This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIA.2014.2300160, IEEE Transactions on Industry Applications
A Hybrid Model for Improved Hysteresis Loss
Prediction in Electrical Machines
Maged Ibrahim, Student Member, IEEE, Pragasen Pillay, Fellow IEEE
Abstract— This paper presents a model for calculating the
hysteresis loss in electrical machines. The model can achieve
accurate and computationally efficient hysteresis loss calculations
by utilizing both analytical equations and the Energetic
hysteresis model. The model results are verified experimentally
by comparing to a series of minor hysteresis loop measurements.
The hybrid model is then implemented to calculate the core losses
in a switched reluctance (SR) machine using finite element (FE)
simulation. The results show that precise machine core loss
prediction requires having a model that is capable of calculating
the hysteresis losses under a variety of minor hysteresis loops.
Index Terms— Core loss, Hysteresis loss, Minor loops, Energetic
model, Switched reluctance machine.
I. INTRODUCTION
The design of high efficient electrical machines requires an
optimization procedure for minimizing the machine losses. In
order to achieve the optimum design, accurate core loss
calculations have to be performed for each candidate design.
Therefore, it is essential to have an accurate and fast core loss
model in order to optimize the machine within a convenient
computation time.
One of the challenges of core loss prediction in electrical
machines is the calculation of the hysteresis losses under
distorted flux. The flux waveforms inside many machines,
e.g., permanent magnet (PM) machines and switched
reluctance (SR) machines are naturally non-sinusoidal and
contain significant harmonic content and considerable DC
component in some regions. Depending on the phase and
magnitude of these harmonics, the resulting flux waveforms in
the machine core may contain local flux reversals causing
minor hysteresis loops to occur inside the main loop. The
minor hysteresis loops can also be produced by non-sinusoidal
supplies, e.g., bipolar pulse width modulation (PWM)
inverters [20].
Many empirical formulas are presented in the literature to
evaluate minor loop hysteresis losses [1]-[4]. These formulas
can provide reasonable estimates of the hysteresis losses under
certain conditions. However, they cannot be relied on to
This work was supported by the Natural Sciences and Engineering
Research Council of Canada and by Hydro-Québec.
M. Ibrahim is with the Department of Electrical and Computer
Engineering, Concordia University, Montreal, QC H3G 1M8, Canada, (e-
mail: ma_ib@encs.concordia.ca).
P. Pillay is with the Department of Electrical and Computer Engineering,
Concordia University, Montreal, QC H3G 1M8, Canada, and also with the
University of Cape Town, Cape Town 7701, South Africa (e-mail:
pillay@encs.concordia.ca).
0093-9994 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
predict minor loop losses under a large variety of practically
encountered flux waveforms in electrical machines. The only
way these formulas can achieve accurate minor loop loss
prediction is to use variable coefficients based on a database
of minor loop measurements [1], [5]. Such a database is
difficult to provide by steel manufacturers, as the minor
hysteresis loop losses are dependent on multiple factors.
Many hysteresis loss models are available for hysteresis
loop modeling, ranging from purely mathematical models [6]
to physics-based models such as the Preisach model [7, 8], the
Jiles-Atherton model [9], and the Energetic model [10]
Although the Preisach model can achieve accurate loss
prediction [11], the long computation time makes it unsuitable
for finite element (FE) electrical machine design, which
requires fast core loss calculation in each mesh element. The
Jiles-Atherton model is adopted in [21] to calculate the static
hysteresis losses of induction motors. The results demonstrate
the importance of considering the harmonics initial phase in
the machine core loss calculation.
In this paper, a hybrid model is proposed to calculate the
hysteresis losses under any flux waveform. The model utilizes
both analytical formulas and the Energetic hysteresis model to
achieve accurate and computationally efficient hysteresis loss
prediction.
Section II describes the technique used for measuring minor
hysteresis loops. Section III presents three models for
hysteresis loss calculations; an analytical model, Energetic
hysteresis model and a hybrid model. In section IV, the
hysteresis losses of a 6/4 SR machine are examined using the
three models.
II. MEASUREMENT CONCEPT
Since it is practically impossible to predetermine the
general properties of the minor loops generated in electrical
machine laminations, the core loss model used should be able
to calculate minor loop losses under all possible conditions.
As illustrated in Fig. 1, the hysteresis loss caused by a minor
loop in a certain magnetic material is dependent on four
factors,
i. The magnitude of the minor loop B .
ii. The position of the minor loop B m .
iii. The peak flux density of the major loop B p .
iv. The quadrant in which the minor loop occurs.
While the shape of the minor loop on the ascending part of
the hysteresis loop is different from the minor loop shape on
the descending part, it is observed from a series of minor loop
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIA.2014.2300160, IEEE Transactions on Industry Applications

1.5 1.5

1 1
B B
Bp dB
0.5
Bm Bm Bp
0.5
Flux density (T)

dt

Flux density (T)

0 0
T1 T2 T3 T4

1.5
1.5

0
0

-1.5
-400 -300 -200 -100 0 100 200 300 400 -1.5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
Field intensity (A/m)
Time (s)
Fig. 1. Measured major and minor hysteresis loops.
Fig. 3. Flux density waveform required to generate a certain minor loop.

measurements that the influence of the minor loop quadrant on
the hysteresis loss is negligible. The same observation is also
reported in [1]. Therefore, a series of minor loop
measurements is performed with all combinations of various
values of the first three parameters B , B m and B p .
The minor loops can be generated in the laminations by
imposing high frequency harmonics to the applied sinusoidal
excitation waveform. The magnitude and position of the minor
loop can be controlled by changing the magnitude and phase
shift of the harmonic frequency. This method is used in [1-3]
for minor loop measurements. However, the measured
hysteresis loops with this method represents the dynamic
hysteresis loops, which includes both the hysteresis and eddy
current losses. Although the hysteresis loss component can
still be separated from the total measured loss, the separated
minor loop hysteresis loss is less accurate, as it includes the
core loss separation errors in addition to the measurement
errors. In order to achieve accurate hysteresis loss
measurements, the excitation waveform must have small
magnetization rate, so that the measured loops would represent
the quasi-static hysteresis losses.
Neglecting skin effect, the instantaneous eddy current loss in a
the lamination can be represented by,
2
 dB 
Pe (t )  K e   (1)
 dt 
where K e is dependent on the material electrical conductivity
and the lamination thickness. In order to keep the
instantaneous eddy current loss constant throughout the
magnetization cycle, the measurements are performed under a
controlled rate of change of magnetization. To generate a
minor loop of a certain magnitude B , position B m and peak
flux density B p , the flux reversal times are calculated based
on a constant dB/ dt and the corresponding flux waveform is
generated in the laminations, as shown in Fig. 2. The applied
voltage is then increased until the desired peak flux density
B p is reached. The hysteresis losses caused by the major and
minor loops are then calculated by measuring the area
enclosed by each loop.
The measurements are performed using an Epstein frame
test fixture. The excitation signal is generated by Matlab
Simulink interfaced with a dSPACE board, and applied to a
high bandwidth amplifier, which excites the Epstein frame
primary winding. The Epstein frame primary current and
secondary voltage are measured and sent back to Matlab
Simulink in order to calculate the hysteresis loops.
III. HYSTERESIS LOSS MODELING
A. Analytical
For the cases where the flux waveforms in the machine core
are symmetric and contain only two flux reversals per cycle,
the hysteresis energy loss can be represented by the modified
Steinmetz equation [12] as,
a  bBˆ  cBˆ
Wh  K h Bˆ p p p
2
(2).
The symmetric hysteresis losses for 0.5 mm-thick silicon
iron (M45G26) are measured and the parameters K h , a ,
b and c are then obtained from a curve fit of measured
hysteresis loss data. The extracted parameters are shown in
table I. As shown in Fig. 3, the calculated hysteresis energy
loss by equation (2) agrees well with the measured data.
The total hysteresis loss is then calculated by simply
multiplying the static hysteresis energy loss by the
operating frequency. However, this simplification is only
valid at lower frequencies, where skin effect is negligible,
as skin effect causes the peak flux density to vary across the
lamination causing the local hysteresis loops, and therefore
the local hysteresis energy loss per cycle to differ at
different points inside the lamination. The hysteresis loss at
high frequencies can be calculated by constructing the
magnetic field distribution inside the lamination, as shown

0093-9994 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIA.2014.2300160, IEEE Transactions on Industry Applications

TABLE I B / B p  1 , and in the unsymmetrical flux waveforms in the

Extracted coefficients for M45G26 steel
SR machine stator core, where B / B p  1 . In addition, the
Kh a b c formula does not account for the minor loop position B m ,
0.015 1.846 -0.585 0.480 which has a noticeable effect on the measured minor loop
losses, as shown in Fig. 4. One example of minor loops
0.06 occurring at different positions of the major loop is the
Measured
0.0533 generated hysteresis loops in laminations exposed to bipolar
Modified Steinmetz
PWM excitation. Fig. 5 shows the measured loops under
Hysteresis energy loss (J/kg)

0.0467
bipolar PWM with modulation frequency ratio mf = 21 and
0.04 modulation index ma = 0.6.
0.0333 Figs. 6 and 7 show the errors of the calculated losses
0.0267
relative to the measured losses for B p equal to 1.2T and 1.6T,
0.02 respectively. It can be seen that the model results have
acceptable errors for only a few cases of minor loops, and the
0.0133
prediction errors can be as high as 150% for minor loops with
0.0067
different positions and magnitudes. Therefore, the model
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
cannot be relied on to predict minor loop losses under a
variety of possible encountered flux waveforms in electrical
Flux density (T)
machines. Subsequently, an alternative method has to be
Fig. 4. Comparison between measured symmetric major loop hysteresis implemented in order to achieve accurate hysteresis loss
energy losses with the losses calculated by equation (2). prediction.
by the authors in [13]. However, this procedure is 0.04
Measured (Bm = Bp)
computationally expensive when the flux density waveforms 0.0356 Measured (Bm = 0.8 Bp)
are non-sinusoidal, as it becomes difficult to obtain the flux Measured (Bm = 0.6 Bp)
Hysteresis energy loss (J/kg)

0.0311
density distribution using the analytical models. Therefore, Calculated by Lavers formula
this method is not suitable for core loss determination in FE 0.0267
machine design, which requires fast core loss calculations at
0.0222
each mesh element. Therefore, the total hysteresis energy loss
is assumed to be only dependent on the flux density amplitude 0.0178

and the flux reversal points. 0.0133

While equation (2) can calculate the symmetric major loop
0.0089
hysteresis loss, the flux waveforms in many electrical
machines may contain additional flux reversals and 0.0044

considerable DC components. The loss prediction in these 0

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
machines requires a separate model for calculating minor
hysteresis loop losses, since equation (2) fails to predict these B / Bp
losses. A popular model for minor loop hysteresis loss Fig. 5. Comparison between measured minor loop losses at different
prediction in electrical machine design is the model developed positions with the losses calculated by equation (3) for Bp = 1.4 T .
by Lavers [2]. According to the model, the hysteresis loss 1.5

caused by a minor loop can be calculated by,

k B 1
Whm  Wh  Bp p  
B

2 Bp (3).
0.5
Lavers suggested that a value of the coefficient k between
Flux density (T)

0.6 and 0.7 is suitable for the cases where B p is in the range
0
of 1.0T to 2.0T and the ratio of B / B p is relatively low. The
hysteresis losses calculated by (3) with k equal to 0.65 are 1.5

compared with the measured minor loop losses for various

combinations of B , B m and B p . As shown in Fig. 4, the 0

model can provide a reasonable estimate of the hysteresis

-1.5
losses caused by relatively small minor loops occurring at the -400 -300 -200 -100 0 100 200 300 400
Field intensity (A/m)
tip of the major loop. However, it underestimates the losses
caused by larger minor loops; these loops can be found in the Fig. 6. Measured hysteresis loops under bipolar PWM excitation with
modulation frequency ratio mf = 21 and modulation index ma = 0.6.
unipolar flux of the SR machine stator pole where

0093-9994 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIA.2014.2300160, IEEE Transactions on Industry Applications

200 computationally expensive. The Jiles-Atherton model is easier

Bm = Bp to implement and can achieve faster hysteresis loss calculation
Bm = 0.8 Bp than the Preisach model. However, the simulation still requires
150 Bm = 0.6 Bp
integration over dH . On the other hand, the Energetic model
Bm = 0.4 Bp
Bm = 0.2 Bp simulation of the hysteresis loops is suitable for fast core loss
100
calculations in electrical machine FE simulations as the
% Error

magnetic field H at each time step can be calculated directly

by only one equation from the flux density waveforms, which
50 are available in the post- processing stage. In addition, the
Energetic model is capable of considering the dependence of
magnetization on temperature, stress, and magnetization
0
direction. These parameters are obtainable by means of FE
simulation, and their influence on the machine core loss can be
 50
considered by the model. The Energetic model is found to be
0 0.5 1 1.5 2
capable of simulating the major hysteresis loops of electrical
B / Bp steels [14]. In this section, the model capability to predict
Fig. 7. Errors in the minor loop hysteresis losses calculated using minor hysteresis loop losses is investigated.
equation (3) relative to the measured losses for Bp = 1.2 T.

200
Bm = Bp In the Energetic model, the magnetic field H is calculated
Bm = 0.8 Bp from the relative magnetization ( m  M / M s ) by,
150 Bm = 0.6 Bp
Bm = 0.4 Bp
Bm = 0.2 Bp

H  N e M s m  sgn( m ) h 1  m 
1 m
1  m 1 m 
g /2

1
 k    q 
100  sgn( m  m 0 )  C r H r   1   exp   m  m 0  
 
% Error

 0 s
M    
(4).
50
The first term of equation (4) represents the linear material
behavior with N e , M s being the demagnetization factor and
0 saturation magnetization. The second term represents the no-
linear material behavior with h and g relating to saturation
field and anisotropy. The third term describes the hysteresis
 50
0 0.5 1 1.5 2 effects, with k relating to hysteresis loss, q to the pinning site
B / Bp density, Cr to the grain geometry. The reversible field
Fig. 8. Errors in the minor loop hysteresis losses calculated using function H r is calculated by,
equation (3) relative to the measured losses for Bp = 1.6 T.

B. Energetic model H r  h{[(1  m ) 1 m (1  m ) 1 m ] g / 2  1} (5).

The reason behind the failure of the analytical formulas to
predict minor hysteresis loop losses is that the magnetic As suggested in [10], an improvement of minor loop
hysteresis process is such a complex phenomenon that it is representation can be achieved by modifying the hysteresis
impossible to accurately predict its loss under any arbitrary loss parameter k based on the peak magnetization M p to,
flux waveform using a single empirical formula. Therefore, (M p / M s )  1
the hysteresis model has to be adapted in order to account for k  k0 (6).
the non-linear behavior of the magnetic material under 2
distorted excitations. The function  in equation (4) describes the influence of the
The suitable hysteresis model for core loss calculation in total magnetic state at points of field reversals. The value of
electrical machine FE simulation should have two main  at the field reversal points is calculated based on its
features. Firstly, the model should be able to accurately previous value 0 from,
simulate the major and minor hysteresis loops. Secondly, the
 q 
hysteresis loop simulation should be computationally efficient,   2   0 exp  m  m0  (7).
since the core loss has to be calculated in each mesh element.  0 
The Preisach model can accurately simulate the major and The model calculation starts with m0  0 and   1 , m is
minor hysteresis loops by a properly designed Preisach then increased stepwise and the corresponding value of H at
function [11]. However, the model identification requires a each step is calculated from equation (4). When m reaches
complex procedure, and the hysteresis simulation is the upper field reversal point,  is calculated by equation (7)

0093-9994 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIA.2014.2300160, IEEE Transactions on Industry Applications

and m 0 is set to the value of m at this point. Then m is 1.5

decreased until the lower field reversal point is reached, and Measured
 and m 0 are recalculated. The same calculation procedure is 1 Simulated

executed for major and minor loop simulation. The only

difference is that minor loop calculations are performed using 0.5

Flux density (T)

the modified hysteresis loss parameter k in equation (6).
The extracted Energetic model parameters, shown in table II 0
are obtained from the measured hysteresis loops at low
magnetization rate. Figs. 8-11 compare the measured and 1.5
simulated hysteresis loops for different minor loop
magnitudes B , positions B m and peak flux densities B p . It is
0
can be seen that the simulated loops agree well with the
measured data, qualitatively. The minor loop hysteresis loss is -1.5
then calculated for numerous minor loops of different B , B m -400 -300 -200 -100 0 100 200 300 400
Field intensity (A/m)
and B p . Figs 12 and 13 show the errors of the calculated Fig. 10. Comparison of measured and simulated hysteresis loops for
losses by the Energetic model compared to the measured B  B p , Bm  B p and Bp  1.4T .
losses for B p equal to 1.2T and 1.6T, respectively. It is clear
1.5
that the Energetic model can achieve improved minor loop Measured
loss prediction compared to the analytical model with Simulated
1
maximum error lower than 25%.

0.5
Flux density (T)

TABLE II
Extracted Energetic model parameters
0

Ne 1.189e-5
1.5
Ms 1.432e6
h 7.332
0
g 9.957
k 82.800
-1.5
q 35.110 -400 -300 -200 -100 0 100 200 300 400
Field intensity (A/m)
Cr 0.342
Fig. 11. Comparison of measured and simulated hysteresis loops for
B  0.2B p , Bm  0.4Bp and Bp  1.4T .

1.5
2
Measured
Measured
Simulated Simulated
1 1.5

1
0.5
Flux density (T)

Flux density (T)

0.5

0
0

1.5 -0.5

-1
0

-1.5

-1.5
-400 -300 -200 -100 0 100 200 300 400 -2
Field intensity (A/m) -1000 -800 -600 -400 -200 0 200 400 600 800 1000
Field intensity (A/m)
Fig. 9. Comparison of measured and simulated hysteresis loops for
Fig. 12. Comparison of measured and simulated hysteresis loops for
B  0.4B p , Bm  Bp and Bp  1.4T .
B  0.4Bp , Bm  Bp and Bp  1.6T .

0093-9994 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIA.2014.2300160, IEEE Transactions on Industry Applications

200 prediction in electrical machines. From the flux density

Bm = Bp waveform, the model first extracts the major loop and minor
Bm = 0.8 Bp loop flux reversals. The symmetric major loop hysteresis
150 Bm = 0.6 Bp
losses are then calculated by the modified Steinmetz equation
Bm = 0.4 Bp
Bm = 0.2 Bp
and the minor loop losses are calculated using the Energetic
model. However, in some cases, the machine can experience
100
% Error

particular cases of major loops with large DC component, e.g.

the unidirectional flux in a SR machine stator pole. The
50 hysteresis losses under these waveforms cannot be predicted
by the modified Steinmetz equation, and accurate loss
prediction requires simulating the hysteresis loops by the
0
Energetic model. On the other hand, having a small DC
component in the flux waveform might not also justify using
 50
the Energetic model for the hysteresis loss calculation. In
0 0.5 1 1.5 2
order to find out the DC component at which the hybrid model
B / Bp should switch from the analytical to the Energetic model in the
Fig. 13. Errors in the minor loop hysteresis losses calculated using the
major loop loss calculation, the hysteresis loops with different
Energetic model relative to the measured losses for Bp = 1.2 T. DC components are measured and compared to the calculated
losses by the Energetic model and the modified Steinmetz
200
equation. It can be inferred from the errors, shown in Fig. 15,
Bm = Bp
Bm = 0.8 Bp
150 Bm = 0.6 Bp 20
Bm = 0.4 Bp Modified Steinmetz
Bm = 0.2 Bp 15 Energetic model
100
% Error

10

5
50
% Error

0 5

 10
 50
0 0.5 1 1.5 2
 15
B / Bp
 20
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7
Fig. 14. Errors in the minor loop hysteresis losses calculated using the
Energetic model relative to the measured losses for Bp = 1.6 T.
Flux density (T)
Fig. 15. Comparison of the errors of symmetric major hysteresis loop
C. Hybrid model calculation by the energetic model and the modified Steinmtz equation.
Although the Energetic model is capable of simulating the
40
major hysteresis loops, the model results are not as accurate as Modified Steinmetz
the calculated losses by the modified Steinmetz equation. Fig. Energetic model
20
14 compares the errors of the calculated loss by the Energetic
model with equation (2) errors. It can be seen that precise 0
symmetric major loop loss prediction can be achieved by the
analytical formula without the need for the multiple iterations  20
% Error

required for simulating the hysteresis loop by the Energetic

model. On the other hand, when it comes to minor loop loss  40

prediction, it is obvious that the Energetic model errors in Figs

 60
12 and 13 are much lower than the analytical model errors in
Figs. 6 and 7. Therefore, it is important to use a hysteresis
 80
model in order to achieve accurate minor loop loss prediction,
as the analytical model can only predict minor loop losses for  100
0 0.2 0.4 0.6 0.8 1 1.2
some specific cases of minor loops, and it cannot be relied on
to predict minor loop losses under a large variety of possible DC component (T)
flux waveforms in electrical machines. As a result, a hybrid Fig. 16. Comparison of the errors of the Energetic model and the
model is developed to achieve accurate hysteresis loss modified Steinmetz equation for different DC components.

0093-9994 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIA.2014.2300160, IEEE Transactions on Industry Applications

that improved major loop prediction is achieved by the 40

Energetic model for hysteresis loops with DC components Phase A
larger than 0.34T. Therefore, the hybrid model generally uses 35 Phase B
Phase C
the analytical equation for calculating the major loop losses, 30
except in the cases where the DC component exceeds 0.34T,
as the hybrid model switches from the analytical to the 25

Current (A)
Energetic model for improved major loop loss determination.
20

IV. CASE STUDY 15

In this section, the three previously discussed models are
10
applied to predict the hysteresis losses in a SR machine using
FE simulation. The difficulty of core loss prediction in SR 5
machines is attributed to the nature of the machine flux, as the
flux waveforms vary in different parts of the machine core. 0
0 120 240 360 480 600 720
The flux waveforms also change with the machine
configuration, e.g., number of stator poles, rotor poles, and Rotor position (Dergree)
number of phases. In addition, the flux waveforms are affected Fig. 17. Simulated phase currentsof a 6/4 SR machine at 900 rpm.
by the operating conditions, such as operating speed,
commutation conditions, and the current waveforms [15-19].
In order to obtain the machine flux density waveforms, FE
simulations are performed on a 1.5 kW 6/4 SRM. The
employed FE mesh consists of 7092 triangular mesh elements.
The field distribution is calculated for every 5 mechanical
degrees of mechanical rotation. The simulated phase currents
at 900 rpm are shown in Fig. 16. Fig. 17 shows the simulated
machine flux distribution. In some parts of the machine, the
flux changes its angle relative to a localized axis [19]. The
modeling of core losses under such rotating flux is much more
complicated than the alternating uniaxial flux, and the
rotational loss data are not available from the manufacturers.
However, due to the excitation sequence of the SR machine,
some parts of the machine, which are considered rotational
zones in conventional machines, e.g. the root of the stator
pole, experience uniaxial flux alternating in two directions.
Fig. 18. Flux distribution of a 6/4 SR machine at 900 rpm.
The flux direction in the root of the pole is mostly alternating
radially when the phase is conducting. When the phase is
turned off and the other phases are conducting, the flux 2
direction changes to be almost entirely circumferential. In
order to account for the change in flux direction, the flux 1.5

density waveform in each mesh element is resolved into radial 1

and circumferential components. The hysteresis loss is then
Flux density [T]

calculated by summing the losses produced by the two 0.5

components. 0
Figs. 18 and 19 show the flux waveforms in some mesh
elements in the stator and rotor, respectively. It can be seen  0.5

from Fig. 18 that the flux waveform in the stator pole is 1

unipolar, while the stator core flux contains a DC component
Stator pole
and two additional flux reversals causing one minor hysteresis  1.5
Stator core
loop per cycle. From the rotor flux in Fig. 19, it can be seen
2
that the flux waveforms in the rotor are symmetric with two 0 120 240 360 480 600 720

minor hysteresis loops per cycle in the rotor core. Fig. 20 Rotor position [Degrees]
shows the simulated hysteresis loops by the Energetic model
Fig. 19. Flux density waveforms in the stator pole and the stator core.
in different points inside the machine core.

0093-9994 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIA.2014.2300160, IEEE Transactions on Industry Applications

2 Energetic model
Rotor pole Modified Steinmetz
1.5 Rotor core Energetic model + modified Steinmetz

1
Flux density [T]

0.5

 0.5

1

 1.5

2
0 120 240 360 480 600 720

Rotor position [Degrees] Fig. 21. Spatial distribution of different calculation techniques used by
the hybrid model.
Fig. 20. Flux density waveforms in the rotor pole and the rotor core.

TABLE III
(a) Stator pole (b) Stator core Comparison between the machine hysteresis losses calculated by three models
2 2

1.5
1.5
1 Stator loss (W) Rotor loss (W) Total loss (W)
Flux density (T)

Flux density (T)

1
0.5
Hybrid model 6.8412 1.7705 8.6118
0

0.5
Analytical model 6.2585 1.8298 8.0883
-0.5

-1
Energetic model 7.0937 1.9940 9.0878
0
-1.5

-0.5 -2
-1000 0 1000 2000 3000 4000 5000 -1000 -500 0 500 1000 1500 2000
Field intensity (A/m) Field intensity (A/m)
hysteresis losses calculated by the analytical model are about
(c) Rotor pole (d) Rotor core 6.5% lower than the hybrid model losses. The main reason of
1.5 1.5

1 1
this divergence is that Lavers formula underestimates the
unipolar flux losses, as can be seen from the formula errors in
0.5 0.5
Flux density (T)

Flux density (T)

Fig. 8. On the other hand, the Energetic model simulation of

0 0
-0.5
-1
-1.5
-200 -150 -100 -50 0 50 100
Field intensity (A/m)
Fig. 20. Simulated hysteresis loops by the energetic model in different
parts of the machine core.
The hybrid model is then applied to calculate the hysteresis
losses in each element of the machine. Fig. 21 shows the
regions at which the hybrid model switches from using the
analytical equations to the Energetic model or uses both of
them for hysteresis loss calculation. It can be seen from Fig.
21 that part of the rotor pole uses only the modified Steinmetz
equation as the flux waveform in this region is symmetric and
the minor hysteresis loop losses are negligible. On the other
hand, in the stator pole, the hybrid model uses only the
Energetic model for simulating both major and minor
hysteresis loops as the main loop contains significant DC
component. The rest of the machine core uses the Energetic
model for calculating the minor loop losses and uses the
modified Steinmetz equation for calculating the major loop
losses, as the major loop DC component is lower than 0.34 T
in these regions.
Table III compares the machine hysteresis losses calculated
by the three models described in section III. The machine
unipolar hysteresis loops is very accurate. While the formula
-0.5
can predict the minor loop losses in the stator core where the
-1 peak flux density is around 1.6 T, it overestimates the minor
150 200
-1.5
-150 -100 -50 0 50 100 150
loop losses in the rotor core, where the peak flux density is
Field intensity (A/m)
about 1.2T. The formula overestimation of smaller minor
loops at 1.2 T can also be observed from the formula errors,
shown in Fig. 6. This explains why the rotor hysteresis loss
calculated by the analytical model is higher than the loss
calculated by the hybrid model. It can also be observed from
table III that the machine losses calculated by using only the
Energetic model are higher than the hybrid model losses. This
divergence is attributed to the fact that the analytical
prediction of major loop losses is more accurate than the
Energetic model simulations.
V. CONCLUSION
A series of minor hysteresis loops of different magnitudes,
positions, and major loop peak flux densities are measured.
The measured hysteresis losses are then compared to the
losses calculated by an analytical model. The model is able to
accurately predict the symmetrical major hysteresis loop
losses. However, it can only predict minor loop losses under
particular conditions. Therefore, an Energetic model is
implemented in order to achieve accurate minor loop loss
prediction. While the Energetic model can achieve
considerable improvement of minor loop loss prediction over
the analytical model, the symmetric major loop losses

0093-9994 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIA.2014.2300160, IEEE Transactions on Industry Applications
calculated analytically are found to be more accurate.
Consequently, a hybrid model is developed to calculate the
hysteresis losses in electrical machines using both the
analytical equations and the Energetic model. The hybrid
model is then applied to calculate the hysteresis losses in a SR
machine. The results show that having a model that is capable
of calculating the hysteresis losses under a variety of minor
hysteresis loops is essential for precise machine core loss
prediction.
ACKNOWLEDGMENT
This project is part of the R&D program of the Natural
Sciences and Engineering Research Council Chair entitled
“Energy Efficiency in Electrical Machines for Small
Renewable Energy Production Systems” established in 2009 at
Concordia University.
REFERENCES
[1] T. Nakata, Y. Ishihara, and M. Nakano, “Iron losses of silicon steel core
produced by distorted flux,” Elect. Eng. Jpn., vol. 90, pp. 10–20, 1970.
[2] J. D. Lavers, P. P. Biringer, and H. Hollitscher, “A simple method of
estimating the minor loop hysteresis loss in thin laminations,” IEEE
Trans. Magn., vol. 14, pp. 386–388, Sept. 1978.
[3] P. Rupanagunta, J. S. Hsu, and W. F. Weldon, “Determination of iron
core losses under influence of third-harmonic flux component,” IEEE
Trans. Magn., vol. 27, pp. 768–777, Mar. 1991.
[4] C. Cho, D. Son, and Y. Cho, “Core loss analysis of nonoriented
electrical steel under magnetic induction including higher harmonics,” J.
Magn., vol. 6, p. 66, 2001.
[5] S.D. Calverley, G.W. Jewell, R.J. Saunders, “Prediction and
Measurement of Core Losses in a High-Speed Switched-Reluctance
Machine,” IEEE Trans. Magn., Vol. 41 , pp. 4288 – 4298, Nov. 2005.
[6] M.L. Hodgdon, ”Mathematical theory and calculations of magnetic
hysteresis curves”, IEEE Trans. Magn. 24(6) (1988).
[7] F. Preisach, “Uber die magnetische nachwrikung,” Zeitschrift fur
Physik, vol. B 94, pp. 277–302, 1935.
[8] I. D. Mayergoyz, Mathematical Models of Hysteresis. New York:
Springer-Verlag, 1991.
[9] D.C. Jiles, D.L. Atherton, “Theory of ferromagnetic hysteresis,” J.
Magn. Magn. Mater. 61 (1986) 48.
[10] H. Hauser, “Energetic model of ferromagnetic hysteresis: Isotropic
magnetization,” J. Appl. Phys. 96, 2753 (2004).
[11] E. Barbisio, F. Fiorillo, and C.Ragusa “Predicting loss in magnetic steels
under arbitrary induction waveform and with minor hysteresis loops,”
IEEE Trans. Magn., vol. 40, pp. 1810 - 1819, Jul. 2004.
[12] Y. Chen and P. Pillay, “An improved formula for lamination core loss
calculations in machines operating with high frequency and high flu
density excitation,” proc. of IEEE 37th IAS Annual Meeting, vol. 2, pp.
759– 766, 13-18 Oct 2002.
[13] M. Ibrahim and P.Pillay, “Advanced Testing and Modeling of Magnetic
Materials Including a New Method of Core Loss Separation for
Electrical Machines,” proc. of Energy Conversion Congress and
Exposition conference, pp. 706 -713, 17-22 Sep 2011.
[14] Y. Zhang, M. Cheng, P.Pillay and B. Helenbrook”High order finite
element model for core loss assessment in a hysteresis magnetic
lamination,”J. Appl. Phys. 106, 043911 (2009)
[15] Y. Hayashi and T. J. E. Miller, “A new approach to calculating core
losses in the SRM,” IEEE Trans on. Ind. Appl., vol. 31, no. 5, pp. 1039–
1046, Sep./Oct. 1995.
[16] P. Materu and R. Krishnan, “Estimation of switched reluctnace motor
losses,” in Proc. Conf. Rec. IEEE Ind. Appl. Soc. Annu. Meeting, vol. 1,
Oct. 2–7, 1988, pp. 79–90.
[17] J. Faiz and M. B. B. Sharafin, “Core losses estimation in a multiple teeth
per stator pole switched reluctance motor”, IEEE Trans. Magn., vol. 30,
no. 2, pp. 189–195, Mar. 1994.
[18] T. L. Mthombeni and P. Pillay, “Lamination core losses in motors with
nonsinusoidal excitation with particular reference to pwm and srm
0093-9994 (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
excitation waveforms”, IEEE Transactions on Energy Conversion, vol.
20, pp. 836 – 843, December 2005.
[19] S. D. Calverly, G. W. Jewell and J. R. Saunders “Prediction and
measurement of core losses in a high-speed switched-reluctance
machine” , IEEE Trans. Magn., Vol. 41 , pp. 4288 – 4298, Nov. 2005.
[20] Z. Gmyrek, A. Boglietti, A. Cavagnino, "Iron Loss Prediction With
PWM Supply Using Low- and High-Frequency Measurements: Analysis
and Results Comparison,", IEEE Transactions on Industrial Electronics,
vol.55, no.4, pp.1722-1728, Apr. 2008.
[21] Z. Gmyrek, A. Boglietti, A. Cavagnino, "Estimation of Iron Losses in
Induction Motors: Calculation Method, Results, and Analysis," IEEE
Transactions on Industrial Electronics, vol.57, no.1, pp.161-171, Jan.
2010.
Maged Ibrahim (S’10) received the B.S. degree from
Alexandria University, Alexandria, Egypt in 2008, and the
M.S. degree in 2011 from Concordia University, Montreal,
Canada, where he is currently pursuing the Ph.D. degree in
electrical engineering.
His research interests include core loss modeling in
magnetic materials, design and control of electrical
machines, and power electronics.
Pragasen Pillay (F’05) received the B.S. degree from the
University of Kwa-Zulu Natal, Durban, South Africa, in
1981, the M.S. degree from the University of Kwa-Zulu,
Natal, Durban, South Africa, in 1983, and the Ph.D.
degree from Virginia Polytechnic Institute and State
University, Blacksburg, in 1987.
Currently, he is a Professor in the Department of
Electrical and Computer Engineering, Concordia
University, Montreal, Canada, where he holds the NSERC/Hydro Quebec
Industrial Research Chair. From 1988 to 1990, he was with the University of
Newcastle-upon-Tyne, Newcastle-upon-Tyne, U.K. From 1990 to 1995, he
was with the University of New Orleans. From 1995 to 2007 he was with
Clarkson University, Potsdam, NY, where he held the Jean Newell
Distinguished Professorship in Engineering. He is also an Adjunct Professor
at the University of Cape Town, Cape Town, South Africa. His research and
teaching interests are in modeling, design, and control of electric motors and
drives for industrial and alternate energy applications.
Dr. Pillay is a member of the IEEE Power Engineering, IEEE Industry
Applications (IAS), IEEE Industrial Electronics, and IEEE Power Electronics
Societies. He is a member of the Electric Machines Committee and Past
Chairman of the IEEE Industrial Drives Committee of the IAS, Past Chairman
of the Induction Machinery Subcommittee of the IEEE Power Engineering
Society, Past Chairman of the Awards Committee of the IAS Industrial Power
Conversion Department. He has organized and taught short courses in electric
drives at IAS Annual Meetings. He is a Fellow of the Institution of Electrical
Engineers and Technologists, U.K., and a Chartered Electrical Engineer in the
U.K. He is also a Member of the Academy of Science of South Africa. He was
a recipient of a Fulbright Scholarship for his Ph.D and received the Order of
Mapungubwe from the President of South Africa in 2008 for contributions to
South Africa in the area of energy conservation.