools were used for
construction and controlling physical objects. However, digital twins (DTs) offer unique
features such as bi-directional data exchange and real-time self-management. The need for
DTs has increased, particularly after COVID-19, as they can be useful in various industries.
The paper aimed to explain the concept of DTs and how they differ from other technologies
and systems. The current state of DT development was also reviewed, and suggestions
for future research were provided. The focus was on the Smart City, engineering and
construction sectors, and the need for DT applications that can provide real-time decision-
making, self-operation, and remote supervision capabilities post-COVID-19 [31].
The CPS system, which is made up of the physical and virtual worlds working together,
is difficult to fully design from the start. It is meant to provide benefits such as real-time
optimization and adaptability [32]. The idea of a digital twin, which is a real-time digital
representation of a physical object, is being explored as a way to better understand and
control the CPS system.
A digital twin Is a virtual replica of a real-world physical system or product that
serves as a digital equivalent for practical purposes such as system simulation, testing,
maintenance, and monitoring. This concept is particularly useful for complex systems,
where each component affects the overall performance of the system. Materials can also
be included in this classification. By using a digital twin in real time and synchronizing
it with the corresponding physical system, such as alloys, it is possible to speed up the
process of modeling and monitoring the system. Simultaneously running detailed system
verification and validation test scenarios on both the digital twin and the physical twin,
in this case, a specific alloy, is a practical way to evaluate the effectiveness of the digital
twin [33]. The idea behind the digital twin of an electrical drive is to create a digital copy of
an electrical drive and use it to improve the real-world performance of that object through
simulations and optimization techniques. This is accomplished by constantly updating the
digital representation with real data. It is a key aspect of the fourth industrial revolution,
specifically in the areas of digitalization and simulation.

2.2. Digital Twins of Electrical Drives in the Metaverse

To clarify the difference between digital twin and the Metaverse, a digital twin is a
virtual replica of an existing physical object or system, whereas the Metaverse is a virtual
universe that comprises multiple digital twins and allows for interactions between them.
In this study, we focus on the characteristics of the digital twin of an electric drive and also
demonstrate some features of the Metaverse, such as value chains, production relationships,
and virtual interactions of digital and physical agents. The digital twin of the related electric
drive, on the other hand, is used as a virtual representation of its physical twin, and we
have collected data from a real system to create it. Figure 1 illustrates the concept of the
proposed method.
The use of digital twins of electrical drive technologies across various industries has
not been thoroughly studied yet due to the diversity of sub-processes and manufacturing
techniques, leading to a vast number of possibilities. The Metaverse, also known as
the Internet of 3-D worlds, has recently garnered significant interest from academia and
industry. Each virtual subworld, operated by a virtual service provider (VSP), offers a
specific type of virtual service. Digital twins, which are digital replicas of physical objects,
are crucial enabling technologies. Generally, a DT belongs to the party that develops it and
establishes the communication link between the two worlds. However, in an interoperable
Metaverse, data-like DTs can be “shared” within the platform, allowing a single set of
DTs to be utilized by multiple VSPs. However, the quality of shared DTs may not always
be sufficient [34–36]. In other words, the Metaverse is a developing technology that
creates virtual environments for users to access a wide range of virtual services, while also
allowing users to experience immersive interactions with the real world. Digital twins,
Sensors 2023, 23, 2006 5 of 21

Sensors 2023, 23, x FOR PEER REVIEW

which 5 of 22
represent assets in this virtual world, are crucial in connecting this environment to
the actual world.

Figure 1. The representation of a physical electrical drive and its digital twin within a Metaverse
Figure 1. The
platform. Therepresentation
use of the termofdigital
a physical electrical
twin refers drive
to the and its
concept digital twin
of creating withincounterpart,
the digital a Metaverse
or twin, of the physical electric drive. In this context, the physical electrical drive and its digital twinor
platform. The use of the term digital twin refers to the concept of creating the digital counterpart,
are existing
twin, simultaneously
of the physical electricwithin
drive. the Metaverse
In this context,platform, creating
the physical a bridge
electrical between
drive and itsthe physical
digital twin
and digital realms. The figure highlights the innovative approach of using the Metaverse
are existing simultaneously within the Metaverse platform, creating a bridge between the physical platform
to bring
and together
digital realms.both
The the physical
figure and digital
highlights aspects of
the innovative an electrical
approach drive.
of using the Metaverse platform to
bring together both the physical and digital aspects of an electrical drive.
The use of digital twins of electrical drive technologies across various industries has
not been thoroughly studied yet due to the diversity of sub-processes and manufacturing
techniques, leading to a vast number of possibilities. The Metaverse, also known as the
Sensors 2023, 23, 2006 6 of 21

The potential of data-driven operational support through predictive analysis is limited.

A new approach is a model-based simulation of operational behavior, which allows for
the simulation of specific physical effects and the monitoring of system behavior even for
data that cannot be directly measured. A simulation model that supports plant monitoring
is known as a digital twin, which provides additional information about the asset state.
Improved knowledge of system behavior increases the availability of the plant and the
ability to predict potential faults during operation [37].
The Metaverse can be used to help with the application of these systems by providing
a platform for simulation, training, and remote collaboration. One way to use the Metaverse
is through simulation. By creating virtual copies of electric drive systems in the Metaverse,
engineers and designers can test the syste’s behavior and try out different configurations
and control strategies before building the physical system. This can help to identify
and solve potential problems before they happen in real life and improve the syste’s
performance. Another way to use the Metaverse is through training. By creating virtual
training environments in the Metaverse, engineers and maintenance workers can learn how
to operate and maintain electric drive systems in a safe and controlled environment. This
can help to reduce the risk of accidents and make maintenance operations more efficient.
The Metaverse can also be used to make it easier for engineers, researchers, and other
experts to work together on electric drive systems. By providing a virtual space where
they can share information, work on designs, and conduct experiments, the Metaverse can
help speed up the development and deployment of electric drive systems. Data-driven
operation support has been a topic of interest for approximately ten years. The effectiveness
of methods such as condition-based monitoring or sensor-based fault detection relies on
the quantity and placement of sensors. The demand for simulative operation support is
relatively new [38].

3. Digital Twin Model of the Drive System

The modeling of specific physical processes enables monitoring of the system operation
even for information that cannot be directly measured. A simulation model which allows
for feasible monitoring is known as a digital twin. In the context of this article, the precise
mathematical model to implement digital twins of the electrical drive system includes
three parts: the IM, three-phase inverter, and EKF estimator. These parts are explained in
detail in the article. In this section, the core task of the digital twin model, which is the
state estimation of speed sensorless induction motor drive, is examined. Our approach
is to employ a practical EKF algorithm to create digital twins for model-based condition
monitoring. This allows for the ability to predict and diagnose potential issues with the
system, providing valuable insights for maintenance and operation. In addition, our
approach also enables the monitoring of the syste’s performance and identifying areas
for improvement.
In the case of an electrical drive system, a digital twin can be created by modeling the
different physical processes that take place within the system, such as the operation of the
IM, the three-phase inverter, and the control algorithms. This model can then be used to
simulate the syste’s performance in real-time. By comparing the simulated results to the
actual performance of the physical system, it is possible to identify potential problems and
predict future failures. One of the main advantages of using a digital twin for an electrical
drive system is the ability to monitor information that cannot be directly measured. For
example, a digital twin can be used to estimate the syste’s state, such as the speed and
position of the motor, even when there are no sensor data. This is accomplished by using
an EKF algorithm which can estimate the syste’s state based on other available information
such as current and voltage measurements. In addition, a digital twin of an electrical drive
can also be used for condition monitoring, which is the practice of monitoring the condition
of a machine or system to detect any signs of wear or degradation. This allows for early
detection of potential issues and helps prevent unexpected failures. Additionally, a digital
Sensors 2023, 23, 2006 7 of 21

twin can be used to optimize the performance of the system, by identifying areas where
energy efficiency can be improved or where the system can be operated more efficiently.

3.1. Three-Phase IM Model

To identify the rotor position of IM and estimate the speed and other states, an
accurate motor model along with possible data coming from sensors and simulation results
is considered in this study. The sixth-order rotor flux based three-phase IM mathematical
model is used for the EKF algorithm in a stationary reference frame (αβ0) to estimate
the stator currents (isα , isβ ), rotor fluxes (ψrα , ψrβ ), speed (ωe ), and load torque (τl ). To
implement the EKF digital twin model on a digital processor or microcontroller platform, it
is better to be defined it in a discrete form. The nonlinear discrete IM model can be given as:

x k +1 = f ( x k , u k ) + w k
, (1)
zk = h( xk ) + vk

The model in (1) can be linearized as follows:


xk+1 = A( xk ) xk + Buk + wk
, (2)
zk = Hxk + vk

where xk is an (n × 1) state vector matrix; zk is an (m × 1) current measurement vector

matrix; H is the (m × n) measurement matrix; f ( xk , uk ) is a known nonlinear function of
the state’s transition vector (Jacobian matrix). In this study, it is assumed that the process
noise term wk is white and zero mean with an (n × n) covariance matrix Q and that the
measurement noise parameter vk is also white and zero mean with an (m × m) covariance
matrix R.
The final discrete mathematical model of IM can be obtained as (3) and (4) using
the general form in (1) and (2), wherein usα and usβ are the measured stator voltages in
stationary reference frame based on the (6). ωe is the electrical angular speed of the rotor
which equals to pole pairs (p p ) time of the mechanical speed of rotor ωm (i.e., ωe = p p .ωm ).
The motor coefficient matrix A( xk ) can be expressed as (5).

isα,k+1 isα,k a5 0
     

 isβ,k+1 


 isβ,k  
  0 a5 
 
 ψrα,k+1   ψrα,k   0 0  usα,k
 = A( xk ) +
 usβ,k + wk , (3)
 

 ψrβ,k+1 


 ψrβ,k  
  0 0 | {z }
 ωe,k+1   ωe,k   0 0 
uk
τl,k+1 τl,k 0 0
| {z } | {z } | {z }
x k +1 xk B

   
isα,k+1 100000
= xk + vk , (4)
isβ,k+1 010000
| {z } | {z }
zk H

a1 − a2 /τr 0 a3 /τr a3 ωe,k 0 0

 

 0 a1 − a2 /τr − a3 ωe,k a3 /τr 0 0  
 a4 /τr 0 1 − Ts /τr − Ts ωe,k 0 0 
A( xk ) =  , (5)

 0 a4 /τr Ts ωe,k 1 − Ts /τr 0 0  
 − a6 ψrβ,k − a6 ψrβ,k 0 0 1 − a7 − a8 
0 0 0 0 0 1
" # v 
2 1 1 sa
− 3 − 3
 
usα 3
= √1 √−1  vsb , (6)
usβ 0
3 3 vsc
Sensors 2023, 23, 2006 8 of 21

In (5), the parameter Ts is the sampling time and the coefficients used in it are listed
as follows:
1 3p2p Lm
a0 = Ls Lr − L2m
, a1 = 1 − a0 Rs Lr Ts , a2 = a0 L2m Ts , a3 = a0 Lm Ts , a4 = Lm Ts , a5 = a0 Lr Ts , a6 = 2Jt Lr Ts ,
Bt pp
a7 = Jt Ts , a8 = Jt Ts , τr = Lr /Rr

where Rs and Ls are the stator resistance and inductance, respectively; Rr and Lr are
the rotor resistance and inductance referred to the stator side, respectively; Lm is the
mutual inductance; Jt is the total moment inertia of the IM and load; Bt is the viscous
friction coefficient.

3.2. Inverter Actual Model

Pulse width modulation (PWM) and space vector pulse width modulation (SVPWM)
are two universal switching techniques for the inverters which are feeding the three-phase
IMs [39–41]. The SVPWM strategy has been widely employed in the IM drive systems
due to several merits such as containing small torque ripple, low operation noise, the vast
range of DC voltage utilization, and ease. For practical modeling of a three-phase inverter
controlled by SVPWM, all switching device0 s time delays and voltage drops must be
rigorously considered to prevent short-circuit faults that may be caused by simultaneously
turning on the two switches of the same bridge leg. Here, a three-phase six-switch voltage
source inverter empowered by SVPWM strategy is used for supplying the stator windings
of IM, and its A-phase current (i a ) paths for positive and negative switching intervals are
illustrated in Figure 2. The current flowing from the phase to the IM, which is indicated
by the blue solid arrow, either through the upper A-phase switch Ta+ or the lower related
diode Da− , is determined as a positive phase current, and the current coming from the IM
Sensors 2023, 23, x FOR PEER REVIEW 9 of 22
to the inverter leg, which is indicated by the blue dotted arrow, either through the lower
− +
A-phase switch Ta or upper diode Da , is determined as a negative A-phase current.

U dc + Ta+ Tb+ Tc+

Da+ Db+ Dc+
2 _
ia > 0 ia
u ao
a ia < 0 ib
IM
ic

U dc Ta− Tb− Tc−

2 Da− Db− Dc−

Figure2.
Figure 2. Electric
Electric diagram
diagram of
of the
the three-phase
three-phasevoltage
voltagesource
sourceinverter.
inverter.

Toturn
To turnONONand andOFF OFFthethe switching
switching devices
devices of the
of the inverter,
inverter, the SVPWM
the SVPWM technique
technique uses
uses eight
eight switching
switching states states
with sixwith six active
active sectorssectors
to formtoa form
PWMasignals
PWM signals
and shapeandthe
shape the
output
output waveform
waveform equivalentequivalent to a sinusoidal.
to a sinusoidal. The voltageThespace
voltage space
vector vector of
diagram diagram
the SVPWMof the
SVPWMbased
strategy strategy based
on the on instruments
Texas the Texas instruments switching
switching patterns patterns (TMS320C24x/F24x
(TMS320C24x/F24x code [42])
code
is [42]) isindepicted
depicted Figure 3, in in
Figure
which 3, in
U1which
− U6 areU 1 the
− U 6basic
are the basic
active active vectors
voltage voltage of
vectors of
the six
the six switching
switching sectors sectors
with two with twovectors
zero O0 andOO0 1 .and
zero vectors UoutOis1 .the
U out is thereference
output output reference
voltage
space
voltage vector
spacewhich
vector is which
synthesized from twofrom
is synthesized adjacent basic voltage
two adjacent basicvectors
voltage(here from
vectors U1
(here
and U ) with interval times
from 2U 1 and U 2 ) with interval of T and T .
1 times2of T and T .
1 2
The A-phase corresponding switching time delays analysis is shown in Figure 4. The
overall principles of the B and C-phase bridge legs is similar to the A-phase. Figure 4
β + −
shows the ideal switching signals for the upper (Ta_ideal ) and lower (Ta_ideal ) switches of
U 3 (010) U 2 (110 )
the A-phase inverter leg, the real switching signals based on the dead-time delay td for the
+ −
upper (Ta_real ) and lower (Ta_real ) switches, the ideal voltage waveform of the A-phase leg
p
(u ao_ideal ), and the real positive value of A-phase voltage (u ao_real ) and negative (unao_real )
phase current. Although the dead-time U outdelay in SVPWM setting for practical drives causes
T1
U 4 (011) O0 (000) U1 (100)
θ α
O1 (111) T2
 Sensors 2023, 23, x FOR PEER REVIEW 9 of 22

Sensors 2023, 23, 2006 9 of 21

U dc + Ta+ Tb+ Tc+

Da+ Db+ Dc+
2 _
distortion inuinverter >0
output iawaveforms ia
and deteriorates the system performance, using
ao
a
the proposed precise EKF algorithm ia < 0 i
this issue is bperfectly addressed. For simplicity of
IM
i
analyzation, it is assumed that the dc-link voltage cof the inverter which is denoted by Udc
U dc constant value.
has Ta− The forward Tb− voltage−drops
Tc− on Dthe active switches and the diodes, which
−
2 Da− Db c
are denoted by UT and UD , respectively, are calculated based on their ON-state resistance
characteristics and the absolute value of related phase current as (7).

Figure 2. Electric diagram of the three-phase  UT =voltage source

Vf T inverter.

rT ia +
, (7)
To turn ON and OFF the switching devices of the inverter, the SVPWM technique
 U D = r D i a + Vf D
uses eight switching states with six active sectors to form a PWM signals and shape the
outputVfwaveform
where T and Vf D equivalent to a sinusoidal.
are the inherent forward voltage The voltage space vector
of the switches diagram
and diodes. Tonofandthe
TSVPWM strategy based on the Texas instruments switching patterns (TMS320C24x/F24x
o f f are the ON- and OFF-states times of the upper switches of the inverter, and t on and
tcode
o f f are the is
[42]) finite turn-on
depicted and turn-off
in Figure times of
3, in which U 1the
− Uupper and lower switches, respectively.
6 are the basic active voltage vectors of
To prevent a short-circuit in the phase bridge legs of the inverter, the dead-time amount
the six switching sectors with two zero vectors O0 and O1 . U out is the output reference
should be chosen as td > (ton + to f f ). In addition, in practice, the turn-off time to f f is a
voltage
little space
greater thanvector which turn-on
the switch is synthesized
time tonfrom
. It is two
also adjacent
necessarybasic voltage vectors
to determine td > to f(here
f to
from U 1 and
avoid simultaneously U 2 ) with interval times
turning on the upper and of T1 and T .
lower2 switches of the bridge legs.

β
U 3 (010) U 2 (110 )

U out
T1
U 4 (011) O0 (000) U1 (100)
θ α
O1 (111) T2

U 5 (001) U 6 (101)

Figure3.3.Voltage
Figure Voltagespace
spacevectors
vectorsbased
basedon
onthe
theSVPWM
SVPWMimplemented
implementedon
onTMS320C24x/F24x.
TMS320C24x/F24x.

The A-phase
From Figure 4,corresponding switching
the output A-phase time delays
voltage whichanalysis is shown
is averaged within Figureto
respect 4. the
The
overall principles
midpoint of the
of the dc-link B aand
over C-phase
switching bridge
period Tswlegs is similar
for both to the
positive andA-phase. Figure 4
negative currents
can be precisely
shows the idealcalculated
switchingas:signals for the upper ( T + ) and lower ( T − ) switches
a _ ideal a _ ideal
of the A-phase inverter leg,
dd the
− 12 real
)Udc switching
− (d − dd )signals
UT − (1based
− d +ondd )the , i a > 0 delay t d
UD dead-time

(d −
û ao_real =+ 1 −
, (8)
for the upper ( Ta _ real(d) +
and − 2 )U(dcTa+_ (real
dd lower 1 −) dswitches,
− dd )UT the
+ (dideal
+ ddvoltage
)UD , i awaveform
<0 of the
p
A-phase
in which dleg is (the
u aoideal ), and the real
_ ideal duty-cycle positive
of the activevalue
switches of A-phase
which isvoltage
derived( ufrom the
ao _ real ) and neg-
control
module and n defined as d = T /T , and d is the overall time delay in a switching period
ative ( u ao _ real ) phase current. on
Although
sw
the
d
dead-time delay in SVPWM setting for prac-
which can be calculated as:
tical drives causes distortion in inverter output waveforms and deteriorates the system
performance, using the proposed precise
dd = (td + tEKF
on −algorithm
to f f )/Tsw ,this issue is perfectly addressed. (9)
For simplicity of analyzation, it is assumed that the dc-link voltage of the inverter which
Based on
is denoted by Figure
U dc has 4 and the inverter
constant value. digital twin model
The forward voltageequation
drops on inthe
(8),active
it is observed
switches
that the ON-state time of upper switches in actual operation should be precisely determined
and the diodes, which are denoted by U and U D , respectively, are calculated based
as td + ton < Ton < Tsw − td − ton ; however,T in the ideal model, the impact of time delays
on their ON-state resistance characteristics and the absolute value of related phase current
as (7).
 U T = rT | ia | +V fT
 , (7)
U D = rD | ia | +V fD
where V fT and V fD are the inherent forward voltage of the switches and diodes. Ton
and Toff are the ON- and OFF-states times of the upper switches of the inverter, and t on
Sensors 2023, 23, 2006 10 of 21
and t off are the finite turn-on and turn-off times of the upper and lower switches, respec-
tively. To prevent a short-circuit in the phase bridge legs of the inverter, the dead-time
amount
and should
voltage dropsbeonchosen as t d devices
switching > (t on + are . In addition,
t off )neglected whichin results
practice,
in the
0 <turn-off
T < time
T ; on sw
t off is a little
thereafter greatervoltage
the output than the switch turn-on
magnitude time tot on
is changed as. follows:
It is also necessary to determine
t d > t off to avoid simultaneously turning on the upper and lower switches of the bridge
(d − 12 )Udc , i a > 0

legs. û ao_ideal = , (10)
(d − 1 )Udc , i a < 0 2

Ta+_ ideal

Ta−_ ideal
td td
Ta+_ real
t
Ta−_ real
t

U dc
u ao _ ideal 2
U dc t
2
t on t off

U dc
p − UT
u ao _ real 2
U dc
+U D
2
t off t on

U dc
n +U D
u ao _ real 2
U dc
− UT
2
Tsw

Figure4.4.Switching
Figure Switchingpulses
pulsesand
andoutput
outputvoltage
voltagefor
forthe
theA-phase
A-phaseduring
duringone
oneswitching
switchingperiod.
period.

From
3.3. EKF Figure 4, the output A-phase voltage which is averaged with respect to the
Algorithm
midpoint
In thisof the dc-link
study, a precise overEKFaalgorithm used forTsw
switchingisperiod forestimation
state both positive
of theand negative
IM drive cur-
system
rents can
based be precisely
on the actual model calculated as:
and values of the inverter in comparison with the previous
established research which relied on 1 the ideal model. The EKF, as the most preferred

nonlinear Kalman filter, provides  ( d − d d a− )U dc − ( dbehavior
stochastic − d d )U T to
− (1the
− dstate )U D , ia > issue,
+ d d estimation 0 and its
ˆ 2 ,
u ao _ real
success directly depends = 
on the drive system and measurement noises evaluated in the (8)
 1
estimation process. In comparison ( d + d −
d to the)U + (1 − d − d
dc other nonlinear
d )U + ( d + d
T estimators, d )U , i < 0
DEKFais still a decent
 2
choice due to less computational complexity and accurate estimation performance. The
operation d is the
in which steps ideal duty-cycle
of conventional EKFofalgorithm
the active based
switches on which is derived IM
the three-phase from the control
model in (3)
module
and and
(4) can bedefined d T / T d
given asasfollows:on sw , and d is the overall time delay in a switching pe-
=
riod Initialization
(1) which can be calculated as:
d d = (t d + t on − t off ) / Tsw , (9)
x̂0 = E[ x0 ], (11)

P0 = E[( x0 − x̂0 )( x0 − x̂0 ) T ], (12)

(2) Linearization

∂ f ( xk , uk )
Fk+1|k = , (13)
∂xk
(3) Estimation or time updating
Sensors 2023, 23, 2006 11 of 21

x̂k− = f ( x̂k−1 , uk−1 ), (14)

Pk− = Fk|k−1 Pk−1 FkT|k−1 + Q, (15)

(4) Correction

−1
Kk = Pk− H T [ HPk− H T + R] , (16)
x̂k+ = ( I − Kk H ) x̂k− + Kk zk , (17)
Pk+ = ( I − Kk H ) Pk− , (18)
where F is the Jacobian matrix to linearize the nonlinear model of IM; P− and P+ are the
priority and the posterior covariance matrices, respectively; K is the Kalman gain; I is the
identity matrix.
From Equations (14) and (15) it can be concluded that the accuracy of EKF-based state
estimation strongly depends on IM model, inputs, and parameters. Motor parameters
Sensors 2023, 23, x FOR PEER REVIEW 12 of 22
are assumed to be constant over the optimum working modes. The main inputs of the
EKF algorithm are stator voltage terms in the stationary reference frame (i.e., usα and usβ )
which can be originated either from the control system multiplied by dc-link voltage (low
precise voltages)
precise voltages) or
or from
from the
thevoltage
voltagesource
sourceinverter
inverteralong with
along withprecise information
precise of of
information
switching devices from SVPWM strategy. Therefore, to practically estimate the IM
switching devices from SVPWM strategy. Therefore, to practically estimate the IM states, states,
allswitching
all switching time
time delays
delaysand
andvoltage
voltagedrops
dropsshould bebe
should considered
consideredin input voltage
in input values.
voltage values.
To better analyze the proposed precise estimation process, switching patterns and
To better analyze the proposed precise estimation process, switching patterns and voltagevoltage
waveforms of the sector I of the SVPWM technique based on Figure 3 are considered as
waveforms of the sector I of the SVPWM technique based on Figure 3 are considered as
shown in Figure 5.
shown in Figure 5.
Tsw
td td td td td td
Ta+_ ideal

Tb−_ ideal

Tc−_ ideal

Ta+_ real

Tb−_ real

Tc−_ real

p
uao _ real

n
ubo _ real

n
uco _ real

Ton Toff Toff Ton Ton Toff

Figure 5.
Figure 5. Analysis
Analysis of
of time
timedelays
delayseffect
effectofofthe inverter
the switching
inverter devices
switching (depicted
devices in yellow
(depicted color)color)
in yellow
during sector I.
during sector I.

From Figure
From Figure 5,
5, the
the dead-time
dead-timeeffect
effectand
andswitching
switchingONON and OFF
and OFFdelays impact
delays on the
impact on the
three-phase inverter performance are analyzed when the reference voltage
three-phase inverter performance are analyzed when the reference voltage vector vector UUoutout is
is located in the sector I wherein ia > 0 , ib < 0 , and ic < 0 . The real positive value of A-
p
phase voltage u ao _ real is shorter than the ideal value u ao _ ideal by time delay
∆T = t d + t on − t off ; for the same reason, the real negative values of B- and C-phase volt-
n n
Sensors 2023, 23, 2006 12 of 21

located in the sector I wherein i a > 0, ib < 0, and ic < 0. The real positive value of A-phase
p
voltage u ao_real is shorter than the ideal value u ao_ideal by time delay ∆T = td + ton − to f f ;
for the same reason, the real negative values of B- and C-phase voltages unbo_real and unco_real
are longer than the ideal value by ∆T.
According to Figure 5, when the switching states of A, B, and C phases are compared,
the actual switching time of the voltage vector in sector I U1 (100) is reduced by 2*∆T in
one PWM period (the left side of the positive value and right side of negative value are
reduced by ∆T). However, the actual switching time of the vector U2 (110) in Figure 3 does
not change at this time. If the voltage vector difference caused by the dead-time effect is
∆U, then the difference between the real voltage vector of the stator Us and the reference
voltage vector Uout at this time is determined as:

∆U = −2 × ∆T × U1 , (19)

Based on the above analysis, it is clear that there is a sensible difference between the
real and ideal voltage which is derived from the three-phase inverter. Hence, the proposed
DT model of the inverter which is relied on the actual model can supply precise voltages to
the EKF algorithm to practically estimate the speed in sensorless drives.
Here, only the sector I of SVPWM is based on the assumptions of i a > 0, ib < 0, and
ic < 0, is analyzed to illustrate that the different positive and negative current polarities
can directly affect the results of the stator voltages due to the dead-time effects. As a result,
the determination of the polarity and value of the three-phase currents in each sector is a
key factor for the analysis of the dead-time effect of the three-phase inverter.
Since during the dead-time intervals there is no information about the three-phase
currents due to the off-state of all switches of the inverter bridge leg, the ideal inverter
models and conventional EKF algorithms cannot perfectly act to precisely estimate the
IM states. The presented DT model of the inverter along with the accurate EKF algorithm
fully addressed the mentioned problem by estimating the currents using the actual voltage
values based on the above analysis and (8). To estimate the states during the dead-time
intervals, the proposed EKF algorithm uses the real current and voltage values from the
previous intervals inside the sectors by modifying the voltage in (14) as follows:

x̂k− = f ( x̂k−1 , ureal

k −1 ), (20)

wherein ureal
k −1 is the actual stator voltage based on the practical DT model of the inverter.

4. Sensorless Control Strategy

The overall block diagram of the sensorless rotor FOC strategy based on the precise
EKF estimator is shown in Figure 6. The IM drive system consists of two control loops for
speed control and rotor flux regulation, SVPWM switching module, DT model of the three-
phase inverter, the mathematical model of the IM, the precise EKF estimator, and various
sensors related to current and voltage along with the Park and Clark transform blocks.
These sensors provide data about different aspects of the IM performance, operational
modes, and more. These data are then transferred to a processing system and applied to the
digital processors. Once informed with such information, the virtual DT model can be used
to run simulations, study performance targets, and produce possible signals/feedbacks
which can then be applied back to the original physical asset. As shown in Figure 6, rotor
speed and position data are obtained using the estimator instead of employing mechanical
sensors, which makes the drive system suitable for cost-effective applications.
Unlike the conventional current-based control strategy which is affected by variation
of rotor resistance results to cause an error in flux value evaluation, for flux amplitude
and angle computation the stationary reference frame voltage model is utilized in this
paper which is insensitive to the rotor resistance variations. In order to better analyze the
control system, the reference angular speed ωm ∗ is compared to the estimated motor speed

ω̂m generated by the EKF estimator, and the speed error is applied to a PI (proportional-
 4. Sensorless Control Strategy
The overall block diagram of the sensorless rotor FOC strategy based on the precise
EKF estimator is shown in Figure 6. The IM drive system consists of two control loops for
speed control and rotor flux regulation, SVPWM switching module, DT model of the
Sensors 2023, 23, 2006
three-phase inverter, the mathematical model of the IM, the precise EKF estimator, and 13 of 21
various sensors related to current and voltage along with the Park and Clark transform
blocks. These sensors provide data about different aspects of the IM performance, opera-
tional modes, and more. These data are then transferred to a processing system and ap-
plied to the
integral) digital
speed processors.
controller. TheOnce informed
output of the with
speedsuch information,
controller, which theisvirtual DT as
considered
model can electromagnetic
reference torque τe∗ , study
be used to run simulations, performance
is used to calculatetargets, and produce
the q-axis reference possible
∗
current Iqs
signals/feedbacks
based on (21) whereinwhichψ̂can
rd isthen be applied
the d-axis back to
estimated the original
rotor flux. Iqs physical
∗ is comparedasset.toAs
theshown
measured
in Figure
q-axis 6, rotor
current Iqsspeed and
, and its position
error data areby
is regulated obtained using the estimator
the PI controller to generate instead of em-
a voltage control
ploying
signal Vmechanical
qs which is sensors,
c utilized which makesq-axis
to produce the drive system
reference suitable
voltage forascost-effective
∗
Uqs expressed inap- (22) in
plications.
which ωs is stator field angular speed in terms of rad/s.

ωm* + Speed
*
I sq + iq current
*
U sq U s*α Ta+_ real
d, q isa
_ controller _ regulator
I sq Tb+_ real isb
ω̂m
isc
ψ r* *
I sd id
*
U sd U s*β Tc+_ real
α, β
ψˆ r I sd

u sα , u sβ
θˆr
d, q
a, b, c tan −1 (ψˆ rβ /ψˆ rα )
isα , isβ a, b, c
ψˆ r2α + ψˆ r2β
iˆsα , iˆsβ ,ψˆ rα ,ψˆ rβ , ωˆ m ,τˆl α,β

Figure
Figure6.6.Block
Blockdiagram
diagramofofthe sensorless
the control
sensorless of the
control IM IM
of the drives withwith
drives EKF-based speed
EKF-based estimator.
speed estimator.

The same methodology with the speed control loop is applied to the second loop
for the rotor field control, which initially compares the reference rotor flux ψr∗ with the
estimated flux ψ̂r to produce a d-axis reference current Ids ∗ using the PI controller. I ∗ is
ds
compared to the measured d-axis current Ids results in an error regulated by the PI regulator
to generate d-axis voltage control signal Vds c for producing d-axis reference voltage U ∗ as
ds
computed in (23). From (21) to (23), it is seen that the main control signals strongly depend
on the IM parameters which are relayed on the accurate DT model in this study.

∗ Lr
Iqs = τ∗, (21)
p p Lm ψ̂rd e

∗ ωs p p Lm
Uqs = Vqsc + ( Ids + ψ̂r ω̂m ), (22)
a0 Lr Lr
∗ c ωs L m Rr
Uds = Vds −( Iqs + ψ̂r ), (23)
a0 Lr L2r
The rotor shaft position which is identical to the rotor field position θ̂r is adopted for
measured stator currents transform to dq0 axes based on the Park transformation and the
reference voltages (Uds∗ , U ∗ ) to αβ reference voltages (U ∗ , U ∗ ). These voltage signals from
qs sα sβ
the control system are entered into the SVPWM module to apply the dead-time effect and
take into account the switching devices0 time delays to generate real switching signals for the
+ + +
upper switches of the inverter, which are denoted by Ta_real , Tb_real , and Tc_real , respectively.
Using these practical switching signals and this precise model of the inverter, as explained
before, three real stator voltages based on (8) will be produced and applied to the IM. The
accurate stator voltages are transformed into two voltages (usα , usβ ) using (6) and utilized as
the main inputs along with the measured currents (isα , isβ ) for the EKF estimator.

5. Results
To evaluate the proposed EKF estimator performance in different operation modes, the
drive system is implemented on MATLAB/Simulink software by employing the DT model
of the IM with parameters that are listed in Table 1. At first glance, common ground can be
found between DT and hardware-in-the-loop (HIL) simulations, as they both run under real-
time simulations; however, the essential difference between DT and HIL implementation is
Sensors 2023, 23, 2006 14 of 21

that for the latter one, a software model should be built as an interface with real hardware
to evaluate the performance of its controllers. For DT simulation, we need to create only
a software model of the drive system that is controlled and then prepare it with inputs
and outputs from the controller being tested to identify how well the controller operates
and whether it runs what it is supposed to be performing. Although simulations and DTs
both use digital models to assess the system’s various processes, a DT is a virtual platform,
which makes it considerably richer for study. The difference between digital twins and
simulation is largely a matter of scale: while a simulation typically studies one particular
process, a DT can run any number of useful simulations to study multiple processes.

Table 1. Main parameters of the sensorless IM drive system.

Three-Phase IM Parameters
P [kW] TL [N·m] pp Rs [Ω] Rr [Ω] Ls [H] Lr [H] Lm [H] Jt [kg·m2 ] Bt [N·m/(rad/s)]
4 25 2 1.3177 1.5097 0.1716 0.1716 0.165 0.11 0.01
Three-phase inverter parameters
rD [mΩ] rT [mΩ] VfT [V] VfD [V] Vdc [V]
1 1 0.8 0.8 400

Here, for SVPWM modeling, the switching frequency is set to 10 kHz, and Ts as
sampling time of the algorithm is 1 µs in all simulations. Moreover, the number of time
delays td , ton and to f f are selected as 4, 1, and 1.5 all in µs, respectively. Fixed-step solver
is used in this simulation with ode5 codes. In the simulations, the initial values of all
estimated states and parameters are set to zero for both operation scenarios. To avoid
the computational burden and simplify the determination of covariance matrices via the
trial-and-error method, the Q, R, and initial P matrices in both scenarios are assumed to be
the same and considered as constant in diagonal form for the EKF algorithm as follows:

Q = diag{2e − 9, 2e − 9, 4.62e − 11, 4.62e − 11, 4e − 9, 1e − 5}

R = diag{6.7e − 3, 6.7e − 3}
P0 = diag{1e − 6, 1e − 6, 1e − 6, 1e − 6, 1e − 6, 1e − 6}

To verify the performance of the proposed speed-sensorless IM drive system, the

simulation studies are carried out in two scenarios. These scenarios are simulated for
different speed and load torque profiles based on the values which are presented in Table 2.
According to Table 2, the drive system is subjected to positive low (10 rad/s) and high
(100 rad/s) speeds as well as negative high speed (−100 rad/s) test along with the variable
positive and negative load torque from no-load (0 N·m) to full-load (25 N·m) values.
Figure 7 shows the EKF algorithm estimation of the IM drive for both scenarios when the
motor is subjected to the nine operation modes based on Table 2. In all figures illustrating
the simulation results, the superscripts Actual, Ideal, and EKF indicate the measured states
(come from the physical twin which includes the real IM, inverter, and control system), the
estimated states by the ideal model of the EKF, and the estimated states by the practical
DT model of the EKF, respectively. Moreover, e(∗) denotes the error defined as between the
actual and the estimated states. Owing to the DT model of the EKF estimator, the states are
estimated without any significant errors in both case studies.

Table 2. Different operation modes of the proposed IM drive system.

Mode 1 2 3 4 5 6 7 8 9
t [s] 0–1 1–2 2–2.5 2.5–3.5 3.5–4 4–5 5–5.5 5.5–6 6–7
ω*m [rad/s] 100 −100 −100 50 50 −50 −50 10 10
TL [N·m] 2.5 2.5 25 25 −25 −25 10 10 0
SensorsSensors
2023, 23, x FOR
2023, PEER REVIEW
23, 2006 16 of 22
15 of 21

100
55 m -Actual m -Ideal m -EKF
50
[rad/s]

50
0
2.9 3.2 3.6
-50
m

-100

0 1 2 3 4 5 6 t [s]

(a)
25
[N.m]

0
L

-25 T T T
T

L -Actual L -Ideal L -EKF

0 1 2 3 4 5 6 t [s]

(b)

20
[A]

0
s

-20
i

i i i
s -Actual s -Ideal s -EKF

0 1 2 3 4 5 6 t [s]

(c)
1

0.5
[Web]

0
r

-0.5
r -Actual r -Ideal r -EKF

-1
0 1 2 3 4 5 6 t [s]

(d)
2 e e
-EKF -Ideal
[rad/s]

-2
e

0 1 2 3 4 5 6 t [s]
10
(e)
10
e e
5 TL -EKF TL -Ideal
[N.m]

-5
L
T
e

-10
0 1 2 3 4 5 6 t [s]

(f)

Figure
Figure7. 7.
TheTheEKF
EKFalgorithm
algorithm estimation resultsfor
estimation results for(a)(a) speed,
speed, (b)(b)
loadload torque,
torque, (c) current,
(c) current, (d) rotor
(d) rotor
flux, (e)speed error, and (f) load torque error of the speed-sensorless IM
flux, (e)speed error, and (f) load torque error of the speed-sensorless IM drive. drive.

Table 2.The
different operation
first scenario modes
deals with of
anthe proposed
ideal inverterIM drive
model system.
without considering the switching
devices time delays, and voltage drops them. From Figure 7, the IM is loaded to a variable
Mode
load of minimum1 −25 N·m 2 to maximum
3 254N·m from5 0 to 7 s6(second 7subplot)8when 9
t [s] 0–1 1–2 2–2.5 2.5–3.5 3.5–4 4–5 5–5.5 5.5–6 6–7
ω*m [rad/s] 100 −100 −100 50 50 −50 −50 10 10
TL [N·m] 2.5 2.5 25 25 −25 −25 10 10 0
Sensors 2023, 23, 2006 16 of 21

Sensors 2023, 23, x FOR PEER REVIEW 17

running with variable speed profile (first subplot) based on Table 2. The performance of
the EKF algorithm depends on the predefined noise and measurement covariance matrices
and motor parameters. Based on the speed and load torque error results (two bottom
subplots), the IM statesThe first scenario deals
are precisely with an
estimated ideal
with inverter model
negligible errorswithout
in steady considering
states; the sw
ing devices time delays, and voltage drops them.
however, during transients, the estimation errors are increased because of predefined noise From Figure 7, the IM is loaded
matrices, and it variable load oftominimum
is not possible determine−25 N·m to maximum
appropriate matrices 25 forN·m from 0 tomodes
all operating 7 s (second su
using the trial-and-error method. The transient errors for the IM speed are about at least 2. The p
when running with variable speed profile (first subplot) based on Table
1.5 rad/s and atmance
most 6ofrad/s,
the EKF andalgorithm
for the load depends
torque onaretheabout
predefined
minimum noise0.5 and N·measurement
m and c
maximum 14 N·m. ance matrices and motor parameters. Based on the speed and load torque error r
(two bottom subplots), the IM states are precisely
The second scenario is simulated based on the feasible practical DT model of the estimated with negligible err
steady
inverter with taking into states;
accounthowever,
all timeduring
delaystransients,
and voltage thedrops
estimation errors According
on switches. are increased becau
to the Figure 7, the performance of the digital twinning EKF algorithm is improved to meetmatrices
predefined noise matrices, and it is not possible to determine appropriate
the physical twin operating modes using
model responses withthe some trial-and-error
extra errors and method. The transient
fluctuations errorsand
in currents for the IM
are about at least 1.5 rad/s and at most 6 rad/s, and
rotor fluxes than the ideal one due to the dead-time effect in steady states operations. Basedfor the load torque are about mini
0.5 N·m and maximum 14 N·m.
on the two bottom subplots, the transient errors for the speed are about 2 rad/s to 8 rad/s,
and for the load torque Theare second
aboutscenario
minimum is simulated
1.2 N·m and based on the feasible
maximum 16 N·m.practical
It is clearDT model of t
that
verter with taking into account all time delays and
the error rates in the second scenario with practical model are a little significant due to the voltage drops on switches. Acco
to the Figure
time delays of switching devices. 7, the performance of the digital twinning EKF algorithm is improv
A zoomed partmeetofthe physical
Figure twin modelinresponses
7 is illustrated Figure 8 along with some extra errors measured
with three-phase and fluctuations in
stator currents inrents
bothand rotorand
ideal fluxes than the
practical ideal one
scenarios due to the
wherein very dead-time effect in steady
small distortions are states
observed in DT ations.
EKF-based Based on thewaveforms
current two bottomdue subplots, the transient
to the effect errorsand
of dead-time for voltage
the speed are ab
drops of inverter rad/s to 8 rad/s,
switching and for
devices whichthe load torque are
are marked with about
blackminimum
circles in1.2 theN·m bottomand maximu
N·m. It is clear that the error rates in the second
subplot of Figure 8. To discriminate the performance of two scenarios, estimated stator scenario with practical model are a
significant due to the time delays of switching
currents and rotor flux in stationary frame are more relevant to the point in which the devices.
A zoomed
delays caused by dead-time part of Figure
of switches 7 is illustrated
and voltage drops are in Figure 8 along with three-phase mea
more considerable.
stator currents in both ideal and practical
It should be noted that the measured errors of the system are the reference scenarios wherein very small distortion
to control
regulation; they observed
might occur in DT EKF-based
in various current waveforms
conditions. Their reasonable due tofluctuation
the effect of dead-time
ranges are and vo
set according todrops of inverter
knowledge switching devices
and experience based onwhich are marked
experimental withand
records black circles in the b
available
subplot
measurements/tests. of Figurethe
However, 8. To discriminatemethod
trial-and-error the performance
can be used of two
to setscenarios,
reasonable estimated
fluctuation ranges. To ensure that the method is convergent, it is possible to examine the test in whic
currents and rotor flux in stationary frame are more relevant to the point
delays
system under various caused byTest
conditions. dead-time of switches
results showed that and voltage drops
the proposed method areismore considerable.
convergent.

100.5
m -Actual m -Ideal m -EKF
[rad/s]

100
m

99.5

0.5 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59 t [s]

(a)
T T T
4 L -Actual L -Ideal L -EKF
[N.m]

2
L
T

0
0.5 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59 t [s]

(b)
10 i i i
s -Actual s -Ideal s -EKF
5
[A]

0
-5
s
i

-10
0.5 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59 t [s]

(c)

Figure 8. Cont.
Sensors 2023, 23, 2006Sensors 2023, 23, x FOR PEER REVIEW 17 of 21 18

0.5

[Web]
r -Actual r -Ideal r -EKF

r
-0.5

-1
0.5 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59 t [s]

(d)
5
[A]

0
-Ideal

-5
abcs
i

t [s]
0.5 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59

(e)
5
[A]

0
abcs -EKF

-5
i

t [s]
0.5 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59

(f)

Figure 8. The EKFFigure 8. Theresults

estimation EKF estimation results
in zoomed in zoomed
version version
for (a) speed, (b)for (a) torque,
load speed, (b)
(c) load torque, (c) alpha
alpha-term
current, (d) rotor alpha-term flux, (e) actual stator three-phase currents for the ideal model, ideal
current, (d) rotor alpha-term flux, (e) actual stator three-phase currents for the and model, a
actual stator three-phase currents for the DT model.
(f) actual stator three-phase currents for the DT model.

6. Discussion It should be noted that the measured errors of the system are the reference to co
regulation; they might
Computer aided simulations have occur in various
been effective conditions.
enough Theirdesign
to address reasonable
issuesfluctuation
and ra
are set according to knowledge and experience based on experimental
modeling process; however, they are restricted in some cases due to the systems complexity records and a
able measurements/tests. However, the trial-and-error method can
and the high volume of processed data. These types of simulations can be implemented in be used to set re
ablesimulations
several in-the-loop fluctuation tools
ranges. To ensure
based that the
on the level method
of the is convergent,
product generation it is possible
such as in to exa
the test system under various conditions. Test results showed that
model-in-the-loop (MIL) form, processor-in-the-loop (PIL) evaluation, software-in-the-loopthe proposed me
is convergent.
(SIL) simulation, and HIL. In comparison with the previous simulation tools, DT technology
is a decent alternative and a powerful development in area of digital implementation by
6. Discussion
connecting the virtual and physical entities. DT has already applied in various applications
Computer aerospace,
such as industry, automotive, aided simulations
healthcare,haveandbeenmedical
effectiveevaluations.
enough to addressAlthoughdesign issue
electric drive is modeling
one of theprocess;
present however, they aremost
DT applications, restricted in some
investors cases
in this due
field to the systems com
concentrate
on motor design, ityelectric
and theinterface,
high volume andof processed data. These
sensing/measuring. types of simulations
Concerning an electriccan be impleme
drive
inaseveral
system, it is still in-the-loop
new topic simulations
to investigate withtools based on theAmong
DT technology. level ofthe the simulation
product generation
forms, the HILas in model-in-the-loop
simulation is the most(MIL)similar form,
toolprocessor-in-the-loop (PIL) evaluation,
to the digital twin technology. HIL softwa
the-loop (SIL) simulation, and HIL. In comparison
simulation is a remarkable solution for evaluating a drive system or at most an asset with the previous simulation tool
technology is a decent alternative and a powerful
performance. It also can be employed in designing or implementing the system and development in area of digital im
mentation by connecting the virtual and physical entities.
applied for some fault identification. However, DTs can be applied for a component, entity, DT has already applied in
or the whole driveioussystem
applications
due tosuch theirasmajor
industry, automotive,
ability to deal with aerospace,
an enormous healthcare, and medical
amounts
uations.
of various datasets. OneAlthough
of the best electric drive is one
implementing of the present
solutions in the caseDT ofapplications, most investo
testing electric
this field
drive systems might be aconcentrate
combiningon motor
both DT design,
and HILelectric interface, and sensing/measuring. Con
technologies.
The basic ing an electric of
configuration drive
our system, it is still
digital twin systema new topic toininvestigate
illustrated with DT techno
Figure 1 consists
Among the simulation forms, the HIL simulation
of three main elements: a physical twin (three-phase IM drive system in reak space), is the most similar tool to the d
twin technology. HIL simulation is a remarkable solution
a digital representation model in cyberspace, and the interface entity for transferring for evaluating a drive syste
at most an asset performance. It also can be employed
information to connect all the spaces. The timeline of the simulation running is also in designing or implementin
shown in Figuresystem
1 withand applied
starting from forthe
some fault identification.
collecting measured data However, DTs can
from physical be applied
twin and for a
ending with DTponent, entity,
modeling. Theoroverall
the whole drive
block systemofdue
diagram theto their major
proposed ability to deal with an
speed-sensorless
IM drive systemmous amounts
simulation of various
model is showndatasets.
in FigureOne6,ofandthe the
bestestimation
implementing solutions in the ca
performances
of EKF algorithm testing electric
for both drive
ideal and systems might
practical be a combining
operation scenariosboth are DT and HILfor
examined technologies.
the
Sensors 2023, 23, 2006 18 of 21

same references. The resulting control performances and the related errors for the EKF-
based version of the proposed drive system are shown in Figures 7 and 8. Moreover, the
numerical implementations of the drive system have been run for nine operation modes
under two scenarios. Considering the estimation process and control performances which
are presented in Figures 6–8, the following outcomes can be encapsulated:
• As discussed before, the DT model as a virtual platform makes the practically im-
plementation of the drive system considerably richer for study and improves the
performance of the IM drive system that might be applicable for development and
research purposes.
• Estimation performances of the EKF algorithm for both ideal and DT models of the
inverter at steady-states are close to each other, except for transients in which the DT
model contains some distortion in currents due to the switch’s time delays.
• In some cases, challenging low- and high-step changes in speed and load torque
are applied to evaluate the proposed drive system’s robustness. Nevertheless, the
proposed drive can perfectly address these variations.
• Although the DT model of the EKF algorithm increases the computational burden
because of estimating the states inside the sectors in narrow time intervals, its com-
putational load can be acceptable since it improves estimation performance during
dead times.
• The main contribution of this work can be practically testing the performance of the
speed-sensorless IM drive system based on a digital model in virtual space to use it for
further analysis purposes and as a predefined module for future studies with no need
to high-cost real test setup along with actual implementation problems and challenges.
• In the case of a technical limitation of the proposed digital drive system, it can be
clarified that the most challengeable part of the work is appropriately connecting the
physical model data to digital twin model and transferring correct information (care
about measurement and system noises) using physical sensors.
• All in all, since DTs offer a lot of functions such as data monitoring, fault detection,
optimization, and lifetime prediction, it seems reasonable to employ the DT model of
the EKF algorithm instead of the conventional ideal model one for electrical drives
especially speed-sensorless drive systems.
In short, although DT technology contains various entities, it is still new and not
fully applied for other applications. In case of our proposed digital model of electric drive
system, one of the appropriate applications is use of the model for further future research
projects with no need for the actual test setup and real physical experiments, which leads
to significant reduction in implementation cost.

7. Conclusions
Digital twin is a type of technology that allows the physical world to be represented
in a virtual digital space, and it can be used to help build a Metaverse. This technology is
a product of the latest advances in information technology. This study involves using a
digital twin based on the Extended Kalman Filter (EKF) to simulate and estimate the state
of a sensorless electric drive system without a speed sensor, specifically an induction motor
(IM), under various operating modes. The Metaverse platform is an appealing option
for connecting virtual and physical models through digital simulation using digital twin
technology, as compared to other simulation models. The proposed digital twin model
significantly improves the performance of the IM drive by providing a practical inverter
model, as demonstrated by the estimation results. In contrast to many previous models,
this system takes into account the effects of time delays and voltage drops of inverter
switches to make it more feasible and reduce current distortions. From the results, it is
concluded that the speed and load torque estimation errors, especially in transients for
practical modeling based on DT, are a little higher than the ideal model due to taking into
account of all time delays of switching devices and voltage drops.
Sensors 2023, 23, 2006 19 of 21

Future work will concentrate on: (1) applying the proposed DT based EKF observer to
other types of electric motors such as permanent magnet synchronous motor and brushless
direct current motor; (2) using the parameter estimation along with states estimation to
improve the overall performance of the drive system; (3) testing the proposed drive system
in different engineering problems, such as robotics, navigation, and data fusion; (4) utilizing
inverter dead-time information to accurate evaluation of conduction and switching losses;
and (5) using different kinds of nonlinear Kalman Filters such as the CKF or the UKF.

Author Contributions: Conceptualization, M.E., J.T. and Z.P.; methodology, M.E. and J.T.; software,
M.E.; validation, M.E. and J.T.; formal analysis, M.E.; investigation, M.E.; resources, M.E.; data
curation, M.E.; writing—original draft preparation, M.E.; writing—review and editing, M.E., M.J. and
H.H.-D.; visualization, M.E.; supervision, Z.P.; project administration, Z.P.; funding acquisition, Z.P.
All authors have read and agreed to the published version of the manuscript.
Funding: This research work was supported by the European Union Research, Development and
Education Program Fund under project name “MOBILITY ZČU” and by the Ministry of Education,
Youth and Sports of the Czech Republic under the project SGS-2021-021.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: We will share our research final data link after full revision and publication.
Acknowledgments: The authors would like to thank Jan Michalík, Vice-Dean for International
Relations and Communication of Faculty of Electrical Engineering, for his kind guidance and insight
during our meetings.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Segovia, M.; Garcia-Alfaro, J. Design, modeling and implementation of digital twins. Sensors 2022, 22, 5396. [CrossRef]
2. Jones, D.; Snider, C.; Nassehi, A.; Yon, J.; Hicks, B. characterizing the digital twin: A systematic literature review. CIRP J. Manuf.
Sci. Technol. 2020, 29, 3652. [CrossRef]
3. Lehmann, J.; Schorz, S.; Rache, A.; Haubermann, T.; Radle, M.; Reichwald, J. Establishing reliable research data management by
integrating measurement devices utilizing intelligent digital twins. Sensors 2023, 23, 468. [CrossRef] [PubMed]
4. Fuller, A.; Fan, Z.; Day, C.; Barlow, C. Digital twin: Enabling technologies, challenges and open research. IEEE Access 2020, 8,
108952–108971. [CrossRef]
5. Neghab, H.K.; Jamshidi, M.; Neghab, H.K. Digital twin of a magnetic medical microrobot with stochastic model predictive
controller boosted by machine learning in cyber-physical healthcare systems. Information 2022, 13, 321. [CrossRef]
6. Yang, B.; Yang, S.; Lv, Z.; Wang, F.; Olofssn, T.; Reichwald, J. Application of digital twins and metaverse in the field of fluid
machinery pumps and fans: A review. Sensors 2022, 22, 9294. [CrossRef]
7. Hu, F. Digital twin-driven reconfigurable fixturing optimization for trimming operation of aircraft skins. Aerospace 2022, 9, 154.
[CrossRef]
8. Zaky, M.S.; Metwaly, M.K.; Azazi, H.Z.; Deraz, H.A. A new adaptive SMO for speed estimation of sensorless induction motor
drives at zero and very low frequencies. IEEE Trans. Ind. Electron. 2018, 65, 6901–6911. [CrossRef]
9. Salim, R.; Mansouri, A.; Bendiabdellah, A.; Chekroun, S.; Touam, M. Sensorless passivity based control for induction motor via
an adaptive observer. ISA Trans. 2019, 84, 118–127. [CrossRef]
10. Ebadpour, M.; Sharifian, M.B.B.; Feyzi, M.R. A cost-effective position sensorless control for four-switch three-phase brushless DC
motor drives using single current sensor. Int. Rev. Autom. Control 2011, 4, 386–393.
11. Jamshidi, M.B.; Roshani, S.; Talla, J.; Sharifi-Atashgah, M.S.; Roshani, S.; Peroutka, Z. Cloud-based machine learning techniques
implemented by microsoft azure for designing power amplifiers. In Proceedings of the IEEE 12th Annual Ubiquitous Computing,
Electronics & Mobile Communication Conference (UEMCON), New York, NY, USA, 1–4 December 2021. [CrossRef]
12. Zuo, Y.; Ge, X.; Zheng, Y.; Chen, Y.; Wang, H.; Woldegiorgis, A.T. An adaptive active disturbance rejection control strategy for
speed-sensorless induction motor drives. IEEE Trans. Transp. Electrif. 2022, 8, 3336–3348. [CrossRef]
13. Ebadpour, M.; Sharifian, M.B.B.; Feyzi, M.R. A simple position sensorless control strategy for four-switch three-phase brushless
DC motor drives using single current sensor. In 2011 2nd Power Electronics, Drive Systems and Technologies Conference; IEEE:
Piscatvie, NJ, USA, 2011; pp. 235–240. [CrossRef]
14. Talla, J.; Leu, V.Q.; Šmídl, V.; Peroutka, Z. Adaptive speed control of induction motor drive with inaccurate model. IEEE Trans.
Ind. Electron. 2018, 65, 8532–8542. [CrossRef]
Sensors 2023, 23, 2006 20 of 21

15. Bahloul, M.; Chrifi-Alaoui, L.; Drid, S.; Souissi, M.; Chaabane, M. Robust sensorless vector control of an induction machine using
multiobjective adaptive fuzzy Luenberger observer. ISA Trans. 2018, 74, 144–154. [CrossRef]
16. Yin, Z.; Gao, F.; Zhang, Y.; Du, C.; Li, G.; Sun, X. A review of nonlinear Kalman filter applying to sensorless control for AC motor
drives. CES Trans. Elec. Machines Syst. 2019, 3, 351–362. [CrossRef]
17. Talla, J.; Peroutka, Z.; Blahnik, V.; Streit, L. Rotor and stator resistance estimation of induction motor based on augmented EKF. In
2015 International Conference on Applied Electronics (AE); IEEE: Piscatvie, NJ, USA, 2015; pp. 253–258.
18. Yildiz, R.; Barut, M.; Zerdali, E. A comprehensive comparison of extended and unscented kalman filters for speed-sensorless
control applications of induction motors. IEEE Trans. Ind. Inform. 2020, 16, 6423–6432. [CrossRef]
19. Zarei, J.; Kowsari, E.; Razavi-Far, R. Induction motors fault detection using square-root transformed cubature quadrature kalman
filter. IEEE Trans. Energy Conv. 2019, 34, 870–877. [CrossRef]
20. Yousefi-Darani, A.; Paquet-Durand, O.; Hitzmann, B. The kalman filter for the supervision of cultivation processes. Adv. Biochem.
Eng. Biotechnol. 2021, 177, 95–125. [CrossRef] [PubMed]
21. Zhang, J.; Lee, J. A review on prognostics and health monitoring of li-ion battery. J Power Source 2011, 196, 6007–6014. [CrossRef]
22. Wang, J.; Moreira, J.; Cao, Y.; Gopaluni, B. Time-variant digital twin modeling through the kalman-generalized sparse identification
of nonlinear dynamics. In 2022 American Control Conference (ACC); IEEE: Piscatvie, NJ, USA, 2022; pp. 5217–5222. [CrossRef]
23. Kurvinen, E.; Kutvonen, A.; Ukko, J.; Khadim, Q.; Hagh, Y.S.; Jaiswal, S.; Neisi, N.; Zhidchenko, V.; Kortelainen, J.;
Timperi, M.; et al. Physics-based digital twins merging with machines: Cases of mobile log crane and rotating machine. IEEE
Access 2022, 10, 45962–45978. [CrossRef]
24. Lim, J.-W.; Bu, H.; Cho, Y. Novel dead-time compensation strategy for wide current range in a three- phase inverter. Electronics
2019, 8, 92. [CrossRef]
25. Özkurt, G.; Zerdali, E. Design and implementation of hybrid adaptive extended kalman filter for state estimation of induction
motor. IEEE Trans. Instrum. Meas. 2022, 71, 1–12. [CrossRef]
26. Ebadpour, M.; Sharifian, M.B.B.; Babaei, E. Modeling and synchronized control of dual parallel brushless direct current motors
with single inverter. Comput. Electr. Eng. 2018, 70, 229–242. [CrossRef]
27. Wang, X.; Yang, J.; Han, K.; Wang, W.; Wang, F.-Y. Metaverses and demetaverses: From digital twins in CPS to parallel intelligence
in CPSS. IEEE Intell. Syst. 2022, 37, 97–102. [CrossRef]
28. Daneshfar, F.; Jamshidi, M. An Octonion-Based Nonlinear Echo State Network for Speech Emotion Recognition in Metaverse.
Available online: https://ssrn.com/abstract=4242011 (accessed on 8 October 2022).
29. Shafiei, A.; Jamshidi, M.; Khani, F.; Talla, J.; Peroutka, Z.; Gantassi, R.; Baz, M.; Cheikhrouhou, O.; Hamam, H. A hybrid technique
based on a genetic algorithm for fuzzy multiobjective problems in 5G, internet of things, and mobile edge computing. Math.
Pronlems Eng. 2021, 2021, 1–14. [CrossRef]
30. Chao, Z.; Zhou, G.; Xu, Q.; Zhibo, W.; Han, C.; Wang, Z. A digital twin defined autonomous milling process towards the online
optimal control of milling deformation for thin-walled parts. Int. J. Adv. Manuf. Technol. 2022, 124, 2847–2861. [CrossRef]
31. Sepasgozar, S.M.E. Differentiating digital twin from digital shadow: Elucidating a paradigm shift to expedite a smart, sustainable
built environment. Buildings 2021, 11, 151. [CrossRef]
32. Hu, F. Mutual information-enhanced digital twin promotes vision-guided robotic grasping. Adv. Eng. Inform. 2022, 52, 101562.
[CrossRef]
33. Khalaj, O.; Jamshidi, M.; Hassas, P.; Hosseininezhad, M.; Masek, B.; Stadler, C.; Svoboda, J. Metaverses and AI twinning of 42SiCr
steel alloys. Mathematics 2023, 11, 4. [CrossRef]
34. Han, Y.; Niyato, D.; Leung, C.; Kim, D.I.; Zhu, K.; Feng, S.; Shen, X.; Miao, C. A dynamic hierarchical framework for IoT-assisted
digital twin synchronization in the metaverse. IEEE Internet Things J. 2023, 10, 268–284. [CrossRef]
35. Ebadpour, M.; Talla, J.; Jamshidi, M.B.; Peroutka, Z. EKF digital twinning of induction motor drives for the metaverse. In 2022
20th International Conference on Mechatronics—Mechatronika (ME); IEEE: Piscatvie, NJ, USA, 2022; pp. 1–6. [CrossRef]
36. Jamshidi, M.B.; Ebadpour, M.; Moghani, M.M. Cancer digital twins in metaverse. In 2022 20th International Conference on
Mechatronics—Mechatronika (ME); IEEE: Piscatvie, NJ, USA, 2022; pp. 1–6. [CrossRef]
37. Brandtstaedter, H.; Ludwig, C.; Hübner, L.; Tsouchnika, E.; Jungiewicz, A.; Wever, U. DIGITAL TWINS FOR LARGE ELECTRIC
DRIVE TRAINS. In 2018 Petroleum and Chemical Industry Conference Europe (PCIC Europe); IEEE: Piscatvie, NJ, USA, 2018; pp. 1–5.
[CrossRef]
38. Eigner, M.; Dickopf, T.; Apostolov, H.; Schaefer, P.; Faißt, K.G.; Keßler, A. System Lifecycle Management: Initial Approach
for a Sustainable Product Development Process Based on Methods of Model Based Systems Engineering. In Product Lifecycle
Management for a Global Market. PLM 2014. IFIP Advances in Information and Communication Technology; Springer: Berlin/Heidelberg,
Germany, 2014; p. 442. [CrossRef]
39. Rahman, K.; Rahman, S.; Bhaskar, M.S.; Iqbal, A.; Khandakar, A.; Tariq, M.; Alamri, B. Field-oriented control of five-phase
induction motor fed from space vector modulated matrix converter. IEEE Access 2022, 10, 17996–18007. [CrossRef]
40. Ebadpour, M.; Sharifian, M.B.B. Cascade H-bridge multilevel inverter with low output harmonics for electric/hybrid electric
vehicle applications. Int. Rev. Electr. Eng. (IREE) 2012, 7, 3248–3256.
Sensors 2023, 23, 2006 21 of 21

41. Hannan, M.A.; Ali, J.A.; Ker, P.J.; Mohamed, A.; Lipu, M.S.H.; Hussain, A. switching techniques and intelligent controllers for
induction motor drive: Issues and recommendations. IEEE Access 2018, 6, 47489–47510. [CrossRef]
42. Zhenyu, Y. Space-Vector PWM with TMS320C24x/F24x Using Hardware and Software Determined Switching Patterns. Texas
Instruments Application Report SPRA524. 1999. Available online: https://www.ti.com/lit/an/spra524/spra524.pdf?ts=167585
8747746 (accessed on 13 January 2023).

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.