Speed Sensorless Nonlinear Adaptive Control of

Induction Motor using Combined Speed and

Perturbation Observer
Yaxing Rena,b , Ruotong Wangb,c , Saqib Jamshed Rindb,d , Pingliang Zenga ,
Lin Jiangb,∗∗
a
Hangzhou Dianzi University, Xiasha, Hangzhou, 310018, China
b
Department of Electrical & Electronics Engineering, University of Liverpool, Liverpool,
L69 3BX, UK
c
Shandong Artificial Intelligence Institute, Qilu University of Technology (Shandong
Academy of Sciences), Jinan, 250316, China
d
Department of Automotive and Marine Engineering, NED University of Engineering
and Technology, Karachi, 75270, Pakistan

Abstract

High performance induction motors (IM) require a robust and reliable

speed controller to maintain the speed tracking performance under various
uncertainties and disturbances. This paper presents a sensorless speed con-
troller for IM based on speed and perturbation estimation and compensation.
By defining a lumped perturbation term to include all unmodeled nonlinear
dynamics and external disturbances, two state and perturbation observers
are designed with combining the model reference adaptive system (MRAS)
based speed observer to estimate the flux and speed states and the flux- and
speed-loop related lumped perturbation terms. The estimated flux, speed

∗
This work was funded by ‘2018 Huzhou South Taihu Elite Program Leading Innovation
Team Project’.
∗∗
Corresponding author.
Email address: L.Jiang@liv.ac.uk (Lin Jiang)
and perturbation terms are used to design an output feedback, speed sen-
sorless nonlinear adaptive controller (SSNAC) for IM. The stability of the
closed-loop system is addressed in Lyapunov theory. Effectiveness of the
SSNAC is verified via simulation and experiment tests. Comparing with the
standard vector control plus MRAS speed observer (VC-MRAS), the pro-
posed SSNAC reduces the speed tracking error by 20% to 30% on average
under model uncertainties and unknown load disturbance due to the estima-
tion and compensation of perturbation terms. The combined observer can
estimate the real rotor speed under speed varying and load changes and thus
makes SSNAC achieve high performance robust speed drive without using
speed sensors.
Keywords: Perturbation estimation and compensation, nonlinear adaptive
control, speed sensorless control, combined speed and perturbation
observer, induction motor

1. Introduction

Induction motors (IM) have been widely used as the main power force
in industry such as various household products and industrial applications,
and also for the developed electric vehicle such as Tesla, due to their sim-
plistic construction and reliable performance in harsh environment (Hu et al.
(2014)). However, control of IM is a challenge task due to its highly coupled
nonlinear nature, the unmeasurable rotor side variables and flux linkage and
uncertainties from the parameters and the load (Finch and Giaouris (2008)).
At present, the most widely applied control approaches for IM are the vector
control (VC) and direct torque control (DTC) (Rehman and Xu (2011)). The

2
DTC is a lookup table based control approach that directly adjusts the sta-
tor voltage using space vector technology to avoid the complex coordination
transformation on stator current and therefore results in a rapid response of
torque (Rind et al. (2017); Costa et al. (2019)). The VC, also called field-
oriented control, can successively regulate the magnetic flux amplitude and
the rotor mechanical speed to track their references (Bhowate et al. (2019)).
However, the traditional VC method can only realize the asymptotic input-
output linearisation of the IM system and cannot fully linearise the nonlinear
dynamic of the IM system and the neglected nonlinear coupling can seriously
affect the dynamics of the magnetic flux loop and the speed loop under ex-
ternal disturbances (Marino et al. (1993); Liu et al. (2013)).
IM traction drives require high-performance controllers for fast transient
response as well as energy optimisation. One of the most used control meth-
ods for nonlinear systems is the input-output linearisation control (IOLC) to
achieve complete decoupling of rotor speed and flux linkage. This method
can simultaneously adjust the speed and magnetic flux without ignoring the
coupling nonlinear dynamics between them. Based on this method, some
approaches for IM speed drive have been proposed, such as the adaptive
input-output linearisation control (Marino et al. (1993)), nonlinear precision
feedback linearisation control (Boukas and Habetler (2004)) that can fully
linearise the flux linkage and speed to decouple their dynamics, and nonlinear
model predictive control to solve the optimal control problem (Englert and
Graichen (2020)). The main drawback of IOLC method is that it requires
detailed system models and accurate model parameters. In practical appli-
cations, the system parameters are difficult to obtain accurately, such as the

3
time-varying rotor/stator resistance (Guzinski and Abu-Rub (2013)). In this
way, the robustness and reliability of the IOLC method are greatly reduced.
In response to this problem, researchers have developed disturbance estima-
tion and compensation methods based nonlinear adaptive control (Jiang et al.
(2004); Chen et al. (2019b,a)). These methods use online estimation and com-
pensation the lumped perturbation term that including the nonlinear parts,
external disturbances, and parameter changes. Thus, its performance does
not depend on the accuracy of the IM model. In addition, there are online
parameter estimation techniques (Proca and Keyhani (2007); Ravi Teja et al.
(2012); Kivanc and Ozturk (2018); Yang et al. (2018); Pyrkin et al. (2019);
Perin et al. (2021)), estimating the rotor resistance via stator temperature to
compensate temperature variation (Sung et al. (2012)), sliding mode control
(Wu et al. (2013); Wang et al. (2018a)), combined sliding mode techniques
with MRAS-type estimator (Tarchala and Orlowska-Kowalska (2018); Ho-
lakooie et al. (2019)), fuzzy control (Suetake et al. (2011); Grabowski et al.
(2000)), and auto-disturbance rejection control (Feng et al. (2004); Li et al.
(2012, 2015)), etc.
On the other hand, the traditional IM speed control is based on speed
feedback from encoder or position sensors. However, due to the reliability of
the sensors as well as the electrical noise of the sensor itself, if the speed is
only obtained from the sensors, the overall reliability of the system will be
reduced, such as the impact of sensor fault (Verrelli et al. (2018)). Therefore,
in addition to installing a speed encoder, it is also considered to use sensorless
control that estimates the rotor speed in high-performance IM and EV ap-
plications (Holtz (2005)). Due to the rapid development of power electronics

4
technology and digital signal processor performance, the sensorless control
method of IM becomes increasing more feasible. There have been a lot of re-
search results on the speed estimation of sensorless control, such as estimated
rotor position and velocity calculated from stationary frame (Nasiri (2007)),
rotor flux observer based algorithm (Marchesoni et al. (2020)), optimization-
based position sensorless control (Callegaro et al. (2018)), model reference
adaptive system (MRAS) speed observer (Schauder (1992); Ohyama et al.
(2005)), MRAS-fuzzy logic observer (Gadoue et al. (2010)), easy implement
PLL-like sensorless observer (Tilli and Conficoni (2016)), sliding mode tech-
niques based speed observer (Li et al. (2005); Zhang (2013); Zaky et al.
(2018)), extended Kalman filter for speed estimation (Habibullah and Lu
(2015)), artificial neural network speed observer (Sun et al. (2013)), and ap-
proximate high gain observer (Wang et al. (2018b)), etc. The MRAS speed
observer based on rotor flux equations is the most preferable scheme due
to its easy implementation and clear physical meaning (Yang et al. (2018)).
Most MRAS speed observer is using the rated parameters of the IM and
assuming the parameters are constant. To solve this issue, some parameter
estimation approaches have been proposed to deal with unknown constant
or slow-varying parameters (Marino et al. (2005, 2008); Tilli and Conficoni
(2016)). But these algorithms work under the constant or slow-varying speed
conditions which is not quite suitable in EV application. Moreover, the
disturbance observer based control algorithm uses a state observer to esti-
mate the lumped nonlinear terms and disturbances, while the speed observer
estimates the rotor speed. Then it will raise a question that, if the distur-
bance observer based control algorithm is combined with the speed sensorless

5
method, whether the speed estimator affects the robustness and control per-
formance of the control algorithm. Another question is whether they can be
combined in some way to estimate speed and disturbance at the same time,
thereby reducing the amount of calculation.
In this paper, a speed sensorless nonlinear adaptive controller (SSNAC)
has been proposed to integrate the nonlinear control techniques and tradi-
tional speed estimation techniques to provide a nonlinear sensorless drive
of an IM. The SSNAC is designed using the MRAS speed observer to es-
timate the rotor speed and two state and perturbation observers (SPO) to
estimate the perturbation terms of the flux and speed loop. Moreover, a
combined speed and perturbation observer (CSPO) is proposed to estimate
the speed and perturbation terms simultaneously. It reduced the usage of
a PI regulator in the traditional MRAS speed estimator for the reduction
of computational load. The estimated perturbation term from the CSPO is
used to compensate the real value of speed perturbation in order to reduce
the dependency of accurate model and parameters for improved robustness
of the control method. The stability of closed-loop system that integrates
the nonlinear control techniques and speed estimation with a combined speed
and perturbation observer is proved using Lyapunov theory.
The remainder of this paper is organised as follows. Section 2 presents
the IM dynamic model in the d-q frame. Section 3 presents the design of the
proposed SSNAC controller. The stability proof of the closed-loop system
is presented in Section 4. The effective application of NAC is validated in
simulation and presented in Section 5 and validated experimentally in Section
6. Finally, the paper concludes in Section 7.

6
2. Model of Induction Motor

A three-phase would-rotor IM can be transformed into rotating frame in

d-q axis via Park transform with the d-axis synchronize with the angle of
rotor flux θr . Then the dynamic model of an IM can be presented as follows
(Bimal (2002)):
·
x = f (x) + Gu (1)

where
h iT
x= isd isq ψrd ψrq ωm

f (x) =
  
Rr L2m

Rs Rr Lm ωr Lm
 − σL + σL L2 isd + ωe isq + σL L2 ψrd + σL L ψrq 
  s s r s r s r 
2

 − Rs R L
r m ω L
r m R L
r m

+ isq − ωe isd − ψrd + ψrq 

 σLs σLs L2r σLs Lr σLs L2r 

 Rr Rr Lm 
 − ψrd + (ωe − ωr )ψrq + isd 

 Lr Lr 

 Rr Rr Lm 
 − ψrq − (ωe − ωr )ψrd + isq 

 Lr Lr 

 3P Lm TL 
(ψrd isq − ψrq isd ) −
2JLr J

T
1

0 0 0 0
 σLs
G=

1 
0 0 0 0
σLs

h iT h iT
u= u1 u2 = vsd vsq

where x is the states vector of IM system with five state variables, isd and
isq are the stator current in d-q frame, ψrd and ψrq are the rotor flux linkage,

7
ωm is the mechanical rotor speed. f (x) is the nonlinear function of state
dynamics. Rs , Rr , Ls and Lr are the resistances and inductances of stator and
rotor; Lm is the mutual inductance between stator and rotor phase winding;
σ = 1 − L2m /Ls Lr is the parameter that indicates the leakage factor. ωe is
the synchronous speed of stator current; ωr is the rotor speed in electrical
that satisfies ωr = P ωm , P is the pole pairs of IM. J is the rotor inertia and
TL is the load torque which is seen as a disturbance. G is the matrix gain of
system inputs. u is the control inputs of IM system with two input variables,
vsd and vsq , the stator voltage inputs. The system output is defined in a
vector y, which is presented as

y = [ |ψr | ωm ]T (2)

where
q
2 2
|ψr | = ψrd + ψrq (3)

and the rotor currents ird and irq can be expressed as


1 L
 ird = ψrd − m isd


Ls Lr (4)
1 L m
 irq = ψrq − isq


Ls Lr

3. Speed sensorless nonlinear adaptive controller

This section introduces the design steps of the perturbation observer

based nonlinear adaptive controller (NAC) and combines it with the rotor-
flux based MRAS speed observer to become a speed sensorless control system
for IM. The NAC design uses the input-output linearisation to fully decou-
ple the interaction between the flux linkage and speed subsystems. Then

8
the system nonlinearities, disturbances, and parameter uncertainties are de-
fined as perturbation terms (Ren et al. (2016)). After that, two state and
perturbation observers (SPOs), one for the flux linkage loop and the other
for the speed loop, are designed to estimate their perturbation terms. The
speed sensorless feedback controller is fed by the speed estimation from the
MRAS speed observer, which uses a PI regulator to eliminate the estimation
error. Moreover, the speed SPO and MRAS speed observer are combined
to a speed and perturbation observer that is able to estimate both the rotor
speed and its perturbation term simultaneously and the usage of PI regulator
in traditional MRAS is then removed.

3.1. Input-output linearisation

In the d − q frame dynamic model of IM in (1), the d-axis is aligned
with the direction of rotor flux. Then it can be assumed that ψrd = ψr , and
ψrq = 0 in steady state (Bimal (2002)).
Differentiate the system output y until the system inputs is separated
with other states, the input-output relationship of the IM system is obtained
as Chen et al. (2014):
     
ÿ1 Lf 1 (x) u1
 =  + B(x)   (5)
ÿ2 Lf 2 (x) u2

R2
 
Lm Rr Rr Rs Lm Rr
Lf 1 (x) = r2 ψr (x) − + isd (x) + ωe isq (x) (6)
σLr σLr Lr Ls Lr
  
3P Lm Rs Rr
Lf 2 (x) = − + ψr (x)isq (x) − ωe ψr (x)isd (x)
2JLr σLs σLr

Lm Rr Lm 2 1
+ isd (x)isq (x) − ωr (x)ψr (x) − ṪL (x) (7)
Lr σLs Lr J

9
 Lm Rr
   
B1 (x) 0
B(x) =  = σLs Lr 
(8)
 3P Lm ψr (x) 
B2 (x) 0
2JσLs Lr
where Lf 1 (x) and Lf 2 (x) represent the system nonlinearities which are func-
tion of system states and parameters. As during the normal operation ψr 6= 0,
|B(x)| =
6 0 and B(x) is a non-singular matrix, thus the feedback linearising
control can be obtained as
     
u1 −L f1 (x) υ
  = B −1 (x)   +  1  , (9)
u2 −Lf 2 (x) υ2
where υ1 and υ2 are the control inputs of the linear system
   
ÿ υ
 1  =  1 . (10)
ÿ2 υ2
3.2. Design of state and perturbation observer design
The first step of design the state and perturbation observer (SPO) is to
define the perturbation terms. The value of B(x) is time varying due to
the changes of states and parameter uncertainties, such as J and τr . De-
fine B(x) = B0 + ∆B, B0 is the nominal value of B(x) and calculated by
using the nominal value of rotor flux ψr0 and all parameters, and ∆B rep-
resents uncertainties of the control gain matrix caused from the parameter
uncertainties and operation point changes. For example, the rotor resistance
can be described as the rated resistance plus the varying of rotor resistance
as Rr = Rr0 + ∆Rr . Moreover, to simplify the controller design, it is as-
sumed that all system dynamic Lf 1 (x) and Lf 2 (x) are unknown as well. The
perturbation terms Ψ1 and Ψ2 are defined as
     
Ψ (x) L (x) u
 1  =  f1  + (B(x) − B0 )  1  (11)
Ψ2 (x) Lf 2 (x) u2

10
 Thus, the perturbation terms include the system nonlinearities, external
disturbances TL and parameter uncertainties.
Then (5) can be rewritten as:
     
ÿ Ψ (x) u
 1 = 1  + B0  1  (12)
ÿ2 Ψ2 (x) u2
Define new state variables as zi1 = yi , zi2 = ẏi , and a fictitious state to
represent the perturbation terms as zi3 = Ψi for flux subsystem and speed
subsystem, system (12) can be represented as two subsystems as:




 żi1 = zi2


 żi2 = zi3 + B0i ui , i = 1, 2 (13)



 żi3 = Ψ̇i


Several types of SPO have been proposed for estimating the perturbation
terms, such as linear and high-gain observers (Chen et al. (2014)), sliding
mode observer (Jiang and Wu (2002)) and nonlinear observer (Han (2009)).
In order to achieve the simplest structure and applicability, in this paper
SPOs are designed via linear Luenberger observer. The SPO can be designed
as: 
ẑ˙i1 = ẑi2 + li1 zei1






 ẑ˙i2 = ẑi3 + B0i ui + li2 zei1 , i = 1, 2 (14)


 ẑ˙i3 = li3 zei1



where ẑij are the estimation of zij , and zei1 = zi1 − ẑi1 are defined as the state
estimation error. The observer gains li1 , li2 , li3 can be parametrised following
method (Yoo et al. (2007)):
h i j k
li1 li2 li3 = 3α0 3α02 α03 (15)

11
where α0 is the observer bandwidth and a tuning parameter. The tuning of
α0 should ensure that the observer response is faster than the controlled plant
(Yoo et al. (2007)). This leads the observer with larger bandwidth than the
bandwidth of the controller plant. On the other hand, the observer should
be applied with a lower bandwidth to filter out the measurement sensor noise
(Ellis (2002)). Therefore, the observer bandwidth setting should compromise
the requirement considering the actual bandwidth of both controller and
sensor noise.

3.3. Design of speed observer

3.3.1. Conventional MRAS based speed observer

In the MRAS based speed observer (SO) based on rotor flux equations, the
reference model is the rotor flux from the real model calculated by the current
feedback, while the adaptive model tracks the rotor flux of the reference
model via adjusting the rotor speed (Gadoue et al. (2010)). When the rotor
flux difference between adaptive model and reference model eliminated to
zero, it is known that the speed estimation is the same with the real speed.
The rotor flux dynamics of the reference model can be presented as:

dψrα Lr Lr Rs σLs Lr disα

= vsα − isα − (16)
dt Lm Lm Lm dt
dψrβ Lr Lr Rs σLs Lr disβ
= vsβ − isβ − (17)
dt Lm Lm Lm dt

where the variables vsα , vsβ , isα and isβ are the voltage control inputs and
current feedback of real IM.
The rotor flux dynamics of adaptive model is designed to estimate the

12
rotor flux variables using rotor circuit equations as (Gadoue et al. (2010)):

dψ̂rα Lm 1
= isα − ψ̂rα − P ω̂m ψ̂rβ (18)
dt τr τr
dψ̂rβ Lm 1
= isβ + P ω̂m ψ̂rα − ψ̂rβ (19)
dt τr τr

where ψ̂rα , ψ̂rβ and ω̂m are the estimation of rotor flux ψrα and ψrβ and
mechanical rotor speed ωm , respectively.
The speed tuning signal  is defined as the difference between the rotor
fluxes imaginary components from the reference model and adaptive model
as Gadoue et al. (2010):

 = ψrβ ψ̂rα − ψrα ψ̂rβ (20)

Then the adaptation mechanism uses a PI regulator to adjust the esti-

mated rotor speed as: Z
ω̂m = kP  + kI  dt (21)

In the MRAS-SO, as shown in the left side of Figure 1(a), when the error
 convergences to zero, the estimated rotor speed ω̂m reaches the real speed.
Then the estimated rotor speed and rotor flux is used to feed the two SPOs
in (14) for perturbation estimation. In the whole controller design, two SPOs
are used for state and perturbation estimation of rotor speed and rotor flux
in different loops.

3.3.2. Combined speed and perturbation observer

In previous section, a MRAS-SO is designed to estimate the rotor speed
and a speed SPO is designed to estimate the speed loop perturbation by
using the estimated rotor speed, as shown in the Figure 1 (a). From the

13
 (a)

(b)
Figure 1: Block diagrams of (a) a traditional MRAS based speed observer and speed
perturbation observer, (b) a combined speed and perturbation observer (CSPO).

0
block diagram, the rotor speed needs to be estimated twice, ω̂m and ω̂m , in
different observers. This undoubtedly increases the computational load.
This section combines the MRAS-SO and speed SPO and proposes a
combined speed and perturbation observer (CSPO) that estimates the speed
and its perturbation term at the same time, as shown in Figure 1 (b). The
motivation of combining the MRAS-SO and speed SPO into a CSPO aims to
simplify the work in speed estimation and its perturbation estimation into a
simpler define, that when  → 0, the estimated speed from CSPO tends to the

14
 0
real rotor speed as ω̂m → ωm , and its estimated perturbation Ψ̂ tends to its
real value at the same time. But in the CSPO, the actual speed signal z21 is
assuming unknown, and the standard observation error ze21 is not achievable.
Thus, it is required to find another driving signal to replace the ze21 in original
speed SPO.
In the original speed SPO, the observer output is driven by the observa-
tion error ze21 = z21 − ẑ21 , where z21 comes from the output of PI regulator in
the MRAS-SO. In order to simplify the proof, the output of PI is simplified
with the product of proportional gain and tuning signal as z21 = kP . Then
speed SPO from equation (14) with i = 2 can be rewritten as

ẑ˙21 = ẑ22 + l21 (kP  − ẑ21 )






 ẑ˙22 = ẑ23 + B02 u2 + l22 (kP  − ẑ21 ) (22)


 ẑ˙23 = l23 (kP  − ẑ21 )



The above equation can be represented in matrix form as

        
ẑ˙21 −l21 1 0 ẑ21 kP l21 0
        
 ˙  
 ẑ22  =  −l22 0 1   ẑ22  +  kP l22   +  B02 u2 (23)
     

        
˙ẑ23 −l23 0 0 ẑ23 kP l23 0

Thus, it proves that the tuning signal  can be used as the driving signal
when ze21 is not available. If using Eq. (23) as the CSPO, it will achieve
the same performance with MRAS-SO and speed SPO. But this paper aims
to reduce the complexity of the direct combination of MRAS-SO and speed
SPO. In order to simplify the CSPO design, the proposed CSPO substituted
the observation error ze21 with the speed tuning signal  from (20). Then the

15
CSPO can be represented as

ẑ˙21 = ẑ22 + l21 






 ẑ˙22 = ẑ23 + B02 u2 + l22  (24)


 ẑ˙23 = l23 



Comparing among (14), (23) and (24), the CSPO maintains the simplicity
of the original SPO, utilizes the tuning signal  to drive the estimation of rotor
speed, and replaces the PI regulator in the original MRAS system. In order
to verify whether this change is stable, the stability analysis of its closed-loop
system will be carried out in later sections.
In the separated approach, the parameters of MRAS-SO and speed SPO
are calculated separately. But the parameterisation of CSPO should take into
account the stability of both speed estimation and perturbation observation,
which is more difficult than the separate MRAS-SO and speed SPO approach.
In order to add more degree of freedom in the tuning of system parameters,
an additional proportional gain is added to the estimation of rotor speed as

ω̂m = ẑ21 + l20  (25)

The final transfer function of the rotor speed estimation can be presented as
 
B02 u2 l23 l22 l21
ω̂m = + + 2 + + l20  (26)
s2 s3 s s

The gains of l20 , l21 , l22 and l23 can be achieved by pole placement of the
closed-loop system by substituting (25) into system (16) to (19).
Then the CSPO consists of equation (24) and (25) to estimate the rotor
speed as well as its perturbation term simultaneously. After the speed tuning

16
signal  converges to zero, the estimation state ω̂m tracks the mechanical
rotor speed and ẑ23 tracks the rotor speed perturbation for speed control and
linearising the IM system.

3.4. Speed sensorless nonlinear adaptive controller

In the flux subsystem, the original SPO (14) is used to estimate its pertur-
bation term ẑ13 . In the speed subsystem, the CSPO (24) is used to estimate
both the rotor speed ω̂m and its perturbation term ẑ23 . Note that the speed
can be estimated by a conventional MRAS-SO defined in (16) to (21) and the
speed loop perturbation be estimated by a speed SPO defined in (14). Then
the control law of the speed sensorless nonlinear adaptive control (SSNAC)
can be obtained as:
     
u1 υ1 ẑ13
  = B0−1  −  (27)
u2 υ2 ẑ23

Substituting (27) into (12), the system can be linearised as

       
ÿ1 Ψ1 (x) ẑ13 υ1
  =  −  +  
ÿ2 Ψ2 (x) ẑ υ2
    23 (28)
υ1 ze
=  +  13 
υ2 ze23

where zei3 indicates the SPO estimation error of perturbation term zi3 as
zei3 = Ψi (·) − ẑi3 .
By compensating the system nonlinear dynamics, parameter uncertainties
and external disturbances by the estimated lumped perturbation terms, the
linearised system can be easily controlled by lots of developed linear control

17
law. The linear control law υ1 and υ2 can be designed as

 υ = z̈ ∗ + k (z ∗ − z ) + k (ż ∗ − ẑ )
1 11 11 11 11 12 11 12
(29)
 υ = z̈ ∗ + k (z ∗ − z ) + k (ż ∗ − ẑ )
2 21 21 21 21 22 21 22

∗ ∗
where zi1 and żi1 are the reference and its derivative of system outputs;
ki1 and ki2 are the gains of linear control law that are determined by pole
placement defined as ki1 = n2 and ki2 = 2n for a given pole location n. The
location of poles can be determined for a second-order linear system with a
given transient dynamic requirement.
The stator voltages, known as the system control inputs, are calculated
from the proposed controller using variables with physical meaning as

σLs τr h ∗ ∗
i

 vsd = k11 (ψr − ẑ11 ) + k12 (ψ̇r − ẑ12 ) − ẑ13
Lm



2JσLs Lr h  ∗
  
vsq = k21 ωm − ẑ21 − l20 ψrβ ψ̂rα − ψrα ψ̂rβ (30)


 3P L m ψ r0

∗

 +k22 (ω̇m − ẑ22 ) − ẑ23 ]

The final control scheme of the SSNAC for IM are shown in Figure 2.
The SSNAC include two control schemes, the first one is a MRAS-SO/SPOs
based SSNAC that includes a conventional MRAS-SO, two SPOs and two
control loops for flux and speed subsystem, respectively. The second control
scheme is a CSPO based SSNAC that includes the flux SPO, the speed CSPO
for and two control loops for flux and speed tracking.

4. Closed-loop Stability Analysis

The previous section proposed two control approaches, the MRAS/SPO

based SSNAC and the CSPO based SSNAC. In this section, the stability

18
 Figure 2: The control scheme of the SSNAC for IM.

analysis of two control schemes has been provided. Section 4.1 and Section
4.2 present the closed-loop stability proof of both control schemes of SSNAC
using Lyapunov theory.

4.1. Closed-loop stability of the MRAS/SPO based SSNAC

The closed-loop system includes five subsystems: the MRAS speed ob-
server defined in (16) to (21), the flux and speed perturbation observers
defined in (14), and their tracking errors in the IM system.
First, the error dynamic of the MRAS speed observer is obtained with
the rotor flux-based speed tuning signal  defined in (20), whose derivative
is calculated from the rotor circuit equations as

˙ = −a1  + a2 (ωm − ω̂m ) + a3 (31)

19
where
2
a1 = ,
τr
a2 = P (ψrα ψ̂rα + ψrβ ψ̂rβ ), (32)
Lm Lm
a3 = (ψrβ − ψ̂rβ )isα − (ψrα − ψ̂rα )isβ
τr τr
Combining the adaptation mechanism in (21), the dynamic of speed tun-
ing signal can be rewritten as

˙ = −h1  + λ (33)

where

h1 = a1 + kP a2 (34)
Z
λ = a2 ωm + a3 − kI  dt (35)

where h1 is a positive value as a1 > 0 and a2 > 0 (as proved in the Appendix
A) and kP > 0 defined in (21).
Second, define the estimation error of the flux SPO and the speed SPO
as zei1 = zi1 − ẑi1 , zei2 = zi2 − ẑi2 , and zei3 = Ψi (·) − ẑi3 , where i = 1 indicating
the flux SPO and i = 2 indicating the speed SPO. The error dynamic of flux
SPO and speed SPO are obtained from (13) and (14) as
      
˙zei1 −l 1 0 ze 0
   i1   i1   
 ˙  
 zei2  =  −li2 0 1   zei2  +  0 (36)
   

      
˙zei3 −li3 0 0 zei3 Ψ̇1 (·)

Then (36) can be rewritten in matrix format as

[ze˙ i ] = [Ai ][e

zi ] + [ηi ] (37)

20
where [Ai ] is the non-singular observer parameter matrix for both SPOs
and [e
zi ] = [e
zi1 zei2 zei3 ]T is the vector of estimation error with and [ηi ] =
[0 0 Ψ̇i ]T indicate the vector of derivatives of perturbation terms.
The perturbation terms Ψi depend on the system parameters and varia-
tions, system input and external disturbance. In the IM, the system parame-
ters and their variations have to be bounded. The system inputs are given by
the control outputs which are designed to be bounded. The external torque
disturbance cannot be guaranteed to be bounded, but infinite external torque
disturbance is out of the scope of the controller design. Due to these, the
derivatives of the perturbation terms Ψ̇i are considered as bounded in this
paper as a precondition.
Third, define the system output tracking error as ei1 = yi∗ − zi1 and
ei2 = ẏi∗ −zi2 , where i = 1 indicating the flux control loop and i = 2 indicating
the speed control loop. The control law in (29) can then be represented as:

υi = ki1 (yi∗ − zi1 + zi1 − ẑi1 ) + ki2 (ẏi∗ − zi2 + zi2 − ẑi2 )

= ki1 (ei1 + zei1 ) + ki2 (ei2 + zei2 ) (38)

From (12) and (38), the closed-loop tracking error dynamics are obtained
as       
ėi1 0 1 ei1 0
 =   +  (39)
ėi2 −ki1 −ki2 ei2 −ξi
where  
zei1
h i  
ξi = ×  zei2 (40)
 
ki1 ki2 1 
 
zei3
which indicates the lumped estimation error from (36).

21
 Similarly, rewrite system (39) in the matrix format as

[ėi ] = [Mi ][ei ] + [Ki ][e

zi ] (41)

where [ei ] = [ei1 ei2 ]T indicates the tracking error of the closed-loop sys-
tem; [Mi ] is the non-singular controller parameter matrix of flux and speed
subsystems; [Ki ] is the matrix depending on control gains.
For the stability proof, assuming that the system input gain B(x) and
its derivative are bounded and B(x) is non-singular when ψr 6= 0; and the
perturbation terms Ψi (x, t) and their derivatives Ψ̇i (x, t) are bounded. This
assumption is reasonable based on the fact that the target system is a physical
system containing mechanical and electrical processes, so its variables will be
within a suitable range instead of infinite values. Then the closed-loop error
system can be proved to be globally uniformly ultimately bounded (GUUB)
as the following theorem.

Theorem 1. Consider the IM system (1) equipped the proposed SSNAC (30)
with the MRAS-SO in (16) to (21) and two SPOs in (14). If the perturbation
terms Ψi (x, t) defined in (11) satisfying

 kΨ̇ (x, t)k ≤ γ
1 1
(42)
 kΨ̇ (x, t)k ≤ γ
2 2

then the estimation error in (33) and (36) and the tracking error in (41) are

22
GUUB, i.e.,



 |(t)| ≤ P1 |λ|


k[e
z1 ](t)k ≤ 2γ1 k[P21 ]k





k[e
z2 ](t)k ≤ 2γ2 k[P22 ]k , ∀t ≥ T (43)


 k[e1 ](t)k ≤ 4γ1 k[K1 ]kk[P21 ]kk[P31 ]k





 k[e ](t)k ≤ 4γ k[K ]kk[P ]kk[P ]k

2 2 2 22 32

Proof. For the error variable  in (33), choose a Lyapunov function as

1
Vso () = P1 2 . The h1 defined in (34) is a positive value, the P1 can be set
2
as a positive value of P1 = h−1
1 .

For the SPOs estimation error [e

zi ] in (37), choose the Lyapunov function
as Vspo,i ([e zi ]T [P2i ][e
zi ]) = [e zi ]. The gains of SPOs (14) are determined by (15),
which means [Ai ] is Hurwitz. Then a feasible positive solution [P2i ] can be
found from the Riccati equation [Ai ]T [P2i ] + [P2i ][Ai ] = −I.
For the tracking error [ei ] in (41), define a Lyapunov function Vt,i ([ei ]) =
[ei ]T [P3i ][ei ] and a [P3i ] as a feasible positive solution from the Riccati equa-
tion [Mi ]T [P3i ] + [P3i ][Mi ] = −I.
The Lyapunov function of closed-loop system can be defined as the sum of
MRAS speed observer, two SPOs, and two control loops as V (, [e
z1 ], [e
z2 ], [e1 ], [e2 ]) =
Vso () + Vspo,1 ([e
z1 ]) + Vspo,2 ([e
z2 ]) + Vt,1 ([e1 ]) + Vt,2 ([e2 ]). Then the derivative

23
of V can be obtained as

V̇ z1 ]T ([A1 ]T [P21 ] + [P21 ][A1 ])[e

= P1 (−h1  + λ) + [e z1 ]
z2 ]T ([A2 ]T [P22 ] + [P22 ][A2 ])[e
+[e z2 ]
+[e1 ]T ([M1 ]T [P31 ] + [P31 ][M1 ])[e1 ]
+[e2 ]T ([M2 ]T [P32 ] + [P32 ][M2 ])[e2 ]
≤ −2 + P1 |||λ| − k[e
z1 ]k2 + 2k[e z2 ]k2
z1 ]kk[η1 ]kk[P21 ]k − k[e
z2 ]kk[η2 ]kk[P22 ]k − k[e1 ]k2 + 2k[e1 ]kk[K1 ]kk[e
+2k[e z1 ]kk[P31 ]k (44)
−k[e2 ]k2 + 2k[e2 ]kk[K2 ]kk[e
z2 ]kk[P32 ]k
≤ −|| (|| − P1 |λ|) − k[e z1 ]k − 2γ1 k[P21 ]k)
z1 ]k (k[e
−k[e z2 ]k − 2γ2 k[P22 ]k)
z2 ]k (k[e
−k[e1 ]k (k[e1 ]k − 2k[K1 ]kk[e
z1 ]kk[P31 ]k)
−k[e2 ]k (k[e2 ]k − 2k[K2 ]kk[e
z2 ]kk[P32 ]k)

Each Lyapunov function in (44) has its own close-loop. In the first section,
as λ defined in (35) is bounded (as proved in Appendix B), when || > P1 |λ|,
the −|| (|| − P1 |λ|) ≤ 0 can be obtained. Thus, there exists a time T1 > 0
to satisfy
|(t)| ≤ P1 |λ|, ∀t ≥ T1 (45)

Similarly, there exists T2 and T3 to satisfy

k[e
z1 ](t)k ≤ 2γ1 kP21 k, ∀t ≥ T2 (46)

k[e
z2 ](t)k ≤ 2γ2 kP22 k, ∀t ≥ T3 (47)

and T3 and T4 to satisfy

k[e1 ](t)k ≤ 4γ1 kK1 kkP21 kkP31 k, ∀t ≥ T4 (48)

k[e2 ](t)k ≤ 4γ2 kK2 kkP22 kkP32 k, ∀t ≥ T5 (49)

24
 Considering (45) to (49) to give T = max (T1 , T2 , T3 , T4 , T5 ) to lead (43).
It proved that the closed-loop system of using the separate SO and speed
SPO is stable and the speed estimation error, the flux and speed perturbation
estimation error and tracking error are bounded.

4.2. Closed-loop stability of the CSPO based SSNAC

In another SSNAC approach that uses the CSPO, its dynamics is obtained
from the MRAS error defined in (33) to (35) and speed SPO in (36) into a
single matrix as
      
˙ −a1 − l20 a2 a2 0 0  a3
      
 ˙  
−l21
   
 ze21   0 1 0   ze21   0 
 ˙ =
     +  (50)
−l22
   
 ze22   0 0 1   ze22   0 
      
ze˙ 23 −l23 0 0 0 ze23 Ψ̇2 (·)

where a1 and a2 are positive values defined in (32) and −a1 − l20 a2 can be
adjusted by gain l20 .
The error system (50) can be rewritten into matrix form as

[ze˙ 3 ] = [Λ][e
z3 ] + [δ] (51)

With the new estimation error (51), the (41) of speed subsystem can be
rewritten as
[ė2 ] = [M2 ][e2 ] + [K3 ][e
z3 ] (52)

where
h i
[K3 ] = 0 k21 k22 1 (53)

The following theorem is summarised.

25
Theorem 2. Consider the IM system (1) equipped the proposed SSNAC
(30) with a flux SPO in (14) and a CSPO in (24). If the perturbation terms
Ψi (x, t) defined in (11) is bounded as in (42), then the estimation error in
(50) and the tracking error in (41) are GUUB, i.e.,



 k[e
z1 ](t)k ≤ 2γ1 k[P21 ]k


 k[e

z ](t)k ≤ 2γ k[P ]k
3 2 4
, ∀t ≥ T̄ (54)


 k[e1 ](t)k ≤ 4γ1 k[K1 ]kk[P21 ]kk[P31 ]k


 k[e ](t)k ≤ 4γ k[K ]kk[P ]kk[P ]k

2 3 3 4 32

where γ3 = max {a3 , γ2 }.

Proof. Define the Lyapunov function for (51) as Vcspo ([e z3 ]T [P4 ][e
z3 ]) = [e z3 ],
where [P4 ] is a feasible positive solution from the Riccati equation [Λ]T [P4 ] +
[P4 ][Λ] = −I.
The Lyapunov function of closed-loop system can be defined as the sum
of the flux SPO, CSPO and two control loops as V̄ ([e
z1 ], [e
z3 ], [e1 ], [e2 ]) =
Vspo,1 + Vcspo ([e
z3 ]) + Vt,1 ([e1 ]) + Vt,2 ([e2 ]). Then the derivative of V̄ can be

26
obtained as

V̄˙ z1 ]T ([A1 ]T [P21 ] + [P21 ][A1 ])[e

= [e z1 ]
z3 ]T ([Λ]T [P4 ] + [P4 ][Λ])[e
+[e z3 ]
+[e1 ]T ([M1 ]T [P31 ] + [P31 ][M1 ])[e1 ]
+[e2 ]T ([M2 ]T [P32 ] + [P32 ][M2 ])[e2 ]
z1 ]k2 + 2k[e
≤ −k[e z1 ]kk[η]kk[P21 ]k
z3 ]k2 + 2k[e
−k[e z3 ]kk[δ]kk[P4 ]k
(55)
−k[e1 ]k2 + 2k[e1 ]kk[K1 ]kk[e
z1 ]kk[P31 ]k
−k[e2 ]k2 + 2k[e2 ]kk[K3 ]kk[e
z3 ]kk[P32 ]k
≤ −k[e z1 ]k − 2γ1 k[P21 ]k)
z1 ]k (k[e
−k[e z3 ]k − 2γ3 k[P4 ]k)
z3 ]k (k[e
−k[e1 ]k (k[e1 ]k − 2k[K1 ]kk[e
z1 ]kk[P31 ]k)
−k[e2 ]k (k[e2 ]k − 2k[K3 ]kk[e
z3 ]kk[P32 ]k)

Similarly, there exists T̄3 to satisfy

k[e
z3 ](t)k ≤ 2γ3 kP4 k, ∀t ≥ T̄3 (56)

and T̄5 to satisfy

k[e2 ](t)k ≤ 4γ3 kK3 kkP4 kkP32 k, ∀t ≥ T̄5 (57)

Considering (46), (48), (56) and (57) to give T̄ = max T2 , T̄3 , T4 , T̄5 to
lead (54).
This verifies that the closed-loop system is stable and the estimation
error of flux SPO and speed CSPO and tracking error of flux and speed are
bounded.

27
 Besides, if [δ] is locally Lipschitz, the observation error and tracking error
can be guaranteed the exponential convergence (Jiang and Wu (2002)) and
finally to give

lim (t) = 0, lim zeij (t) = 0 and lim eik (t) = 0 (58)
t→∞ t→∞ t→∞

where i = 1, 2 indicates different subsystems, j = 1, 2, 3 indicates the state

orders in estimation and k = 1, 2 indicates the state orders in controller.
This guarantees that the estimation error from both the MRAS speed
estimator and SPOs and the close-loop error produced by the unknown dis-
turbance are finally converged to zero to make the IM system stable.
The internal dynamic of the IM system is anlysed uding a zero-dynamic
technique. In zero-dynamic, the estimated speed and rotor flux as well as
their derivatives are well controlled, i.e. [e] + [e ˙ + [e
z ] = 0 and [e] z˙ ] = 0 in
(38), the estimated states tracks their real value as ẑij = zij and system
outputs track their reference as zi1 = yi∗ . Thus, the system outputs can be
∗
represented as ωm = ωm , ψr = ψr∗ , ω̇m = 0, ψ̇r = 0, ẑi1 = yi∗ and ẑi3 = Ψi .
Then combining with (5) to (8), the control inputs defined in (30) can be
represented as
  
Rr Ls Ls ∗
 vsd0 = Rs + isd − σLs ωe isq − ψ


 Lr  Lm τ r r
Rr Ls P ωm Lm ∗ σLs Lm Rr
 vsq0 = Rs + isq + σLs ωe isd + ψr − isd isq


Lr Lr Lr ψr∗
(59)
Substitute (59) into the original IM system (1), one can derive that i̇sd = 0
and i̇sd = 0 and the steady state currents can be obtained as

ψr∗
lim isd (t) = (60)
t→∞ Lm

28
 2TL Lr
lim isq (t) = (61)
t→∞ 3P Lm ψr∗
After the states ωm and ψr and their derivatives are stable, the corre-
sponding states, such as the stator currents isd and isq , are also stable. The
zero-dynamic of the internal system of IM is stable and, therefore, the closed-
loop system error dynamics is stable as well.

5. Simulation Results

The control performance of SSNAC for IM is verified in simulation using

MATLAB/Simulink and in hardware implementation on dSPACE hardware
platform. The simulation tests are carried out in the time-continuous do-
main for both the observer and controllers. At first, in order to validate
the effectiveness of the CSPO, the speed estimation performance of NAC
with separated traditional MRAS speed observer and SSNAC with CSPO
are compared in simulation study. As the MRAS speed estimator is parame-
ter sensitive, the variation of IM parameters has been tested of both the rotor
resistance and load torque in order to test the performance reduction of the
speed sensorless approach affected by parameter variation. After that, the
case studies of forward and reverse motoring and time-varying load torque
disturbance have been given in both the software simulation and hardware
implementation. The parameters of target IM system have been given in
Table 1. The controller parameters of SSNAC as well as two SPOs are listed
in Table 2.

29
 Table 1: System parameters of IM
Parameter Value Unit
Rs 0.1607 Ω
Rr 0.1690 Ω
Ls 6.017 mH
Lr 5.403 mH
Lm 5.325 mH
J 0.000145 kg
P 2

Table 2: Controller parameters of SSNAC

Name Parameter & Value
l11 = 9 × 103
Flux SPO l12 = 2.7 × 107
l13 = 2.7 × 109
l20 = 2 × 103
l21 = 6 × 103
Speed SPO
l22 = 1.2 × 107
l23 = 8 × 109
k11 = 1.5 × 104
Flux Controller
k12 = 2.5 × 102
k21 = 1 × 104
Speed Controller
k22 = 2 × 102

5.1. Sensorless speed tracking performance test

This test aims to verify the effectiveness of using CSPO to replace the
PI based adaptation mechanism in conventional MRAS speed observer. The
comparison between separated MRAS-SO plus speed SPO and CSPO in sim-
ulation result is shown in Figure 3. The result demonstrates that both the

30
Figure 3: Speed tracking performance comparison between seperated SPO and combined
SPO.

separated SO plus speed SPO and CSPO has good performance in estimat-
ing the rotor speed from 0 to 150 rad/s. In Figure 3 (b), the comparison of
estimation error shows that the speed estimation performs better in higher
speed than lower speed in both method. As the separated MRAS-SPO and
CSPO has similar performance in speed estimation, only the results of CSPO
based SSNAC are presented in all the following case studies.

5.2. Sensitivity test to parameter variation

In IM systems, the rotor resistances are typically uncertain since they

are possible varied during operations due to rotor heating, especially in the
wound-rotor IMs (Marino et al. (2005); Li et al. (2015)). The SSNAC un-
der varied rotor resistance have been validated in simulation. The result of
SSNAC under different constant rotor resistance variation and a rated con-

31
Figure 4: Simulation result of SSNAC under (a) rated load torque and different rotor
resistance variation, and (b) 10% mismatch of rated rotor resistance and different load
torque.

stant load torque has been shown in Figure 4 (a), and that of a constant
10% rotor resistance variation and different load torque has been shown in
Figure 4 (b). When the rotor resistance varies with a constant mismatch to
its normal value, the speed controller performs the same dynamic response.
Up to 20% mismatching of Rr have been tested, with the estimation perfor-
mance degraded and steady-state shift with less than 4% estimation error,
which is acceptable. Thus, under a constant uncertain load torque and rotor
resistance, the dynamic response and stability of an IM controlled by SSNAC
will not be affected.

32
5.3. Speed tracking under constant load disturbance

In the IM application, the speed tracking under constant load is the most
common condition. Figure 5 shows the result of different control methods for
speed tracking under constant load, which can be either positive or negative
load torque. In the test case, from t=1.0 s to t=2.0 s, the load torque rises
from 0 to 0.4 Nm and keep constant. The rotor speed rises from 0 to 80 rad/s
and stays at this speed. After the speed reached the steady state, the load
torque reduced from positive to negative 0.4 Nm. During this period, the IM
should also be able to keep the speed. At last, from t=6.0 s, the IM start to
decelerate from 80 rad/s to 0 under the negative load applied on the rotor.
The load torque is shown as in Figure 5(a) and the rotor speed is shown as
in Figure 5(b). As the rotor speed is under the limit of the rated value, the
IM is not using in the field weakening area. Thus, the flux reference is set
to constant in its rated value at 0.0265 Wb during the speed tracking. The
rotor flux is shown in Figure 5(c).
In the control performance of rotor speed and rotor flux, it can be obvi-
ously found that the SSNAC tracks reference better than that of traditional
speed sensorless method of VC with MRAS speed observer. In order to com-
pare their relative tracking error, Figure 5(d) and 5(e) show the tracking error
of rotor speed and rotor flux in percentage. In the comparison, the maxi-
mum speed error of SSNAC is less than 1% at t=4.5 s while the maximum
speed error of VC plus MRAS is around 8% caused by the load disturbance
changed from positive to negative. When rotor speed is rising or decreasing,
the control performance of SSNAC is also better than the traditional control
method with faster error elimination. The result of rotor flux tracking error

33
 (a) (b)

(c) (d)

(e) (f)
Figure 5: Simulation results of IM speed tracking under constant load disturbance. (a)
Load disturbance, (b) rotor speed, (c) rotor flux, (d) speed tracking error, (e) flux tracking
error, and (f) stator voltage control input of SSNAC.

also shows that the SSNAC obviously reduced the flux tracking error caused
by varying of speed and load disturbance. In order to compare the control
performance numerically, the detailed indices are given in Table 3. The con-
trol input of stator voltage of SSNAC is given in Figure 5(f), which shows
that even the SSNAC performs much better in speed and flux tracking, the
control actuator is not over used as the stator voltage is within its limit.
As the indices summarised in Table 3, the maximum flux error of SSNAC

34
is 98% less than that of the traditional VC plus MRAS method while the
flux integral-absolute-error (IAE) of SSNAC has reduced 99%. In the speed
tracking indices, the maximum speed error and IAE of speed regulation are
reduced by 79% and 81%, respectively.
Even though the control performance has been validated as improved a
lot, the estimation performance of the CSPO will need to be tested. Figure
6(a) shows the speed estimation result in rotor speed tracking. The absolute
speed estimation error is shown in Figure 6(b). The result of speed esti-
mation shows that the maximum estimation error is less than 0.01 rad/s in
speed acceleration and deceleration. When the load changes from positive
to negative, the maximum estimation error is less than 0.004 rad/s at the
speed of 80rad/s. The reason of which is the absolute accuracy of model
and parameters between reference and adaptive model. But in practice, it is
difficult to obtain the exact model as well as its accurate parameters.
The effectiveness of perturbation estimating and compensating also needs
to be verified. Figure 6(c) and 6(e) show the estimation performance of
perturbation terms of flux and speed, where the real value of perturbation
terms in the nonlinear IM system is presented by the solid green lines and
the estimated perturbation terms from SPOs are presented by the dashed red
line. Their estimation errors are given in Figure 6(d) and 6(f). The result
shows that the average estimation error of flux perturbation term of around
2% and that of speed perturbation term is around 0.3%.

5.4. Time-varying load torque disturbance test

After the speed tracking under constant load disturbance has been tested,
another most common case is the constant speed regulation under time-

35
 (a) (b)

(c) (d)

(e) (f)
Figure 6: Estimation performance of the CSPO under constant load disturbance speed
tracking.

varying load torque as unknown external disturbance. In this case, both the
speed and flux are expected to be kept constant, but the load disturbance is
time-varying and fast changes. This is to verify the stability of the control
system under continuous disturbances. From the start, the rotor speed is
rising and stays at 100 rad/s. Then at t=4 s, the load torque applied on IM
with a sharp vibration in sinewave with amplitude of 0.3 Nm and frequency
of 0.5 Hz, as shown in Figure 7(a).

36
 The control performance of rotor speed and rotor flux regulation are
shown in Figure 7(b) and 7(c). Both the SSNAC and traditional VC plus
MRAS are affected by the load disturbance with speed sharply reduced. Then
the SSNAC responses fast to regulate the rotor speed back to its reference
while the traditional VC plus MRAS method has larger regulation error.
The flux regulation performance is more obvious that the time-varying load
has less impact to the SSNAC as it fully decoupled the interaction between
speed and flux. The relative regulation errors of rotor flux and rotor speed
are shown in Figure 7(d) and 7(e). And Figure 7(f) shows the stator voltage
of SSNAC as its control inputs. In the speed tracking error, an interesting
finding is the phase leading of SSNAC to VC plus MRAS due to the esti-
mation and compensation of perturbation terms that includes the prediction
of load torque. The performance indices are also summarised in Table 3, in
which it shows that the SSNAC reduced the maximum regulation error of
flux and speed by 99% and 87% and reduced the IAE of flux and speed by
99% and 88%, respectively.
The speed estimation performance of the CSPO is shown in Figure 8(a)
and its estimation error is shown in Figure 8(b). The result shows that the
maximum estimation error is less than 0.009 rad/s under the speed of 100
rad/s. That is also because of the accurate model and parameters used. The
two lumped perturbation terms of flux and rotor speed are estimated and
compared with their real value in Figure 8(c) and 8(e) while their estimation
error are given in Figure 8(d) and 8(f). The average estimation error of
perturbation term of flux is around 2% and that of speed is around 0.8%.
Thus, the perturbation terms are well estimated to compensate their real

37
 (a) (b)

(c) (d)

(e) (f)
Figure 7: Simulation results of IM constant speed regulation under time-varying load
torque disturbance. (a) Load disturbance, (b) rotor speed, (c) rotor flux, (d) speed tracking
error, (e) flux tracking error, and (f) stator voltage control input.

value for fully linearise the coupled states in the nonlinear IM system.

6. Experiment Results

6.1. Experimental platform

The experimental setup of SSNAC for IM hardware implementation is
shown in Figure 9. The target motor is a 200 W, 2 pole pairs, three phase
wound-rotor IM from Motorsolver. Other devices include a 42 V power elec-
tronics (PE) converter unit and current transducers, two power supplies to

38
 (a) (b)

(c) (d)

(e) (f)
Figure 8: Estimation performance of the CSPO under time-varying load torque distur-
bance.

provide power to both the motor and PE board in different voltage level,
and a DS1104 dSPACE controller with breakout box. In the motor bench,
the target IM is coupled with a DC motor that produces the expected load
torque. In the practical implementation, the first-order Euler discretization
method is used for compiling the simulation model into C programming code
in the dSPACE DS1104 controller board, whose CPU clock is 250 MHz.
Considering both the sampling rate and computational capability of the con-
troller hardware, the sampling time adopted for practical implementation is

39
Table 3: Simulation performance indices and reductions of VC plus MRAS and SSNAC
PP
PP Method VC+MRAS SSNAC Reduc.
PP
PP
Indices PP
P
Max. Flux Error 6.9% 0.13% 98.1%
Flux IAE (×10−3 ) 8.2 0.1 99%
Case 1
Max. Speed Error 7.8% 1.6% 79.5%
Speed IAE 20.4 3.8 81%
Max. Flux Error 3.7% 0.052% 99%
Flux IAE (×10−3 ) 3.0 0.0042 99%
Case 2
Max. Speed Error 2.2% 0.29% 87%
Speed IAE 6.3 0.77 88%

set to 0.0001 s. The observer and controller are operating under the same
implementation environment and sampling time. Thus, the whole control
system with proposed algorithm takes less than 25,000 computational clock
cycles in total. If the proposed algorithm is implemented in a computational
unit with lower clock frequency, the sampling time should be increased cor-
respondingly to allow the computational unit to have enough time to run
out the algorithm. In the experiment, the nominal value of motor parame-
ters has been used. The unmeasured parameter varying during operations is
considered as the uncertainty and external disturbance.

6.2. Speed tracking under constant load disturbance

The control parameters of experiments are the same with that in simu-
lation. The load torque disturbance and the speed references use the same
profiles as in simulation, as shown in Figure 10(a) and 10(b). In the hard-
ware implementation, the speed estimation performance and its estimation
error are given in Figure 10(c) and 10(d). From the speed estimation result,

40
 Figure 9: Experimental setup of IM speed sensorless control.

the maximum speed estimation error are around 2 rad/s during speed rising
and less than 0.7 rad/s in steady state. As the real perturbation of the IM
system is difficult to be obtained in experimental test, the performance of
perturbation estimation cannot be given.
In the rotor flux and speed tracking, the rotor flux is controlled to stay at
0.0265 Wb while rotor speed increases from 0 to 80 rad/s and stays at this
speed. Their performance is shown in Figure 11(a) and 11(b). During the
period of speed rising, the flux and speed tracking error of VC plus MRAS
can reach 18.2% and 8.6rad/s, respectively. As shown in Figure 11(c) and
11(d), the SSNAC can reduce the flux regulation error to 13.2% and speed
tracking error to 3.1 rad/s. More than half of the speed tracking error has
been reduced.

41
 (a) (b)

(c) (d)
Figure 10: Experimental results of speed tracking under constant load disturbance. (a)
Load disturbance, (b) speed reference, (c) speed estimation, (d) estimation error.

42
 (a) (b)

(c) (d)

(e) (f)
Figure 11: Experimental results of speed tracking under constant load disturbance. (a)
rotor flux performance, (b) rotor speed performance, (c) flux error, (d) speed error, (e)
stator voltage control input of VC plus MRAS, and (f) stator voltage control input of
SSNAC.

43
 In addition, after the rotor speed reaches the steady state, the change of
load torque disturbance from positive 0.4 Nm to negative 0.4 Nm causes the
speed varying at t=6 s. The regulation performance of flux and speed under
the change of load torque using VC plus MRAS method is about 3.9% and
2.8 rad/s, respectively. While the SSNAC reduces the flux regulation error
to 0.8% and speed tracking error to 0.5 rad/s. It verifies that the SSNAC has
obvious improvement in the sensorless tracking of both flux and rotor speed.
Figure 11(e) and 11(f) show the control inputs of stator voltage of VC plus
MRAS and SSNAC. From which it can be found that the SSNAC has less
voltage input in average and its peak value is less than 6V while that of VC
plus MRAS is close to 8 V.
The numerical comparison of performance indices are summarised in Ta-
ble 4 case 1. The results show that, in speed tracking under constant load
torque, the SSNAC performs better than that of VC plus MRAS method
with the reduction of maximum and IAE of flux regulation error by 21% and
77%, respectively. And the reduction of that on speed tracking error by 33%
and 75%. The reduction of SSNAC in experiments is less than that in simu-
lation. The reason of that is caused by the uncertainty of IM parameters in
rotor-flux model. In experiments, the effective parameters of IM system can
be varying with ambient temperature and the delay of current feedback from
transducers, which is assumed ideal in simulation test. In order to compare
the reduction of flux and speed tracking error in an obvious way, the bar
chart of performance indices is given for this case in Figure 14(a).

44
 (a) (b)

(c) (d)
Figure 12: Experimental results of constant speed regulation under time-varying load
torque disturbance. (a) Load disturbance, (b) speed reference, (c) speed estimation, (d)
estimation error.

6.3. Time-varying load torque disturbance test

In constant speed regulation, the IM rotor speed is expected to be main-

tained at 100 rad/s, as in Figure 12(b). A sinewave shape time varying load
torque disturbance is then applied to the IM as shown in Figure 12(a). Under
this case, the rotor speed estimation performance of the CSPO is shown in
Figure 12(c) and its estimation error is shown in Figure 12(d). The result
shows that the designed observer can estimate the rotor speed of the target
IM system under time-varying load torque disturbance with its estimation
error less than 1.7% in any time and 0.8% in average.
The flux and speed regulation performance and their relative regulation
error are shown in Figure 13(a) and 13(b) and their relative regulation error

45
Table 4: Experiment performance indices and reductions of VC plus MRAS and SSNAC
PP
PP Method VC+MRAS SSNAC Reduc.
PP
PP
Indices PP
P
Maximum Flux Error 18.3% 13% 21%
Flux IAE (×10−3 ) 9.3 2.1 77%
Case 1
Maximum Speed Error 4.2% 2.8% 33%
Speed IAE 31.0 7.7 75%
Maximum Flux Error 3.7% 0.78% 79%
Flux IAE (×10−3 ) 2.3 0.46 80%
Case 2
Maximum Speed Error 3.2% 1.6% 52%
Speed IAE 8.9 6.4 28%

are shown in Figure 13(c) and 13(d). The peak regulation error of flux and
speed in VC plus MRAS method are 3.7% and 3.2%, while that of the SSNAC
are reduced to 0.8% and 1.6%. The control input of SSNAC does not have
any increment to achieve the improvement of control performance in flux and
speed regulation as shown in Figure 13(e) and 13(f)
The performance indices are summarised and compared between SSNAC
and traditional VC plus MRAS method in Table 4 case 2 and Figure 14(b).
The results show that the SSNAC reduced the flux regulation error and its
IAE by 79% and 80% while the reduction of rotor speed regulation error is
52% in peak and 28% in IAE. The effectiveness of the speed sensorless control
using SSNAC to fully decouple the interaction of flux and rotor speed has
been verified with obvious better performance.

46
 (a) (b)

(c) (d)

(e) (f)
Figure 13: Experimental results of constant speed regulation under time-varying load
torque disturbance. (a) rotor flux performance, (b) rotor speed performance, (c) flux
error, (d) speed error, (e) stator voltage control input of VC plus MRAS, and (f) stator
voltage control input of SSNAC.

47
 (a)

(b)
Figure 14: Comparison of performance indices of experimental results in (a) speed tracking
under constant load disturbance, (b) constant speed regulation under time-varying load
disturbance.

48
7. Conclusion

This paper proposed a speed sensorless nonlinear adaptive control to

achieve the fully decoupling control of the flux and speed of IM without the
usage of speed sensor. The proposed nonlinear adaptive controller has been
used together with the traditional rotor-flux MRAS speed estimator. Then
a combined speed and perturbation observer (CSPO) is designed to estimate
the rotor speed and its perturbation term simultaneously. The estimated
speed is used as feedback in the speed loop and the estimated lumped per-
turbation terms is used to compensate the real perturbation that contains the
nonlinear dynamics, external load disturbance and parameter uncertainties.
Moreover, the CSPO replaced the PI regulator in the MRAS speed observer
and thus reduced the complexity of SSNAC controller. With the estimated
speed and perturbation terms from the CSPO, the IM system can be fully
decoupled and controlled without speed sensor. The stability of SSNAC with
CSPO has been proved using Lyapunov theory. The effectiveness of SSNAC
has been verified in both simulation and experiment studies. The results
show that the SSNAC performs better in speed tracking and under unknown
load disturbance without speed sensors comparing with traditional VC plus
MRAS method. It validated that the estimation of speed and its perturba-
tion term using the CSPO can provide speed sensorless control as well as the
fully decoupling of interactions in the nonlinear IM system.
In summary, the main advantages of the proposed control approach are
given as below:

ˆ The proposed control approach integrates nonlinear control technology

and speed estimation technology, so it not only reduces the use of PI

49
 regulator in the traditional MRAS speed estimator, but also enables
overall stability analysis.

ˆ The proposed control method can estimate the rotor speed and its dis-
turbance term simultaneously, thus, it reduces the dependence of speed
sensors, accurate models and parameters to improve the robustness of
the control system.

ˆ The proposed control approach can perform much better speed tracking
performance under model uncertainty and unknown load disturbance.

The limitations of this work are listed as follows. First, the proposed
approach requires a high-gain observer in the SPO design. Therefore, when
speed estimation and perturbation estimation are combined, the selection of
the optimal gains becomes difficult. The best way to adjust the observer gain
will be considered in future studies. Secondly, the speed sensorless technology
in this article uses the MRAS method, which requires accurate models and
parameters. Although the perturbation estimation approach reduces the
dependency of accurate model and parameters, more other speed estimation
methods will be studied in the future to choose the most robust way to
combine the speed sensorless techniques with the perturbation estimation
approach.
In future work, the authors will study the method of finding the optimal
gains for the observer and the controller in both theoretical and practical
aspects. And compare the combination of different types of speed sensorless
methods and perturbation estimation-based control methods under different
conditions and application scenarios. In addition, the authors will continue

50
to study the sensitivity of this method to system parameter uncertainty and
fault detection methods in hardware implementation.

Appendix A. Proof of a2 >0

Since the ψrα and ψrβ are in stationary frame, assuming that the angle
of rotor flux is θr . Then it can be represented as ψrα = |ψr | sin θr and ψrβ =
|ψr | cos θr . Then assuming that the estimation error of rotor flux angle is θer ,
   
and one can get ψ̂rα = ψ̂r sin θr + θer and ψ̂rβ = ψ̂r cos θr + θer . On the
basis of these, it can be proved that

a2
= ψrα ψ̂rα + ψrβ ψ̂rβ (A.1)
P    
= |ψr | sin θr · ψ̂r sin θr + θer + |ψr | cos θr · ψ̂r cos θr + θer
   
= |ψr | ψ̂r · sin θr · sin θr + θr + |ψr | ψ̂r · cos θr · cos θr + θr
e e
 
= |ψr | ψ̂r sin2 θr cos θer + cos2 θr cos θer

= |ψr | ψ̂r cos θer

Thus, if θer < 90◦ , a2 can be proved as a positive value. In normal condition
of IM control, the estimation error of rotor angle will be far less than 90
degree. Thus, in this paper, it assumes that a2 > 0 in all conditions.

Appendix B. Proof of λ is bounded

From equation (35), since a2 ωm and a3 are calculated from the physical
variables, they can be assumed as bounded based on the fact that the target
system is a physical system containing mechanical and electrical processes,
and its variables will be within a suitable range instead of infinite values.

51
 R
Thus, this section will only prove the boundness of kI  dt. From (20), 
can be rewritten as

 = ψrβ ψ̂ rα − ψrα ψ̂rβ (B.1)

   
= ψ̂r sin θr + θer · |ψr | cos θr − ψ̂r cos θr + θer · |ψr | sin θr
    
= ψ̂r |ψr | · cos θr · sin θr + θer − sin θr · cos θr + θer

= ψ̂r |ψr | sin θer

Then it is easy to find that

Z
 dt = ψ̂r |ψr | cos θer ≤ ψ̂r |ψr | (B.2)
R
Thus  dt is proved as bounded since the rotor flux ψr has the upper limit
according to the design and manufacturing. And on the basis of this, λ can
be proved as bounded then.

References

Bhowate, A., Aware, M., Sharma, S., 2019. Predictive torque control with
online weighting factor computation technique to improve performance of
induction motor drive in low speed region. IEEE Access 7, 42309–42321.
doi:10.1109/ACCESS.2019.2908289.

Bimal, K., 2002. Bose. Modern power electronics and AC drives .

Boukas, T., Habetler, T., 2004. High-performance induction motor speed

control using exact feedback linearization with state and state deriva-
tive feedback. IEEE Transactions on Power Electronics 19, 1022–1028.
doi:10.1109/TPEL.2004.830042.

52
Callegaro, A.D., Nalakath, S., Srivatchan, L.N., Luedtke, D., Preindl, M.,
2018. Optimization-based position sensorless control for induction ma-
chines, in: 2018 IEEE Transportation Electrification Conference and Expo
(ITEC), IEEE. pp. 460–466. doi:10.1109/ITEC.2018.8450137.

Chen, J., Jiang, L., Yao, W., Wu, Q., 2014. Perturbation estimation based
nonlinear adaptive control of a full-rated converter wind turbine for fault
ride-through capability enhancement. IEEE Transactions on Power Sys-
tems, 29, 2733–2743. doi:10.1109/TPWRS.2014.2313813.

Chen, J., Yao, W., Ren, Y., Wang, R., Zhang, L., Jiang, L., 2019a. Nonlinear
adaptive speed control of a permanent magnet synchronous motor: A per-
turbation estimation approach. Control Engineering Practice 85, 163–175.
doi:10.1016/j.conengprac.2019.01.019.

Chen, J., Yao, W., Zhang, C.K., Ren, Y., Jiang, L., 2019b. Design of robust
MPPT controller for grid-connected PMSG-based wind turbine via pertur-
bation observation based nonlinear adaptive control. Renewable Energy
134, 478–495. doi:10.1016/j.renene.2018.11.048.

Costa, B.L.G., Graciola, C.L., Angélico, B.A., Goedtel, A., Castoldi, M.F.,
de Andrade Pereira, W.C., 2019. A practical framework for tuning DTC-
SVM drive of three-phase induction motors. Control Engineering Practice
88, 119–127.

Ellis, G., 2002. Observers in control systems: a practical guide. Elsevier.

Englert, T., Graichen, K., 2020. Nonlinear model predictive torque con-

53
 trol and setpoint computation of induction machines for high performance
applications. Control Engineering Practice 99, 104415.

Feng, G., Liu, Y.F., Huang, L., 2004. A new robust algorithm to im-
prove the dynamic performance on the speed control of induction mo-
tor drive. IEEE Transactions on Power Electronics 19, 1614–1627.
doi:10.1109/TPEL.2004.836619.

Finch, J., Giaouris, D., 2008. Controlled ac electrical drives. IEEE Transac-
tions on Industrial Electronics, 55, 481–491. doi:10.1109/TIE.2007.911209.

Gadoue, S., Giaouris, D., Finch, J., 2010. MRAS sensorless vector con-
trol of an induction motor using new sliding-mode and fuzzy-logic adapta-
tion mechanisms. IEEE Transactions on Energy Conversion, 25, 394–402.
doi:10.1109/TEC.2009.2036445.

Grabowski, P., Kazmierkowski, M., Bose, B., Blaabjerg, F., 2000. A sim-
ple direct-torque neuro-fuzzy control of PWM-inverter-fed induction mo-
tor drive. IEEE Transactions on Industrial Electronics, 47, 863–870.
doi:10.1109/41.857966.

Guzinski, J., Abu-Rub, H., 2013. Speed sensorless induction motor drive with
predictive current controller. IEEE Transactions on Industrial Electronics
60, 699–709. doi:10.1109/TIE.2012.2205359.

Habibullah, M., Lu, D.D.C., 2015. A speed-sensorless FS-PTC of induction

motors using extended kalman filters. IEEE Transactions on Industrial
Electronics 62, 6765–6778. doi:10.1109/TIE.2015.2442525.

54
Han, J., 2009. From PID to active disturbance rejection con-
trol. IEEE Transactions on Industrial Electronics, 56, 900–906.
doi:10.1109/TIE.2008.2011621.

Holakooie, M.H., Ojaghi, M., Taheri, A., 2019. Modified DTC of a six-
phase induction motor with a second-order sliding-mode MRAS-based
speed estimator. IEEE Transactions on Power Electronics 34, 600–611.
doi:10.1109/TPEL.2018.2825227.

Holtz, J., 2005. Sensorless control of induction machines- with or without

signal injection? IEEE Transactions on Industrial Electronics, 53, 7–30.
doi:10.1109/TIE.2005.862324.

Hu, Y., Song, X., Cao, W., Ji, B., 2014. New SR drive
with integrated charging capacity for plug-in hybrid electric vehicles
(PHEVs). IEEE Transactions on Industrial Electronics 61, 5722–5731.
doi:10.1109/TIE.2014.2304699.

Jiang, L., Wu, Q., Wen, J., 2004. Decentralized nonlinear adaptive control
for multimachine power systems via high-gain perturbation observer. IEEE
Transactions on Circuits and Systems I: Regular Papers, 51, 2052–2059.
doi:10.1109/TCSI.2004.835657.

Jiang, L., Wu, Q.H., 2002. Nonlinear adaptive control via sliding-mode state
and perturbation observer. IEE Proceedings - Control Theory and Appli-
cations 149, 269–277. doi:10.1049/ip-cta:20020470.

Kivanc, O.C., Ozturk, S.B., 2018. Sensorless PMSM drive based on

stator feedforward voltage estimation improved with MRAS multi-

55
 parameter estimation. IEEE/ASME Transactions on Mechatronics
doi:10.1109/TMECH.2018.2817246.

Li, J., Ren, H.P., ru Zhong, Y., 2015. Robust speed control of in-
duction motor drives using first-order auto-disturbance rejection con-
trollers. IEEE Transactions on Industry Applications 51, 712–720.
doi:10.1109/TIA.2014.2330062.

Li, J., Xu, L., Zhang, Z., 2005. An adaptive sliding-mode observer for in-
duction motor sensorless speed control. IEEE Transactions on Industry
Applications 41, 1039–1046. doi:10.1109/ICEMS.2019.8922535.

Li, S., Xia, C., Zhou, X., 2012. Disturbance rejection control method for per-
manent magnet synchronous motor speed-regulation system. Mechatronics
22, 706–714. doi:10.1016/j.mechatronics.2012.02.007.

Liu, Y., Zhao, J., Wang, R., Huang, C., 2013. Performance improvement
of induction motor current controllers in field-weakening region for elec-
tric vehicles. IEEE Transactions on Power Electronics 28, 2468–2482.
doi:10.1109/TPEL.2012.2217757.

Marchesoni, M., Passalacqua, M., Vaccaro, L., Calvini, M., Venturini, M.,
2020. Performance improvement in a sensorless surface-mounted PMSM
drive based on rotor flux observer. Control Engineering Practice 96,
104276.

Marino, R., Peresada, S., Valigi, P., 1993. Adaptive input-output linearizing
control of induction motors. IEEE Transactions on Automatic Control 38,
208–221. doi:10.1109/9.250510.

56
Marino, R., Tomei, P., Verrelli, C., 2005. Adaptive control for speed-
sensorless induction motors with uncertain load torque and rotor resis-
tance. International Journal of Adaptive Control and Signal Processing
19, 661–685. doi:10.1002/acs.874.

Marino, R., Tomei, P., Verrelli, C.M., 2008. An adaptive tracking

control from current measurements for induction motors with uncer-
tain load torque and rotor resistance. Automatica 44, 2593–2599.
doi:10.1016/j.automatica.2008.02.023.

Nasiri, A., 2007. Full digital current control of permanent magnet syn-
chronous motors for vehicular applications. IEEE Transactions on Ve-
hicular Technology 56, 1531–1537. doi:10.1109/TVT.2007.896969.

Ohyama, K., Asher, G., Sumner, M., 2005. Comparative analysis

of experimental performance and stability of sensorless induction mo-
tor drives. IEEE Transactions on Industrial Electronics 53, 178–186.
doi:10.1109/TIE.2005.862298.

Perin, M., Pereira, L.A., Pereira, L.F., Sogari, P.A., Nicol, G., 2021. A
method to estimate parameters of five-phase induction machines including
the third-harmonic airgap field. Control Engineering Practice 111, 104792.

Proca, A., Keyhani, A., 2007. Sliding-mode flux observer with online rotor
parameter estimation for induction motors. IEEE Transactions on Indus-
trial Electronics, 54, 716–723. doi:10.1109/TIE.2007.891786.

Pyrkin, A., Bobtsov, A., Vedyakov, A., Ortega, R., Vediakova, A., Sinetova,
M., 2019. DREM-based adaptive observer for induction motors, in: 2019

57
 IEEE 58th Conference on Decision and Control (CDC), IEEE. pp. 648–
653. doi:10.1109/CDC40024.2019.9029739.

Ravi Teja, A., Chakraborty, C., Maiti, S., Hori, Y., 2012. A new model
reference adaptive controller for four quadrant vector controlled induction
motor drives. IEEE Transactions on Industrial Electronics, 59, 3757–3767.
doi:10.1109/TIE.2011.2164769.

Rehman, H., Xu, L., 2011. Alternative energy vehicles drive system: Control,
flux and torque estimation, and efficiency optimization. IEEE Transactions
on Vehicular Technology, 60, 3625–3634. doi:10.1109/TVT.2011.2163537.

Ren, Y., Li, L., Brindley, J., Jiang, L., 2016. Nonlinear PI control for
variable pitch wind turbine. Control Engineering Practice 50, 84–94.
doi:10.1016/j.conengprac.2016.02.004.

Rind, S.J., Ren, Y., Hu, Y., Wang, J., Jiang, L., 2017. Configurations and
control of traction motors for electric vehicles: A review. Chinese Journal
of Electrical Engineering 3, 1–17. doi:10.23919/CJEE.2017.8250419.

Schauder, C., 1992. Adaptive speed identification for vector control of in-
duction motors without rotational transducers. IEEE Transactions on In-
dustry Applications 28, 1054–1061. doi:10.1109/28.158829.

Suetake, M., da Silva, I.N., Goedtel, A., 2011. Embedded DSP-based

compact fuzzy system and its application for induction-motor V/f speed
control. IEEE Transactions on Industrial Electronics 58, 750–760.
doi:10.1109/TIE.2010.2047822.

58
Sun, X., Chen, L., Yang, Z., Zhu, H., 2013. Speed-sensorless vector con-
trol of a bearingless induction motor with artificial neural network inverse
speed observer. IEEE/ASME Transactions on mechatronics 18, 1357–1366.
doi:10.1109/TMECH.2012.2202123.

Sung, W., Shin, J., Jeong, Y.s., 2012. Energy-efficient and robust control
for high-performance induction motor drive with an application in elec-
tric vehicles. IEEE transactions on vehicular technology 61, 3394–3405.
doi:10.1109/TVT.2012.2213283.

Tarchala, G., Orlowska-Kowalska, T., 2018. Equivalent-signal-based sliding

mode speed MRAS-type estimator for induction motor drive stable in the
regenerating mode. IEEE Transactions on Industrial Electronics 65, 6936–
6947. doi:10.1109/TIE.2018.2795518.

Tilli, A., Conficoni, C., 2016. Semiglobal uniform asymptotic stabil-

ity of an easy-to-implement PLL-like sensorless observer for induc-
tion motors. IEEE Transactions on Automatic Control 61, 3612–3618.
doi:10.1109/TAC.2016.2521728.

Verrelli, C.M., Lorenzani, E., Fornari, R., Mengoni, M., Zarri, L., 2018.
Steady-state speed sensor fault detection in induction motors with uncer-
tain parameters: A matter of algebraic equations. Control Engineering
Practice 80, 125–137.

Wang, H., Liu, Y.c., Ge, X., 2018a. Sliding-mode observer-based speed-
sensorless vector control of linear induction motor with a parallel secondary

59
 resistance online identification. IET Electric Power Applications 12, 1215–
1224. doi:10.1049/iet-epa.2018.0049.

Wang, Y., Zhou, L., Bortoff, S.A., Satake, A., Furutani, S., 2018b.
An approximate high gain observer for speed-sensorless estimation
of induction motors. IEEE/CAA Journal of Automatica Sinica
doi:10.1109/TMECH.2012.2202123.

Wu, L., Zheng, W.X., Gao, H., 2013. Dissipativity-based sliding mode control
of switched stochastic systems. IEEE Transactions on Automatic Control
58, 785–791. doi:10.1109/TAC.2012.2211456.

Yang, S., Ding, D., Li, X., Xie, Z., Zhang, X., Chang, L., 2018. A decoupling
estimation scheme for rotor resistance and mutual inductance in indirect
vector controlled induction motor drives. IEEE Transactions on Energy
Conversion doi:10.1109/TEC.2018.2880796.

Yoo, D., Yau, S.T., Gao, Z., 2007. Optimal fast tracking observer bandwidth
of the linear extended state observer. International Journal of Control 80,
102–111. doi:10.1080/00207170600936555.

Zaky, M.S., Metwaly, M.K., Azazi, H.Z., Deraz, S.A., 2018. A new adaptive
SMO for speed estimation of sensorless induction motor drives at zero
and very low frequencies. IEEE Transactions on Industrial Electronics 65,
6901–6911. doi:10.1109/TIE.2018.2793206.

Zhang, X., 2013. Sensorless induction motor drive using indirect vector con-
troller and sliding-mode observer for electric vehicles. IEEE Transactions
on Vehicular Technology 62, 3010–3018. doi:10.1109/TVT.2013.2251921.

60