You are on page 1of 5

IEEE TRANSACTIONS ON MAGNETICS, VOL. 54, NO.

3, MARCH 2018 1300105

Core Loss Calculation Based on Finite-Element Method


With Jiles–Atherton Dynamic Hysteresis Model
Yang Li 1, Lihua Zhu 1, and Jianguo Zhu2
1 Tianjin
Key Laboratory of Advanced Electrical Engineering and Energy Technology, Tianjin
Polytechnic University, Tianjin 300384, China
2 Faculty of Engineering and Information Technology, University of Technology Sydney, Ultimo, NSW 2007, Australia

For accurate computation of core losses, the Jiles–Atherton ( J– A) dynamic hysteresis model accounting for hysteresis, eddy
current and excess losses is incorporated into the finite-element method (FEM). The J– A dynamic hysteresis model is constructed
by combining the traditional J– A hysteresis model with the models of instantaneous eddy current and excess losses. The J– A model
parameters and dynamic loss coefficients are determined by fitting the models to the measurement data of a single sheet tester
(SST 500) and Epstein frame tester. To find the robust best fit, the particle swarm optimization algorithm is employed. By using the
proposed J– A dynamic hysteresis model and FEM, the magnetic characteristics of a magnetic core is simulated and the core loss
distribution within the core obtained. The calculated and measured results are compared to show the accuracy and effectiveness of
the proposed model.
Index Terms— Core losses, finite-element method (FEM), Jiles–Atherton ( J– A) dynamic hysteresis model.

I. I NTRODUCTION the equivalent elliptical loop method [8], which is inaccurate


to account for the magnetic hysteresis.
C ORE loss is a key and difficult component to calculate
accurately in the design optimization of power transform-
ers to achieve high efficiency and avoid local overheating [1].
This paper proposes a new method to calculate the core
loss and its distribution in a magnetic core by the FEM
In a magnetic core, the total core loss can be attributed incorporated with the J – A dynamic hysteresis model, which
to the magnetic hysteresis and eddy currents. Because of the integrates the hysteresis, eddy current and excess losses.
existence of magnetic domains, the eddy current loss has two To verify the calculation accuracy, the calculated core loss
components due to the global and local eddy currents, known is compared with the experimental results obtained by both
as the eddy current and excess losses, respectively. For core the single-sheet and Epstein frame testers.
losses under sinusoidal excitations at different frequencies,
II. J – A DYNAMIC H YSTERESIS M ODEL
the Bertotti loss separation formula has been used to separate
the total core loss into three components corresponding to the A. Traditional J – A Hysteresis Model
hysteresis, eddy current and excess losses [2]. The traditional J – A model assumes that in a ferromagnetic
When the excitation is non-sinusoidal, however, a general material, the magnetization M has two components: an irre-
dynamic model is required. In [3], the core loss is calculated versible magnetization Mirr due to the discontinuities in the
by integrating the area of dynamic hysteresis loop. In [4], material structure and a reversible magnetization Mrev due to
the hysteresis model and the finite-element method (FEM) are the elastic bending of domain wall [9], [10].
coupled. Based on this, the FEM is employed for 3-D eddy By the law of energy conservation, the energy balance in a
current analysis [5], and it was claimed that this method could magnetization process can be expressed as
  
improve greatly the accuracy of core loss calculation, but the d Mirr
μ0 Md He = μ0 Man d He − μ0 kδ d He (1)
influence of the excess loss was not considered, and some d He
discrepancy still exists between the theoretical and experi- where
mental results. A Jiles–Atherton ( J – A) dynamic model was     
H + αMan a
proposed for the accurate simulation of hysteresis loops [6]. Man = Ms coth − (2)
a H + αMan
Due to improper method of parameter determination, the
simulation is inaccurate. In [7], the dynamic hysteresis model where Man is the anhysteretic magnetization, μ0 the vacuum
is derived to calculate the core loss, but the loss distribution permeability, He = H + αM the effective field, k the
of the transformer cannot be calculated, and thus the local pinning parameter, δ(δ = 1 for d H /dt > 0, and δ = −1
overheating still remains a problem. A dynamic core loss for d H /dt < 0) the direction coefficient, Ms the saturation
model was proposed for loss distribution calculation based on magnetization, a the Langevin parameter, and α the local field
parameter.
Manuscript received June 19, 2017; revised September 19, 2017; accepted
October 16, 2017. Date of current version February 21, 2018. Corresponding B. J – A Dynamic Hysteresis Model
author: L. Zhu (e-mail: zhulihua@tjpu.edu.cn). In order to consider the effects of eddy current and excess
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. losses on the hysteresis loop, the traditional J – A hysteresis
Digital Object Identifier 10.1109/TMAG.2017.2765704 model is extended to the J – A dynamic hysteresis model.
0018-9464 © 2018 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.
1300105 IEEE TRANSACTIONS ON MAGNETICS, VOL. 54, NO. 3, MARCH 2018

The classical eddy current instantaneous power loss per unit


volume can be expressed as
 
e2 σ d B 2
pe = (3)
2β dt
where e is the thickness of material, σ the conductivity of
material, β the geometric coefficient, and β = 6 when the
core is laminated in the structure.
The excess loss is generated by the local eddy currents
ensuing from domain wall motion. According to Bertotti’s
statistical theory of losses [11], [12], the excess loss can be
calculated by

4σ GSV0 d B dB
Pex = 1+ −1 (4)
(n 0 V0 )2 dt dt
where n 0 is the effective number of active magnetic objects
under dc excitation, S the cross-sectional area of steel sheet,
G = 0.1356 the dimensionless coupling constant, and V0 the
statistical coupling field parameter determining the ability of
applied field to increase n 0 with increasing frequency.
Since

4σ GSV0 d B
1+  1. (5)
(n 0 V0 )2 dt
Equation (4) can be reduced as
3/2
dB
Pex = (G SV0 σ )1/2 . (6)
dt
To incorporate the effects of eddy current and excess losses,
(1) can be rewritten as
   Fig. 1. Flow chart of J – A dynamic hysteresis model combined with FEM.
d Mirr
μ0 Md He =μ0 Man d He − μ0 kδ d He
d He
 2  2  To incorporate the J – A dynamic hysteresis into FEM,
e σ dB d B 3/2
− dt − (G SV0 σ )1/2 dt. the equation of magnetic field considering hysteresis is
2β dt dt
discretize as
(7)
[K][An ] + [M(An−1 )] = [F] (10)
Substituting He = H + αM and B = μ0 (H + M) into (7),
one obtains the differential expression of the J – A dynamic where [M] is the magnetization matrix obtained by the J – A
model as follows: dynamic model, [K] is the stiffness matrix, [F] is the excitation
ckδ d Man
1 matrix, and subscripts n and n − 1 are the number of current
dM M − Man − 1−c d He + k e dt + k ex δ dt
dB dB 2
and last iterations.
= (8)
dB ckδ d Man μ0 kδ The flowchart of J – A dynamic hysteresis model combined
μ0 (α − 1) Man − M + 1−c d He − 1−c
with the FEM is shown in Fig. 1.
where ke = e2 σ/2β and kex = (G SV0 σ )1/2 are the dynamic
parameters determined by eddy current and excess losses, III. D ETERMINATION OF J – A DYNAMIC
respectively, and c is the domain rotation loss parameter. M ODEL PARAMETERS
A. Non-Oriented Silicon Steel Sheet Test
C. FEM Considering Magnetic Hysteresis Fig. 2(a) shows the single sheet tester (SST 500), and
Fig. 2(b) the Epstein frame tester used in this paper. The B–H
By the Maxwell’s equations, the magnetic field in terms of
relationship and the corresponding core losses of non-oriented
the vectorial magnetic potential A in domain  is governed by
silicon steel sheets, 50ww470, were measured and calculated.
 : ∇ × ∇ × A − μ0 ∇ × M = μ0 J The sample dimensions are 600 × 100 × 0.5 mm3 for the SST,
1 : A = 0 and 16 pieces of 300 × 30 × 0.5 mm3 for the Epstein frame.
2 : H × n = 0 (9)
B. Determination of Traditional J – A Hysteresis Parameters
where J is the excitation current density, 1 and 2 are the The J – A model can accurately simulate the largest major
boundaries of the first and second boundary conditions. loop known as the limiting loop. When smaller major loops are
LI et al.: CORE LOSS CALCULATION BASED ON FEM 1300105

TABLE I
PARAMETERS OF J – A DYNAMIC H YSTERESIS M ODEL

Fig. 2. (a) Single-sheet tester (SST 500). (b) Epstein frame tester.

simulated, however, the simulation and experimental results


can be quite different if the parameters are incorrectly deter-
mined. To overcome this deficiency of J – A model, a parame-
ters correction method is presented.
To find the robust best fit, the five parameters corresponding
to different flux densities are determined by the particle swarm
optimization (PSO) algorithm. It is found that parameters
a or k are related exponentially to the difference between peak
and saturation flux densities B and Bs or

am = as eγ (B−Bs ) (11)
km = ks eη(B−Bs ) (12)

where am and km are the modified parameters, as and ks the


Fig. 3. Dynamic hysteresis loops of different magnetic flux densities at 50 Hz.
constant parameters, and γ and η the correction factors.

C. Determination of Dynamic Loss Coefficients


The error function of the algorithm is defined as
A specific issue in core loss calculation is the consideration
of eddy current and excess losses. In a laminated magnetic 
 N
core, the dynamic eddy current and excess loss coefficients  (1 − Bcn /Bmn )
 n−1
vary with the flux density, and hence, a correction coefficient s Err = (17)
varying with the peak flux density, is introduced to account N −1
for this effect. The equations for determining these parameters where N is the number of hysteresis loop data points, Bm the
can be modified as measured data, and Bc the calculated data obtained by J – A
dynamic model with different variable vectors.
e2 σ s
ke = (13) The relationship between the coefficient s and peak flux
2β density was determined by the curve fitting method as follows:
kex = (GSV0 σ s)1/2 . (14)
s = −6.3 × [sin(B − π) + 0.33 × (B − 10)2 ] − 30 (18)
The coefficient s versus flux density can be found by the
PSO algorithm. where B is the peak flux density for a certain applied field.
The variable of PSO is the vector x[i ] which contains the When the peak flux density exceed 1.5 T, the value of s
parameter a, k, and s. In the i th iteration, a single PSO remains unchanged.
particle updates its velocity and position by the following
IV. C ALCULATION AND V ERIFICATION
equations [13]:
A. Single Sheet Test Simulation and Verification
v[i ] = wv[i ] + c1 rand(Pbest[i ] − x[i ]) To verify the accuracy of the proposed model, the calculated
c + c2 rand(Gbest − x[i ]) (15) results are compared with the measured data of SST. By fitting
x[i + 1] = x[i ] + v[i ] (16) the J – A dynamic hysteresis model to the experimental data,
seven parameters of the model were determined. In the process
where v[i ] is the velocity of the current particle at this of SST simulation, the parameters of J – A dynamic model
iteration, c1 and c2 are the cognitive and social acceleration are shown in Table I. Fig. 3 compares the measured and
factors, respectively, x[i ] is the current solution, w the inertia calculated major loops with the peak flux densities of 0.5, 1.0,
weight factor, rand is the random number in the range [0, 1], and 1.5 T, respectively. The core losses were calculated and
Pbest is the best value of current particle till now, and Gbest compared with the measured results. The detailed results are
is the best value of all particles. shown in Table II.
1300105 IEEE TRANSACTIONS ON MAGNETICS, VOL. 54, NO. 3, MARCH 2018

TABLE II TABLE III


C OMPARISON OF C ALCULATED R ESULTS W ITH PARAMETERS OF J – A DYNAMIC H YSTERESIS M ODEL
S INGLE S HEET T ESTS AT 50 Hz

TABLE IV
C OMPARISON OF C ALCULATED R ESULTS W ITH E PSTEIN
F RAME T ESTS AT 50 H Z

Fig. 4. Dynamic hysteresis loop at 50 Hz (Epstein).

As shown in Table II, the errors between the calcu-


lated and measured results are less than 5% in most cases.
However, in particular cases, such as the peak flux densities
of 0.1 and 1.8 T, the error increased rapidly, most probably
due to the inappropriate assumption of J – A model or the
measurement error when the flux density is too large or too
small. Fig. 5. Core loss distribution of Epstein frame.

B. Epstein Frame Test Simulation and Verification For research on local overheating of a magnetic core,
To further confirm the accuracy of the proposed model, the core loss distribution needs to be considered. Fig. 5 shows
the experimental results of Epstein frame tests are used the core loss distribution in the test sample of the Epstein
to compare with the theoretical results. Fig. 4 shows the frame when the flux density is 1.5 T. As shown, the core loss
calculated major loops corresponding to different peak flux distribution is reasonably uniform except at the four corners,
densities. In the Epstein frame simulation, the parameters of confirming the validity of the Epstein frame tests.
J – A dynamic model are listed in Table III. Table IV tabulates
the calculated and measured results in detail. As shown, C. Simulation and Verification of Minor Loops
the proposed model is sufficiently accurate for engineering To test the implementation of the proposed model under the
applications. influence of harmonics, the model of SST has been compared
LI et al.: CORE LOSS CALCULATION BASED ON FEM 1300105

the J – A dynamic hysteresis model can provide accurate


results with error less than 5%, which is sufficiently accurate
for most engineering applications. In the case of Epstein frame
test, the calculated core loss distribution is reasonably uniform
in the test samples.

ACKNOWLEDGMENT
This work was supported by the National Natural Science
Foundation of China under Grant 51507.
R EFERENCES
[1] L. Cao, J. He, and B. Zhang, “Dynamic hysteresis loss model of power
transformer under dc current biasing and its verification,” Proc. CSEE,
vol. 28, no. 24, pp. 141–146, 2008.
[2] G. Bertotti, “General properties of power losses in soft ferromag-
netic materials,” IEEE Trans. Magn., vol. 24, no. 1, pp. 621–630,
Jan. 1988.
[3] J. G. Zhu, “Numerical modelling of magnetic materials for computer
Fig. 6. Dynamic hysteresis loops of SST with excitation consisting of a aided design of electromagnetic devices,” Ph.D. dissertation, Univ.
fundamental and a third harmonic component. Technol. Sydney, Sydney, NSW, Australia, 1994.
[4] F. Liu, “The computation of magnetic field taking in account hys-
with the measured results with minor loops. Fig. 6 depicts the teresis and the measurement techniques of magnetization properties,”
Ph.D. dissertation, Hebei Univ. Technol., Tianjin, China, 2001.
calculated and measured results with the peak flux density [5] F. Liu, Q. Yang, and W. Yan, “Magnetic field numerical calculation
of 1.5 T when the excitation consists of a fundamental method considering hysteresis characteristic,” J. North China Electr.
component (50 Hz) superposed by a third harmonic component Power Univ., no. 32, pp. 59–61, 2005.
[6] S. E. Zirka et al., “On physical aspects of the Jiles–Atherton hysteresis
(150 Hz), generating a pair of minor loops. For peak flux models,” J. Appl. Phys., vol. 112, no. 4, p. 043916, 2012.
density of 1.5 T, the core loss is 3.28 W/kg in the normal [7] Y. Wang and Z. Liu, “Estimation model of core loss under DC
state without minor loops, but under the effect of the third bias,” IEEE Trans. Appl. Supercond., vol. 26, no. 7, Oct. 2016,
Art. no. 0608905.
harmonics, the measured core loss increases to 4.39 W/kg. [8] D. Lin, P. Zhou, W. N. Fu, Z. Badics, and Z. J. Cendes, “A dynamic
The calculated core loss is 4.18 W/kg with a loss breakdown core loss model for soft ferromagnetic and power ferrite materials in
as the calculated hysteresis loss of 1.92 W/kg, the eddy current transient finite element analysis,” IEEE Trans. Magn., vol. 40, no. 2,
pp. 1318–1321, Mar. 2004.
loss of 1.90 W/kg, and the excess loss of 0.37 W/kg. The error [9] D. C. Jiles, J. B. Thoelke, and M. K. Devine, “Numerical determination
of core loss calculation is 5.02%. of hysteresis parameters for the modeling of magnetic properties using
the theory of ferromagnetic hysteresis,” IEEE Trans. Magn., vol. 24,
no. 1, pp. 27–35, Jan. 1992.
V. C ONCLUSION [10] Z. Li et al., “Queries on the J-A modeling theory of the magnetization
The core loss is calculated by incorporating the process in ferromagnets and proposed correction method,” Proc. CSEE,
vol. 3, pp. 124–131, 2011.
J – A dynamic hysteresis model into the FEM. To eliminate [11] D. C. Jiles, “Modelling the effects of eddy current losses on frequency
the error caused by inaccurate parameter determination, dependent hysteresis in electrically conducting media,” IEEE Trans.
the J – A dynamic model parameters are fit to the measured Magn., vol. 30, no. 6, pp. 4326–4328, Nov. 1994.
[12] E. Barbisio, F. Fiorillo, and C. Ragusa, “Predicting loss in mag-
data by the PSO algorithm, and the methods for modifying netic steels under arbitrary induction waveform and with minor hys-
the parameters are presented. teresis loops,” IEEE Trans. Magn., vol. 40, no. 4, pp. 1810–1819,
The proposed J – A dynamic model was experimentally Jul. 2004.
[13] K. Salah, “A generic model order reduction technique based on particle
verified by comparing the major and minor loops and core swarm optimization (PSO) algorithm,” in Proc. IEEE EUROCON,
losses at different flux densities. The comparison showed that Jul. 2017, pp. 1–4.

You might also like