Template for Electronic Submission to ACS Journals


The Journal of Engineering Research (TJER), Vol. 16, No. 1 (2019) 63-76 

 
  

 

CARRIER BASED PWM TECHNIQUE AND ADAPTIVE NEURAL 

NETWORK BASED ROTOR RESISTANCE ESTIMATOR FOR THE 

PERFORMANCE ENHANCEMENT OF VECTOR CONTROLLED 

INDUCTION MOTOR DRIVES  

 
A. Venkadesan  

Department of Electrical and Electronics Engineering, National Institute of Technology Puducherry, 

Thiruvettakudy, karaikal-609609, India.  
 

ABSTRACT: In this paper, the carrier based Pulse Width Modulation (PWM) technique and neural network 

based rotor resistance estimator are proposed for vector controller Induction motor (IM) drives. The popular sine 

PWM is used for induction motor drive. The popular sine PWM has poor harmonic profile and (DC) utilization. 

The space vector modulation (SVM) technique overcomes the disadvantages of sine PWM. But SVM is 

computationally complex. Hence a simple PWM technique namely “carrier based PWM technique” similar to 

SVM is identified and proposed for vector controlled IM drive. The experimental set up is built up and the 

performance of carrier based PWM is validated using FPGA processor. The adaptive neural network based rotor 

resistance estimator in predictive mode is proposed for the vector controlled induction motor drive. The 

performance enhancement of the drive with carrier based PWM and rotor resistance estimator is 

comprehensively presented.  

 
 

Keywords:  Carrier based PWM technique; Rotor resistance estimator; MRAS; Adaptive neural network; Vector 

control; Induction motor drive. 
 

تقنية تعديل عرض النبضة القائمة على الناقل ومقدار مقاومة الدوران القائم على الشبكة العصبية التكيفية من 

 الحثية موجهة التحكم للمحركاتأجل تحسين األداء 

 
 فينكاديسانأ. 

 
العصبية التكيفية باقتراح تعديل عرض النبضة القائمة على الناقل مقاوم الدوران القائم على الشبكة  : قمنا الملخص

ويتسم جيب زاوية جهاز تعديل عرض النبضة بسمة  توافقية ضعيفة عند  استخدام التيار  الحثية موجهة التحكم.  للمحركات

المباشر.  ويعمل استخدام تقنية تعديل اتجاه النطاق على التغلب على عيوب جيب زاوية تعديل عرض النبضة. لكن يُتعد 

بتحديد  تقنية بسيطة لتعديل عرض النبضة وهي  ق أمرا معقدا من الناحية الحسابية. ومن ثم ، قمنا تقنية تعديل اتجاه النطا

"تقنية  تعديل عرض النبضة القائمة على الناقل المشابهة لتقنية تعديل  اتجاه النطاق، واقترحنا محرك حثي قابل للتحكم فيه. 

أداء ناقل الحركة القائم على تقنية تعديل اتجاه نطاق  و قمنا بإعداد التركيب التجريبي والتحقق من مدى صحة

باستخدام معالج تنسيق قابل للبرمجة لنطاق المدخل. ونقترح استخدام  تقنية مقاومة الدوران القائم على الشبكة  الحركة 

ر تحسين األداء الحثية موجهة التحكم. و عملنا على عرض شامل  لمقدا العصبية التكيفية في الوضع التنبئي   للمحركات

 الحثية موجهة التحكم مع تعديل عرض النبضة القائم على الناقل ومقدّر مقاومة الدوران. للمحركات
 

 

 التكيفية العصبية الشبكة التكيف لنظام مرجعي نموذج  الدوران مقاومة مقدار  : تعديل عرض النبضةالكلمات المفتاحية

 . التحكم موجهة الحثية المحركات الناقل متحكم

 
 
 
 
 
 
 
 
 

Corresponding author’s e-mail: venkadesan@nitpy.ac.in  

 

 

 

 

 

 

             DOI: 10.24200/tjer.vol.16iss1pp63-76 



The Journal of Engineering Research (TJER), Vol. 16, No. 1 (2019) 63-76 

  

NOMENCLATURE 
 

s
qsv,

s
ds

v
 

- d-axis Stator voltage, q-axis Stator 

voltage 

s
qsi,

s
ds
i

 

- d-axis Stator current, q-axis Stator 

current 

s
qr,

s
dr

  - d-axis Rotor flux, q-axis Rotor flux 

rR,sR  - Stator resistance, Rotor Resistance 

rL,sL  - Stator inductance, rotor inductance 

mL  - Magnetization inductance 

sT  -Sampling Time 

r  -Rotor Speed (rad/sec)  

Superscript s -Stationary Reference Frame 

Superscript e 

 

-Synchronous Reference Frame 

 

1. INTRODUCTION 
 

Three phase induction motor drives are popularly 

employed for variable speed control applications in 

many industries. This is due to their ease in 

maintenance (Abdelkarim Ammar et al. 2017; 

Ahmed A.Z. Diab 2014; Venkadesan A et al. 2016; 

Venkadesan A et al. 2017; Witold Pawlus et al. 

2017). Now-a-days, for high performance 

applications, vector control provides good alternative 

as compared to scalar control methods. Various pulse 

width modulation techniques are used to generate 

pulse for three phase inverter in vector controlled IM 

drives. Out of which, space vector modulation 

technique (Benchabane et al. 2012; Bimal K. Bose 

2005; Durgasukumar G et al. 2012; Govindasamy et 

al. 2014; Hannan M. A  et al. 2017; Mahmoud 

Gaballah et al. 2013; Sabah V.S et al. 2015) is 

attaining popularity because it has the advantages of 

having better DC utilization and better harmonic 

profile as compared to sine-PWM technique. The 

space vector modulation technique is computationally 

rigorous and complex as it involves many 

mathematical calculations (Govindasamy et al. 2014; 

Hannan M.A et al. 2017; Sabah V.S et al. 2015). The 

carrier based PWM technique with a common mode 

voltage injection method is proposed for a three 

phase inverter Keliang Zhou et al. 2002). In the paper 

by Keliang Zhou et al. (2002), it is mathematically 

proved that the carrier based PWM technique with 

the common mode voltage injection method performs 

similar to SVM technique. But the carrier based 

PWM technique is simple for on-line digital 

implementation. In the paper Keliang Zhou et al. 

(2002), the application of the technique is not 

demonstrated in vector controlled IM drives. The 

same technique is proposed for open loop speed 

control of multilevel inverter fed permanent magnet 

motor Shriwastava R.G et al. (2016) and not 

demonstrated in vector controlled IM drives.  

     In  this  paper, the  same technique is proposed for  

 

 

vector controlled IM drive. The proposed carrier based 

PWM technique with the common mode voltage 

injection method technique (CB-PWM-CMV) is also 

implemented in FPGA (SPARTAN 6 board). The 

performance of CB-PWM-CMV is also validated 

experimentally. The vector control especially indirect 

type, is sensitive to rotor resistance variation (Chitra A 

et al. 2015; Karanayil B et al. 2007). Hence, rotor 

resistance estimation using a neural network approach 

is proposed to track variation in the rotor resistance. 

The neural network based rotor resistance estimator is 

proposed for sine PWM based vector controlled IM 

drive Chitra A et al. (2015). In the proposed work, 

neural network based rotor resistance estimator is 

proposed for CB-PWM-CMV based vector controlled 

IM drive. Also in the paper (Chitra A et al. 2015; 

Karanayil B et al. 2007), the adaptive NN is used in 

simulation mode. But the proposed Adaptive NN based 

rotor resistance estimator is used in predictive mode. 

The idea of using adaptive NN in predictive mode is 

inspired from the speed estimator proposed by 

Maurizio Cirrincione et al. (2005). In the paper by 

Maurizio Cirrincione et al. (2005), it is applied for 

speed estimation. In this proposed work, it is applied 

for rotor resistance estimation which is novel in this 

paper.  

 

2. VECTOR CONTROLLED IM DRIVE 
 

The vector control is preferred for high performance 

applications. The advantages of vector control are well 

presented in Krishnan R (2007). The indirect vector 

controlled IM drive with proposed PWM technique 

(CB-PWM-CMV) and rotor resistance estimator is 

shown in  Fig. 1. For closed operation, the actual speed 

is compared with the reference speed. The speed error 

is  processed   through   the   PI   controller.  The torque  

 

 
 

Figure 1. Indirect Vector Controlled IM Drive showing the 

Rotor Resistance Estimator. 

 

 

64 



A.Venkadesan  

 

command ( *T
e

) is generated. The reference flux 

producing component of current ( e*id ) is generated 

from the reference flux ( φ
ref

) using equation (1). The 

reference torque producing component of current ( e*iq

) is generated using the equation (2). The slip 

frequency (
sl

ω ) is obtained using equation (3). The 

transformation angle (
e

θ ) is obtained using the 

equation (4).  The actual three phase stator current is 

transformed to actual flux producing component ( eid ) 

and actual torque producing component ( eiq ). The (
e

id

) is compared with ( e*id )and (
e

iq ) is compared with (

e*iq ) to produce the corresponding command voltage 

(Vd*, Vq*). The reference three phase voltage (Va*, 

Vb*, Vc*) is generated using the command voltage 

and transformation angle through d-q transformation 

theory. The three phase duty cycle profiles are 

generated using the CB-PWM-CMV. The duty cycle 

profile/control signal is compared with the triangular 

signal to generate the switching pulses to the inverter. 

 

φ
refe*=id L
m

                                 (1) 

 

*4e*=iq
3p φ

ref

L T
r e

L
m

 
 
 
 

            (2) 

r

sl

r

e*Riq
ω =

e*Lid

             (3) 

 e sl rθ = ω +ω dt              (4)  

 

2.1 Carrier Based PWM With Common Mode 
Voltage Injection Technique  

     In the carrier based PWM with the common 

mode voltage injection technique, the common 

mode voltage is injected into the fundamental 

sinusoidal reference signal. The resultant signal is 

used as the modulating signal and it is compared 

with the carrier signal for the generation of 

switching pulses to the inverter. The addition of the 

common mode voltage provides lower harmonic 

distortion, higher fundamental output voltage 

Srirattanawichaikul W et al. (2011).  The common 

mode voltage can be generated using three phase 

sinusoidal reference voltages as (5). The duty cycle 

for each phase can be generated using (6). 

 

   Max *, *, * Min *, *, *a b c a b c
*

2
CMV

v v v v v v
v


        (5) 

 

 * * ,  , ,P P CMVD v v P a b c              (6) 

 

 * sina av m t               (7) 

 

2
* sin

3
b a

v m t



 

  
 

                          (8) 

 

2
* sin

3
c a

v m t



 

  
 

                           (9) 

 

     Where ma is the modulation index. The range of ma 

is 0<ma ≤1 for under modulation ranges. The Fig. 2 

shows the block diagram for the extraction of the duty 

cycle profile. The equations involve  only  simple  

arithmetic  operations.  The equations  are  simple,  

less complex and give ease in digital implementation. 

The Fig. 3 presents the duty cycle profile of CB-PWM-

CMV. 

 

 
 

Figure 2. Duty Cycle profile Generation using CB-PWM- 

CMV Technique. 

 

     The performance of the proposed CB-PWM-CMV 

technique is compared with the SVM and Sine PWM 

technique in terms of DC bus utilization, harmonics 

profile, efficiency and complexity. The motor is 

operated with m=0.9, frequency=50Hz under a rated 

load condition. The performance comparison is shown 

in the Table 1.  

 
Table 1. Comparison of CB-PWM-CMV with Sine PWM 

and SVM. 

 

V
D

C
 

(v
o

lt
s)

 

M
e
th

o
d

 

O
u

tp
u

t 
 V

o
lt

a
g

e
 

V
fu

n
d

a
m

e
n

ta
l 

(v
o

lt
s)

 

O
u

tp
u

t 
 V

o
lt

a
g

e
 

V
T

H
D

 

(%
) 

%
 E

ff
ic

ie
n

c
y

 

C
o

m
p

u
ta

ti
o

n
a
l 

c
o

m
p

le
x
it

y
 

 

 

582.5 

 

Sine 

PW

M 

455.6 79.10% 78.20% Simple 

SV

M 

528 

 

63.68% 83.43% Comple

x  

CB-

PW

M-

CM

V 

528 63.68% 83.43% Simpler 

than 

SVM 

65 



The Journal of Engineering Research (TJER), Vol. 16, No. 1 (2019) 63-76 

  

 
(a) 

 
(b) 

 
(c) 

 
(d) 

 

Figure 3. (a) Duty cycle profile of CB-PWM-CMV (b) Max(Va, Vb, Vc) (c) Min(Va, Vb, Vc) (d) Common mode voltage.

66 



A.Venkadesan 

  

 

 

     The Sine PWM is simple, but has poor DC bus 

utilization and harmonic profile. The SVM has better 

DC bus utilization of 15.89% as compared to sine 

PWM. It has an output voltage THD of 63.69% which 

is lesser as compared to sine PWM. The efficiency of 

the drive with SVM and Sine PWM is also computed 

in Table 1.  For efficiency calculation, the electrical 

input power of the drive is computed using discrete 3-

phase Positive-Sequence Active and Reactive Power 

block in MATLAB Simulink. It is seen that the 

efficiency of the drive is improved with SVM as 

compared to the drive with sine PWM. But 

conventional SVM is computationally complex which 

involves many sector calculations. The proposed CB-

PWM-CMV shows a similar performance as that of 

SVM but it is simple and less complex. Hence, the 

proposed PWM technique has better DC bus 

utilization, a better harmonic profile and a better 

efficiency. Besides, it is computationally simple and 

offers easy on-line digital implementation. 

     The experimental set up is built up to validate the 

performance of CB-PWM-CMV. The prototype 

Laboratory setup is shown in Fig. 4. The CB-PWM-

CMV is implemented on Spartan-6 FPGA board. The 

clock frequency for FPGA implementation is chosen 

as 20MHz. The reference sinusoidal signals are stored 

in a look up table with 8-bits precision as an integer. 

The duty cycle profiles are generated with 8-bits 

precision from the reference sine signals using the 

proposed CB-PWM-CMV technique. The VHDL 

coding is developed to realize the CB-PWM-CMV.  

The gate pulses are given to the three phase inverter 

through  the  pulse  driver  circuit and opto-isolator to  

 

drive the induction motor. The switching is chosen as 

10kHz. The single phase bridge rectifier [DB107], 

Buffer IC [CD4081], Opto-isolator [6N137] are used 

for hardware implementation. Two coupling 

capacitors of 330 μF connected in parallel (equivalent 

capacitor value is 660 μF) are used. The dead time for 

the inverter is chosen as 6μs. 

The performance of CB-PWM-CMV fed induction 

drive is tested for various operating conditions. The 

sample results obtained are presented. The modulating 

waveform and switching pulses obtained from the 

FPGA processor are shown in Fig. 5(a) and Fig. 5(b) 

respectively. The inverter output voltage obtained for 

modulation index=0.96 and frequency=50Hz is shown 

in Fig. 6(a). The inverter output voltage obtained for 

modulation index=0.48 and frequency=25Hz is shown 

in Fig. 6(b). 

     The two-level inverter is used in this paper. Hence, 

two levels are obtained in the output line-line voltage. 

It is observed that as soon as the ratio of m/f is 

reduced to 50%, the fundamental output voltage 

magnitude and frequency of the line-line voltage is 

also reduced to around 50%. The effect of variation in 

the fundamental output voltage magnitude and 

frequency for various m/f ratios is reflected in the 

rotor speed. Since the rotor speed can be easily 

measured using tachometer, the rotor speed for 

various m/f ratios is shown in Table 2. The measured 

speed increases proportionally with increase in m/f 

ratio. From this table, it is shown that the drive 

performs very well under a practical condition with 

CB-PWM-CMV for various m/f ratios. 

 

 

 

 

 
 

 

Figure 4. Hardware setup of SVM based inverter fed IM drive with FPGA controller. 

 

 

67 



The Journal of Engineering Research (TJER), Vol. 16, No. 1 (2019) 63-76 

 

 
(a) 

 
(b) 

 

Figure 5. (a) Modulating waveform (b) Pulses to IGBT switches. 

 

 

 
(a) 

 
(b) 

 

Figure 6. (a) Inverter output voltage (L-L) for m=0.96,f =50 Hz  (b) Inverter output (L-L) for m=0.48, f=25 Hz. 

68 



A.Venkadesan 

  

 

 

2.2 Adaptive Neural Network Based Rotor 
Resistance Estimator   

     The design of an indirect vector controller, to a 

large extent, depends on the accuracy of the rotor 

resistance. The slip frequency depends on the 

knowledge of rotor resistance which is evident from 

(3). The transformation angle or flux angle depends 

on the slip frequency which is evident from (4). In 

real time, the rotor resistance may vary up to 100 due 

to the rotor heating Karanayil B et al. (2007). Any 

mismatch between the rotor resistance of motor and 

the  rotor resistance  value  in  the  vector  controller  

leads to the loss of the decoupled control (Chitra A et 

al. 2015; Karanayil B et al. 2007). Hence to track 

variation in the rotor resistance, the rotor resistance 

estimator using Model Reference Adaptive system 

(MRAS) with adaptive neural network scheme is 

proposed. For simulation study, the motor is modeled 

in d-q frame to introduce variation in the rotor 

resistance.     

The proposed MRAS scheme for rotor resistance 

estimation is shown in Fig. 7. The MRAS uses a 

voltage and a current model for rotor resistance 

estimation. The output from both models is d and q-

axis rotor fluxes in stationary reference frame. It is 

well known that the voltage model is independent of 

rotor resistance and the flux estimated will be 

independent of the rotor resistance. Hence, the voltage 

model is used as a reference model. Accordingly,  the 

equations (10) and (11) do not contain rotor 

resistance. The three phase voltages and currents are 

measured and transformed into d and q axis stationary 

reference frame and given as inputs to the voltage 

model. 

The current model is dependent on the rotor 

resistance and the flux estimated will be dependent on 

the rotor resistance. Therefore, the current model 

equation can be used as the adaptive model. 

Consequently, the equations (12) and (13) contain 

rotor resistance.  The three currents and rotor speed 

are measured. The three phase currents are 

transformed into d and q axis stationary reference 

frame. As a result, the inputs to the current model are 

d and q axis current and rotor speed. 

The current model equations (12), (13) are 

represented as a two-layer neural model and used as 

the adaptive model because it is dependent on the 

rotor resistance. The weight of the adaptive NN model 
 

Table 2. Measured Rotor speed with CB-PWM-CMV 

technique for various m/f ratios. 

 

Modulation 

Index 

Frequency (Hz) Measured Rotor 

Speed (RPM) 

0.96 50 1478 

0.768 40 1176 

0.48 25 743 

0.325 16.93 498 

 
is the function of the rotor resistance (Rr). The error 

between the reference model flux and the adaptive 

model flux is back propagated to adjust the weights of 

the adaptive model to estimate the rotor resistance. 

The error between the reference and adaptive model is 

minimized using the BPM (Back Propagation with 

Momentum Algorithm). 

 
2 2

r r s r s m
s

m m r s

L L L L -L Ls s s s= ( -R )dtφ v i idr ds ds dsL L L L

  
           

   (10) 

 
2 2

r r s r s m
s

m m r s

L L L L -L Ls s s s= ( -R )dtφ v i iqs qsqsqr L L L L

  
       

   
   (11) 

 

r m r
r

r r

s
φ R  L R s sdr s= - -ωφ φi qrds drdt L L

          (12) 

 

r m r
r

r r

s
φ R  L Rqr s ss= - +ωφ φiqs qr drdt L L

          (13) 

 

     The discrete form of Current model equations can 

be obtained using backward Euler’s rule. Using this 

rule, the rate of change of rotor flux can be expressed 

as in (14) 

 

   s 's 'sT T 1φ φ φ
dr dr dr=

dt T
s

 
                 (14) 

 

     The discrete form of Current model equations can 

be obtained using backward Euler’s rule and are given 

in (14) and (15) 

    
s

1 2 3 ds
's 's 's(T) = W (T-1)-W (T-1)+W  i (T-1)φ  φ  φqrdr dr

      (15) 

s

1 2 3 qs
's 's 's(T) = W (T-1)+W  (T-1)+W  i (T-1)φ  φ φqr qr dr

     (16) 

 

Equations (14) and (15) can be represented as a 

two-layer neural network with a linear activation 

function (Fig. 8). This is used as the adaptive model. 

It contains four inputs { 's (T-1)φ
dr

, 's (T-1)φqr ,
s (T-1)ids

,

s (T-1)iqs } and two outputs {
's (T)φ
dr

, 's (T)φqr }. The 

adaptive model is used in a predictive mode instead in 

simulation mode. In predictive mode. In the predictive 

mode, the past values of rotor flux required as the 

input for the adaptive model is obtained from the 

reference voltage model itself rather than from the 

adaptive model as in simulation mode. This gives ease 

in digital implementation of the adaptive model and 

overcomes the problem of instability Maurizio 

Cirrincione et al. (2005). The inputs and outputs are 

connected through the weights w1, w2, w3. The weight 

w2 is kept constant. The other two weights w1 and w3 
are proportional to the Rr they are updated using BPM 

69 



The Journal of Engineering Research (TJER), Vol. 16, No. 1 (2019) 63-76 

  

to minimize the error E (17). 

The BPM is coded in MATLAB-m-file. The 

learning rate () and momentum factor () are chosen 

as 5×10
-5

 and 2×10
-5

 respectively. The E is chosen as 

1×10
-6

.  The weights of NN are updated using (20) 

and (21).  

 

2 2s 's
φ φ(T) (T)r r

1
E=  [ε(T)] =0.5 [ - ]  

2
         (17) 

1 d q
s sΔw (T) = α [ε (T) (T-1) + ε (T) (T-1)] φ  φqrdr

         (18) 

 

3 d q
s sΔw (T) = [ε (T) (T-1) + ε (t) (T-1)]i  iqsds

         (19) 

 

1 1 1 1
w (T)=w (T-1)+α Δw (T)+η Δw (T-1)          (20) 

 

3 3 3 3
w (T)=w (T-1)+α Δw (T)+η Δw (T-1)          (21) 

 

where, 

 

s
φ (T)r

s (T)φ
dr

=
s (T)φqr

 
 
 
  

; 'sφ (T)r

's (T)φ
dr

=
's (T)φqr

 
 
 
  

; 

d
s 'sε (T)= (T)- (T)φ φ
dr dr

; 
q

s 'sε (T)= (T)- (T)φ φqr qr
 

 

The rotor resistance can be obtained either from 

w1 or w3 (Chitra A et al. 2015; Karanayil B et al. 

2007). 

 

r 3
r

m s

L  w
R  = 

L  T

 
 
 

                               (22) 

 rr 1
s

L
R  = 1-w

T
                          (23) 

     The effect of sampling time on the performance of 

the proposed rotor resistance estimator assumes 

importance for on-line digital implementation. Hence, 

the same is tested for various sampling time. The 

drive is operated with the speed of 148 rad/s under 

100% load torque. The rotor resistance is increased to 

50%. The switching frequency of the inverter is 

chosen as 10kHz. Hence the switching time is 100µs.  

 

 

 
 
Figure 7.   NN-MRAS scheme for rotor resistance estimation. 

 

   
 

Figure 8.  Two-layer neural network.  

 

 

 

70 



A. Venkadesan 

  

 

The steady state error is computed between the 

actual and estimated rotor resistance for various 

sampling time and presented in Table 3. Lesser 

sampling time leads to more accuracy but increases 

computation time as the number of samples per cycle 

is more. On the other hand, larger sampling time 

decreases computation time but it decreases accuracy 

as the number of samples per cycle is less. In this 

paper, the sampling time is chosen same as the 

inverter switching time for study. 

The estimator is tested for a step change in rotor 

resistance. The actual and estimated rotor resistance 

for step variation (24) is shown in Fig. 9(a). The 

sample weight evolution plot for step change is also 

shown in Fig. 9(b). It is understood that the network 

learns to estimate the rotor resistance. The weight and 

the rotor resistance are linearly related. The step 

change is the extreme case condition. It is used to 

show the capability of the proposed estimator to 

estimate Rr even in the extreme case condition.  

But in the practical real time, the rotor resistance 

varies slowly.  Accordingly,  the proposed estimator is 

tested for trapezoidal variation (25) in the Rr and also 

with exponential variation Beguenane R et al. (1999) 

as in (26). The result obtained for trapezoidal 

variation and exponential variation is shown in the 

Fig. 9(c) and Fig. 9(d) respectively. The NN-MRAS 

tracks the variation in Rr for the various rotor 

resistance variation profile very well with good 

accuracy. 

 










sec5.11mno,tRmno,tR

sec5.11mno,tR
)t(rR       (24) 

 





























sec4tsec3
)mno,rRmno,rR(mno,rR

t)mno,rR(

sec3tsec2mno,rRmno,rR

sec2tsec1
mno,rR)mno,rRmno,rR(

)mno,rR(t

sec)4t(orsec)1t(mno,rR

)t(rR

                                                                                 (25) 

)
t2.1

1(mno,rR)t(tR


 e                                     (26) 

 

     The estimated Rr value for various %changes in Rr 

is consolidated and presented in the Table 4 along 

with the steady state error for 100% and 50% load 

torque. It is found that the proposed estimator tracks, 

the change in Rr with good accuracy. 

 

2.3 Adaptable Neural Network Based Rotor 
Resistance Estimator  

     The performance of the drive with and without a 

rotor resistance estimator is studied. The motor is 

operated with the speed of 148 rad/sec, torque of 7.5 

Nm and rotor flux of 0.9 Wb. The step change in rotor 

resistance is effected at 1.5 sec. Without the Rr 

estimator, the rotor flux, torque and three phase stator 

current are shown in the Fig. 10 (a), Fig. (b), Fig. (c) 

respectively.  It  is  observed   that   without  the  rotor  

resistance estimator, the error  in the rotor flux is 

found to be 16.66% and the actual flux deviates from 

the reference flux. The ripple in the machine torque is 

found to be 30.46% and oscillates about the load 

torque which is evident from Fig. 10(b). The 

sinusoidal shape of the three phase stator current gets 

distorted. This type of response is not desirable in  

high performance drives. With the rotor resistance 

estimator, the actual rotor flux tracks the reference 

flux with the error of 0.133% after about 0.4sec as 

soon as the step change in rotor resistance is applied at 

1.5sec which is evident in Fig. 11(a). With the rotor 

resistance estimator, the ripple in the machine torque 

is reduced significantly to 3.382% which can be 

observed in Fig. 11 (b). 

     The sinusoidal shape of the current is maintained 

which can be observed from in Fig. 11(c). Hence, the 

performance of the CB-PWM-CMV based vector 

controlled IM drive is significantly enhanced with 

rotor resistance estimator. 

 
Table 3. Effect of sampling time on the rotor resistance 

estimator for switching frequency 10kHz. 

 

Sampling 

Time 

Actual 

Rr 

(ohms) 

Estimated 

Rr using 

Proposed 

NN-

MRAS 

Scheme 
(ohms) 

%Steady 

State 

Error 

50µs 9.127 9.123 0.043826 

100µs 9.114 0.142435 

200µs 9.052 0.821738 

300µs 8.958 1.851649 

400µs 8.828 3.275994 

500µs 8.657 5.149556 

 

3. CONCLUSION 
 

A carrier based PWM technique and Neural Network 

based rotor resistance estimator are proposed for the 

performance enhancement of induction motor drives. 

The CB-PWM-CMV technique is computationally 

less complex and gives ease in digital implementation 

as compared to space vector modulation technique. 

The CB-PWM-CMV technique is shown to improve 

the efficiency of the drive as compared to Sine-PWM. 

The CB-PWM-CMV technique is implemented on 

Spartan-6 FPGA processor and experimentally 

validated.  

     A neural Network based Model Reference 

Adaptive system in prediction mode is proposed for 

the rotor resistance estimation. The proposed NN-

MRAS is shown to estimate Rr with good accuracy 

which is evident from Table 4. With rotor resistance 

estimator, the performance of the drive is significantly 

improved. Hence, it is concluded that the performance 

of the drive is enhanced significantly with CB-PWM-

CMV and rotor resistance estimator. 

71 



The Journal of Engineering Research (TJER), Vol. 16, No. 1 (2019) 63-76 

  

 

 
(a) 

 
(b) 

 
(c) 

 
(d) 

 

Figure 9. (a) Rotor resistance for step variation (b) Weight evolution plot for step variation (c) Rotor resistance for 

trapezoidal variation (d) Rotor resistance for exponential variation. 

 

72 



A.Venkadesan  

 

 
(a) 

 
(b) 

 
(c) 

 

Figure 10. Without Rr Estimator: (a) Rotor Flux (b) Torque (c) Current. 

 

 

 

 

73 



The Journal of Engineering Research (TJER), Vol. 16, No. 1 (2019) 63-76 

 

 
(a) 

 
(b) 

 
(c) 

 

Figure 11. With Rr Estimator: (a) Rotor Flux (b) Torque (c) Current. 

74 



The Journal of Engineering Research (TJER), Vol. 16, No. 1 (2019) 63-76 

  

 

 
Table 4. Rotor resistance estimation using NN MRAS for various % changes in rotor resistance. 

 

 

%
 V

a
r
ia

ti
o

n
 i

n
 R

r
 

A
c
tu

a
l 

R
r
 

(o
h

m
s)

 

100% Load Torque 50% Load Torque 

E
st

im
a

te
d

 R
r
 u

si
n

g
 

P
r
o

p
o

se
d

 N
N

-

M
R

A
S

 S
c
h

e
m

e
 

(o
h

m
s)

 

%
S

te
a

d
y
 S

ta
te

 

E
r
r
o

r
 

E
st

im
a

te
d

 R
r
 u

si
n

g
 

P
r
o

p
o

se
d

 N
N

-

M
R

A
S

 S
c
h

e
m

e
 

(o
h

m
s)

 

%
S

te
a

d
y
 S

ta
te

 

E
r
r
o

r
 

10 6.69 6.672 0.328653 6.663 0.463101 

20 7.30 7.284 0.246508 7.264 0.520405 

30 7.91 7.983 -0.92288 7.889 0.265487 

40 8.51 8.500 0.223031 8.487 0.375631 

50 9.12 9.114 0.142435 9.098 0.317739 

60 9.73 9.713 0.236237 9.711 0.256779 

70 10.34 10.33 0.096712 10.32 0.193424 

80 10.95 10.94 0.091324 10.92 0.273973 

90 11.56 11.54 0.17301 11.53 0.259516 

100 12.17 12.15 0.164339 12.14 0.246508 

 

 

CONFILCT OF INTEREST 
 

Author declares no conflicts of interest. 

 

FUNDING 
 

No funding was received for this research. 

 

ACKNOWLEDGMENT 
 

I thank Dr. Smrutisikta Mishra, Assistant Professor, 

Department of Humanities and Social Sciences, NIT 

Puducherry, Karaikal for her valuable suggestions in 

English language. 

 

REFERENCES 
 

Abdelkarim Ammar, Amor Bourek, Abdelhamid 

Benakcha (2017), Sensorless SVM-Direct torque 

control for induction motor drive using sliding 

mode observers. Journal of Control, Automation 

mand Electrical Systems 28(2): 189-201. 

Ahmed A.Z.D (2014), Real-time implementation of 

full-order observer for speed sensorless vector 

control of induction motor drive. Journal of 

Control. Automation and Electrical Systems 25 

(6): 639-648. 

Beguenane R, Benbouzid M. E. H (1999), Induction 

motors thermal monitoring by means of rotor 

resistance identification. IEEE Transactions on 

Energy Conversion 14(3): 566-570. 

Benchabane F, Titaouine A, Bennis O, Yahia K, Taibi 

D (2012), Direct field oriented control scheme for 

space vector modulated AC/DC/AC converter fed  

 

       induction motor. Frontiers in Energy 6(2): 129- 

       137. 

Bimal K. Bose (2005), Modern power electronics and 

AC drives. Prentice Hall of India. 

Chitra A, Himavathi S (2015), A modified neural 

learning algorithm for online rotor resistance 

estimation in vector controlled induction motor 

drives. Frontiers in Energy 9(1): 22-30. 

Durgasukumar G, Pathak M.K (2012), Comparison of 

adaptive neuro-fuzzy-based space-vector 

modulation for two-level inverter. Electrical 

Power and Energy Systems 38: 9-19. 

Govindasamy Renukadevi, Kalyanasundaram 

Rajambal (2014), Field programmable gate array 

implementation of space-vector pulse-width 

modulation technique for five-phase voltage 

source inverter. IET Power Electronics 7(2): 376-

389. 

Hannan M. A, Jamal Abd Ali, Azah Mohamed, 

Mohammad Nasir Uddin (2017), A random forest 

regression based space vector PWM inverter 

controller for the induction motor drive. IEEE 

Transactions on Industrial Electronics 64(4): 

2689-2699. 

Karanayil B, Rahman M.F and Grantham C (2007), 

Online stator and rotor resistance estimation 

scheme using artificial neural networks for vector 

controlled speed sensorless induction motor drive. 

IEEE Transactions on Industrial Electronics 54 

(1): 167-176. 

Keliang Zhou and Danwei Wang (2002), Relationship 

between space-vector modulation and three-phase 

carrier-based PWM: A comprehensive analysis. 

IEEE Transactions on Industrial Electronics 49 

75 

 



The Journal of Engineering Research (TJER), Vol. 16, No. 1 (2019) 63-76 

 

(1): 186-196. 

Krishnan R (2007), Electric motor drives modeling, 

analysis and control. Prentice Hall of India. 

Mahmoud Gaballah, Mohammed El-Bardini (2013), 

Low cost digital signal generation for driving 

space vector PWM inverter. Ain Shams 

Engineering Journal 4: 763-774. 

Maurizio Cirrincione, and Marcello Pucci (2005), An 

MRAS-based sensorless high-performance 

induction motor drive with a predictive adaptive 

model. IEEE Transactions on Industrial 

Electronics 52(2): 532-551. 

Peter Vas (1998), Sensorless vector and direct torque 

control. Oxford University Press. 

Sabah V.S, Charles Baby T, Krishna Prabhakar Lall, 

Balamurugan M,Sarat Kumar Sahoo, Chitra 

A,Prabhakar, Karthikeyan, I. Jacob Raglend 

(2015), A simplified space vector pulse width 

modulation technique for multilevel inverter fed 

induction motor drive. IEEE International 

Conference on Circuit, Power and Computing 

Technologies (ICCPCT). 

Shriwastava R.G, Daigavane M. B, Daigavane P. M  

(2016), Simulation analysis of three level diode 

clamped multilevel inverter fed PMSM drive 

using carrier based space vector pulse width   

modulation. (CB-SVPWM),  Procedia Computer 

Science (Elsevier). 79: 616-623. 

Srirattanawichaikul W, Kumsuwan Y, 

Premrudeepreechacharn S, Wu B (2011), Carrier-

based PWM strategy for three-level neutral-point-

clamped voltage source inverters. IEEE PEDS 

2011, Singapore, 948-951. 

Venkadesan A, Himavathi S and Muthuramalingam A 

(2016), A novel NN based rotor flux MRAS to 

overcome low speed problems for rotor resistance 

estimation in vector controlled IM drives. 

Frontiers in Energy 10(4): 382-392. 

Venkadesan A, Himavathi S, Sedhuraman K, 

Muthuramalingam A (2017), Design and field 

programmable  gate   array   implementation    of  

cascade neural network based flux estimator for 

speed estimation in induction motor drives. IET 

Electric Power Applications 11(1): 121-131. 

Witold P, Huynh V.K, Michael R.H (2017), 

Temperature rise estimation of induction motor 

drives based on loadability curves to facilitate 

design of electric powertrains. IEEE 

Transactions on Industrial Informatics 13(3): 

985-994. 

 

 

Appendix 1 
 

 

Induction Motor Parameters: 3 phase, 4 poles, 1100W, 415V, 50Hz, 7.5Nm. 

 

 

Rs 

Rr 

Lm 

Lsl 

Lrl 

J 

B 

6.03Ω 

6.085Ω 

0.4893H 

0.0138H 

0.0138H 

0.011787kg.m
2 

0.0027kg.m
2
/s 

 
 

 

 

 
 

 

 

76