A Review of Magnetic Core Materials Core Loss Modeling and Measurements in High-Power High-Frequency Transformers
A Review of Magnetic Core Materials Core Loss Modeling and Measurements in High-Power High-Frequency Transformers
A Review of Magnetic Core Materials Core Loss Modeling and Measurements in High-Power High-Frequency Transformers
classical Steinmetz equation (SE), and advanced core loss mea- frequency frequency frequency M
AC AC
surement techniques. Pros and cons for the magnetic materials,
modeling and measurement methods are comprehensively ana-
(b)
lyzed and summarized to provide a sufficient application insight.
Index Terms—Core loss measurement, core loss modeling, high-fre- Fig. 1. Schematic of railway traction system. (a) Traction system based on
quency transformers, soft magnetic materials. LFTs. (b) Traction system based on HFTs [1].
10000
50 Hz P = 1 MW
I. Introduction
S
1000
OLID-STATE transformers (SSTs, also known as power Arkansas
NC state
Volume (li ter )
TABLE I
The Magnetic Parameters of Commonly Used SMMs [6]
the advantages of reactive power compensation, power flow the heat leakage through the chamber wall will reduce the
control, voltage sag compensation, fault current limitation, etc. measurement accuracy of the calorimeter methods, while the
In SST, the dc-dc converter that plays the role of voltage high-frequency excitation or high-quality factor core will make
change and isolation is the core topological, and they are usually the two-winding method very sensitive to phase discrepancy.
composed of dual active bridges (DAB) or series resonant In recent years, researchers have made many improvements to
converters (SRC). The efficiency and power density of a dc- the above-mentioned shortcomings. Section IV reviews four
dc converter greatly depends on the HFTs. Compared with types of core loss measurement techniques. Pros and cons of
LFTs, the design of HFTs requires additional consideration of various techniques are discussed and compared.
material selection, eddy current loss, leakage inductance, stray
capacitance and thermal management, etc. Among them, due II. Magnetic Core Materials in High-Power
to the nonlinearity and the complex physical mechanism, the High-Frequency Transformers
problems related to the magnetic core have been regarded as the
focus and difficulty of HFTs. The performance of magnetic cores depends largely on the
In LFTs, Fe-Si alloys are commonly used core materials. properties of materials. Basically, the desired characteristics for
However, as the frequency increases, the eddy current loss in an HFT core material includes [3]-[9]:
the core becomes prominent, which limits further improvement 1) High saturation flux density Bs;
of power density. Compared with Fe-Si alloys, soft ferrites, 2) High Curie temperature Tc;
amorphous and nanocrystalline alloys have the advantages 3) Low magnetostriction λ;
of low core loss, and are generally used as core materials for 4) High permeability μ;
medium- and high-frequency applications. In fact, an excellent 5) Low coercivity Hc;
core material should have the advantages of low core loss, 6) High resistivity ρ;
7) Low loss Ploss;
high saturation flux density, and high continuous operating
8) Low cost.
temperature, etc. Therefore, in order to make an appropriate
The magnetic core in HFT is made of soft magnetic
choice, it is necessary to understand the properties of various
materials (SMMs). SMMs can be further divided into four
materials. In this paper, Section II introduces in detail the
categories: traditional ferromagnetic alloys, soft ferrites,
magnetic parameters, properties, and applications of Fe-Si
amorphous/nanocrystalline alloys and soft magnetic composite
alloys, soft ferrites, amorphous/nanocrystalline alloys for
materials (SMCs). Generally, SMCs are seldom applied in
readers to refer to.
HFTs due to higher core loss and lower relative permeability
Generally, engineers prefer to use the classic SE to calculate [3]. In the following, properties, comparison and applications
the core loss. Unfortunately, the SE is only valid under of the first three materials – traditional ferromagnetic alloys,
sinusoidal excitation, and the voltage waveform on the HFTs soft ferrites, amorphous/nanocrystalline alloys – are discussed.
is rectangular or composed of rectangles. In this case, if SE is
still used to calculate the core loss, it will cause a large error.
In order to obtain an accurate calculation result, researchers A. Traditional Ferromagnetic Alloys
have proposed many improved methods based on SE, such Traditional ferromagnetic alloys are earliest SMMs and
as MSE, GSE, iGSE, WcSE, ISE, etc. In this article, Section still occupy a large market share in SMMs [4]. The traditional
III points out the defects of using SE to calculate core loss ferromagnetic alloys mainly include iron-silicon alloys, iron-
under typical nonsinusoidal waveform, reviews various loss nickel alloys and iron-cobalt alloys, etc. By adding silicon
modeling methods based on SE, compares the accuracy of into iron, iron-silicon alloys show higher resistivity and lower
these methods, and finally systematically classifies them. magnetostriction, at the expense of lower saturation flux density
Advanced measurement technology is the key to verifying and Curie temperature and increased brittleness. Generally, the
the core loss model. The calorimeter methods and the two- content of Si element in Fe-Si alloys does not exceed 6.5% [5].
winding method are the most commonly used. However, Table I shows the magnetic parameters of various SMMs [6].
Z. LI et al.: A REVIEW OF MAGNETIC CORE MATERIALS, CORE LOSS MODELING AND MEASUREMENTS 361
It can be seen that Fe-Si alloys have a higher saturation flux 1000
density and Curie temperature than other SMMs. However, Fe-Co-V
the resistivity is very small, so the application frequency of Fe- Grain orientation
Si alloys usually does not exceed 1 kHz. Iron-nickel alloys are 800 Fe-Si
characterized by low coercivity, high saturation flux density, high Fe-Si 6.5%
Fe
Soft ferrites are the most commonly applied magnetic material 200 Ni-Zn soft ferrites
in power electronics. Soft ferrites are made up of iron oxide
(common is Fe2O3), mixed with other metals such as Mn, Zn, Ni, Mn-Zn soft ferrites
Co or Cu, etc. The two most commonly used soft ferrites are Mn-
Material
Zn and Ni-Zn. The attractive feature of soft ferrites is the ultra-
high resistivity (see Table I), which significantly reduces eddy
Fig. 3. Compare the Curie temperature of various SMMs [6].
current loss in high-frequency. For example, under the sinusoidal
excitation of Bm = 0.1 T, f = 100 kHz, the loss of EPCOS’s soft
Ferromagnetic alloys f = 1 kHz
ferrite N87 at T = 100 °C is 50 kW/m³, while the loss of ultra-thin Nanocrystalline
5
silicon steel 10JNHF600 from JFE Steel Corporation of Japan 5×10 Amorphous
exceeds 1500 kW/m³. Usually, the typical resistivity range of Soft ferrites
Mn-Zn soft ferrites is 0.1 to 20 Ωm, which makes it applicable 2×105 FINEMET
to hundreds of kHz to several MHz range. The typical resistivity
NANOPERM
range of Ni-Zn soft ferrites is 103 to 107 Ωm, which makes it a 1×105 Co-based
good candidate for applications above 1 MHz [8].
Relat ive per meabi li ty μr
However, soft ferrites have some disadvantages. Firstly, the 5×104 Fe-Ni
Alloys
saturation flux density is the lowest in SMMs (usually 0.2-0.5 T).
(Fe,Ni)-
Secondly, soft ferrites have low Curie temperature, which is 2×104 based
HITPERM
Ferromagnetic alloys Based on the above parameters, we can conclude that Fe-
Nanocrystalline
2.5 Fe-50%Co Si alloys have large eddy current losses due to low resistivity,
Amorphous
Soft ferrites
therefore, usually used in low- to medium-frequency applications.
Saturat ion magn etic flu x density Bs (T)
Fe-3%Si Carbon Steel Soft ferrites have low eddy current loss in high-frequency, but
2.0 Fe its saturation flux density is low, which makes it have a larger
HITPERM
Fe-base size under the same transmission power. Besides, the Curie
Fe- 6.5%Si
NANOPERM temperature limit must be considered. Amorphous alloys have
1.5
FINEMET higher unit volume loss, but they have a high Curie temperature,
magnetic permeability, and saturation flux density, so they may
Fe-(40% 50%)Ni
1.0 FeSiAl well lead to high-power density in high-power applications.
(Fe,Ni)-base Nanocrystalline alloys have very low loss, high saturation flux
Fe-(70% 80%)Ni density and permeability, which ensures high efficiency and high-
0.5 Co-base Ni-Zn power density. However, if the design of the HFTs is based on
Mn-Zn
soft ferrites cost, amorphous or soft ferrites may be the best choice because
soft ferrites the cost of nanocrystalline cores is relatively high.
0.0
10 100 101 102 103
Coercivity Hc (A·m ) 1 III. Modeling of Magnetic Core Loss
The losses in HFTs mainly include winding loss and core
Fig. 5. Saturation magnetic flux density versus coercivity for typical SMMs
loss. For nonsinusoidal excitation, the winding loss can be
[5]-[7], [17], [18].
estimated by the improved Dowell formula [23], [24] and the
Fourier transformation. However, the core loss calculation is
105 f = 100 kHz, T = 100 °C
more difficult due to the nonlinearity and the complex physical
Nanocrystalline (Vitroperm 500F) mechanism in the magnetic materials. Therefore, the optimum
Fe-amorphous (Metglas 2605SA1)
104
Silicon Steel (JFE 10JNHF600)
design of HFTs requires an accurate investigation of core loss,
Core los s PV (kW·m3)
U B U B 220
SE
200
MSE
iGSE
180
WcSE
Measured
PV (kW·m3)
t t 160
DT/2 DT 140
T/2 T
120
(a) (b)
100
Fig. 7. Voltage/magnetic flux density waveforms of two typical nonsinusoidal
excitations in HFTs. D denote duty cycle ranging from 0 to 1. T denote the 80
switching period. (a) The symmetrical three-level voltage waveform. (b) The
two-level voltage waveform. 60
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Duty cycle D
is a function of the length (time) and width (amplitude) of
the pulse. Then, the losses generated by different excitation Fig. 8. Verify the accuracy of SE, MSE, iGSE and WcSE methods under two-
waveforms are stored in a Herbert curve for engineers to level voltage waveform [46].
refer to.
4) SE and improved methods based on SE [32]-[45]: For core loss is directly related to the rate of change of flux density
sinusoidal excitation, the core loss may be treated with the dB/dt. The author uses the relationship between sinusoidal
Steinmetz equation [32] excitation and nonsinusoidal excitation to find the equivalent
frequency feq under nonsinusoidal excitation and replaces the
PV = Kf α Bmβ (3) frequency f in SE
where the Steinmetz parameters K, α and β are constants PV = ( K ⋅ feqα −1 ⋅ Bmβ ) ⋅ f (4)
obtained from manufacturers or curve fitting. f indicates
the excitation frequency and Bm represents the peak flux where the equivalent frequency feq is expressed in dB/dt:
density. PV is the time-average power loss per unit volume.
For nonsinusoidal excitation, Fig. 7 shows two typical 2 T dB
2
where K, α and β are Steinmetz parameters and B(t) is the NSE are exactly the same.
instantaneous flux density. In order to improve the inaccuracy of Steinmetz parameters
However, the accuracy of GSE is not higher than that of in a wide span of frequencies, [38] proposed DNSE on the
MSE [42], [48], and GSE has not corrected defects 2). In order basis of NSE. The author separates the hysteresis loss Ph from
to expand the application frequency range, expanded GSE the total loss, and other losses are calculated by NSE (iGSE)
(EGSE) was proposed in [35]. However, EGSE method has not
been verified in high-frequency. In addition, a new parameter is
added to the formula, which is not provided in the data sheet. (12)
T ∫0
PV = ki ∆B dt (8) loss PV. This method believes that the magnetic field H in the
dt static hysteresis loop can be decomposed into two components:
the reversible component Hrev and the irreversible component
In order to be consistent with SE, the parameter ki should be: Hirr. In fact, Hrev can be regarded as reactive power, and Hirr can
be regarded as the actual static hysteresis loss Ph
K
ki = 1 dB 1 dB
α (9) T T
∫ (H + H irr ) ∫
2ππ
( 2ππ ) ∫0
α −1
cosθ 2 β −α dθ Ph = rev dt = Hirr dt (13)
T 0 dt T 0 dt
where K, α and β are Steinmetz parameters. Then, the author defined the equivalent elliptical loop
Both GSE and iGSE have made corrections to the base number (EEL), and the static hysteresis loss Ph is calculated by tracing
and are consistent with SE in form so that it can express the loss an elliptical loop having the same area as that of the original
characteristics under nonsinusoidal waveforms. The iGSE is hysteresis loop. After that, the author extended the method to
considered an excellent calculation method [46]-[48]. It can be calculate the total core loss PV
seen from Fig. 8 that compared to SE, MSE and WcSE, iGSE can
better follow the changing trend of core loss with duty cycle D. dB
α
However, when the duty cycle D is very low or very high, there is PV ( t ) = C ⋅ (14)
dt
still an error between the measured value and the calculated value
[43]. One of the reasons may defects 2).
where C is
D. Natural Steinmetz Equation (NSE) and Double Natural
1 β −α
Steinmetz Equation (DNSE) C=± K ⋅ Bm cos (θ ) (15)
Cαβ
After iGSE, NSE was proposed in 2004 [37]. NSE is
2 π2
Cαβ = ( 2ππ ) ⋅
α
ππ ∫0
β −α α
cos β θ dθ (16)
∆B kN T dB
PV =
2
T ∫
0 dt
dt (10)
where K, α and β are Steinmetz parameters.
EEL does not add new parameters. However, when the duty
where kN is
cycle D is very low or very high, it will face the defects 2) [47].
K
kN = F. Waveform-Coefficient Steinmetz Equation (WcSE)
2π α (11)
( 2ππ ) ∫0
α −1
cos θ dθ
The WcSE [40] is simpler in thinking than the previous
method. This method first finds the ratio η of the average of
In fact, through simplification, we can find that iGSE and the flux density waveform under nonsinusoidal excitation to
Z. LI et al.: A REVIEW OF MAGNETIC CORE MATERIALS, CORE LOSS MODELING AND MEASUREMENTS 365
the average of the flux density waveform under sinusoidal Steinmetz parameter, which are obtained by curve fitting. τ and
excitation, where the two excitations maintain the same qr are parameters related to materials. dB(t-)/dt and dB(t+)/dt
peak magnetic flux density Bm. Then, multiply the core loss are the flux density before and after the switching, respectively.
under sinusoidal excitation by the ratio η to find the core t1 is the duration of the zero-voltage phase.
loss under nonsinusoidal excitation, in which the core loss From (19), we can observe that compared to iGSE, i²GSE
under sinusoidal excitation can be obtained by SE. Since η is adds a loss caused by the magnetic relaxation effect. The
related to the average value of the flux density waveform, it is i2GSE is mainly used for excitation with zero-voltage phase
recorded as FWC waveform [Fig. 7(a)] and asymmetric triangular magnetic flux
waveform [Fig. 7(b), D ≠ 0.5]. It should be noted that although
PV = FWC ⋅ Kf α Bmβ the accuracy of i2GSE is higher than that of iGSE, five new
(17)
parameters are added, and these parameters usually do not
where K, α and β are Steinmetz parameters. appear in the data sheet.
Take the nonsinusoidal excitation in Fig. 7(b) as an example,
where the duty cycle D = 0.5. The coefficient FWC is π/4 at the H. Rectangular Extension of Steinmetz Equation (RESE)
moment. Therefore RESE [42] is a core loss calculation method only for
rectangular voltage excitation. The idea of this method is
PV = π ⋅ Kf α Bmβ (18)
similar to WcSE. Both methods use the core loss PV_sin under
4 sinusoidal excitation to multiply a coefficient η to obtain the
core loss PV_rec under rectangular excitation. The difference
between WcSE and RESE is that the coefficient η of the former
It can be seen from Fig. 8 than under square wave excitation
is the ratio of the average value of the flux density waveform,
[Fig. 7(b), D = 0.5], WcSE has the highest calculation accuracy,
while the coefficient η of the latter is directly related to the core
but only in this case. When the duty cycle D in Fig. 7(b) is far
loss. As shown in (23), the detailed derivation process can be
away from 0.5, the calculation accuracy of WcSE will become
referred to [42].
worse and worse [46], [47]. In addition, WcSE lacks theoretical
support in principle. 8
PV = γ +1
Kf α B β (23)
G. Improved-Improved Generalized Steinmetz Equation (i2GSE) π 4 D (1 − D )
2
0
cos θ 2
β −α
dθ
two rectangular pulses. Take Fig. 7(b) as an example, under
this excitation waveform, the instantaneous flux density
n
B(t) is shown in Fig. 9(a). In a period T, the total area can be
where K, α and β are Steinmetz parameters. ∑Q
l =1
P is the
el rl
subdivided into two areas: the first A1 during DT and the second
new loss term due to the magnetic relaxation effect. Prl and Qrl A2 during (1 - D)T. The area A1 can also be regarded as the half-
are further defined as period area of the new signal B1(t) [see Fig. 9(b)]. Similarly,
the area A2 can also be regarded as the half-period area of the
1 d
αr
−1
t new signal B2(t) [see Fig. 9(c)]. The new signals B1(t) and B2(t)
B ( t− ) ( ∆B )
βr
Prl = kr 1 − e τ (21) have frequencies f1 = 1/T1 = f / 2D and f2 = 1/T2 = f / 2(1 - D),
T dt respectively, and have the same duty cycle of 0.5.
− qr
dB ( t+ ) / dt According to CWH, the core loss PV can be expressed as
Qrl = e
dB ( t− ) / dt (22)
240 iGSE
PV (kW·m³) Material: Soft ferrites 3F3
Area A 1 Area A 2 100 kHz 100 °C 0.1 T ISE
200 Measurement data
B(t)
160
t
DT 120
T
(a) 80
Term A 40
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Duty cycle D
B 1(t)
t
Fig. 10. Verify the accuracy of iGSE and ISE methods under two-level voltage
2TD waveform [43].
(b)
Sinusoidal excitation SE
Term B Non-sinusoidal excitation 1892 [32]
2T(1D)
B 2(t)
CWH
(c) iGSE(NSE) EGSE 2010 [30]
2002 [36][37] 2009 [35]
ISE
Fig. 9. Flux waveforms and their corresponding periods considered for core 2017 [43]
losses calculation. (a) Instantaneous magnetic flux density B(t) of Fig. 7 (b). (b) DNSE i²GSE FGSE
2005 [38] 2012 [41] 2018 [45]
New signal B1(t). (c) New signal B2(t) [43].
(26)
(25)
where K1, α1, β1 and K2, α2, β2 are two sets of Steinmetz
parameters, corresponding to the frequency f1 = f / 2D and f2 = f where ke , ki , α , β, kf , αf , and βf are constants. The last term on the
/ 2(1 - D). right side of the equation represents the core loss due to relaxation
In addition, [43] provides a graph of Steinmetz parameters effects. dBf /dt represents the filtering time derivative of the flux
changes with frequency f. If the frequency f and the duty density waveform, which corresponds to the dB/dt after the first-
cycle D are known, we can find the core loss PV by putting the order transfer function filter with a time constant of τf .
corresponding Steinmetz parameters into (25). Due to the introduction of filtered time derivative dBf /dt of
The advantage of ISE is that it considers the influence of the flux density waveform, it is called FGSE. In addition, like
duty cycle D and frequency f on Steinmetz parameters at the i²GSE, FGSE has also added many new parameters.
same time, and solves the defects 2) by cleverly combining Fig. 11 summarizes the above improvement methods
CWH and WcSE. Fig. 10 compares the calculation accuracy and related references. This section classifies them from the
of ISE and iGSE, and it can be clearly seen that when the duty different perspectives.
cycle D is very low or very high, the accuracy of ISE is higher 1) According to the calculation domain, the GSE, iGSE(NSE),
than that of iGSE. DNSE, EEL, i2GSE, FGSE methods calculate the core
loss in the time domain, while the MSE, WcSE, RESE,
J. Filtered-Improved Generalized Steinmetz Equation (FGSE) ISE methods calculate the core loss in the frequency
domain;
[45] proposed a continuous waveform calculation method 2) In terms of whether new parameters are added, the MSE,
that includes the core loss caused by the magnetic relaxation GSE, iGSE(NSE), EEL, WcSE, ISE methods do not add
effect. This method separates the eddy current loss from the new parameters, while new parameters are introduced in
total loss, where the eddy current loss is expressed in the DNSE, i2GSE, RESE, FGSE methods;
Z. LI et al.: A REVIEW OF MAGNETIC CORE MATERIALS, CORE LOSS MODELING AND MEASUREMENTS 367
TABLE II Secondary
Compare Some of the Above Improvement Methods T out T in coolant
C
CUT CUT N1 : N 2
Power + R1 L1 +
Amp. Rref
1:1 + R1 L1 * * L2 R2 + +
* * R core
Digital v1 Rcore Lm v2 v2
Oscilloscope
Lm
vR v2 v1
vR + iR v3
Rref
(a) (b) Cr
vR
+ iR
Fig. 14. Two-winding measurement method and equivalent model. (a) Two-
winding method. (b) Equivalent model [66].
Fig. 15. The improved of two-winding method with series resonant capacitance
[66].
The balance calorimeter based on the open-type calorimeter
is the first improved method, as shown in Fig. 13(a). The
method is mainly divided into two steps [52]: the main test where v2 is the secondary voltage, vR is the current sensing
and the calibration test. The main test is the same as the open- voltage, Rref is the current sensing resistance, and T is the period
type calorimeter, and the calibration test uses a heater (such as of the excitation frequency.
resistance) to replace the CUT. It needs to inject a known and The advantage of this method is the elimination of winding
accurate power into the heater and measure the temperature rise. loss. However, when the magnetic core has high-quality factor
In theory, the precise power injected in the heater is approximately Q or high operating frequency, this method will very sensitive
equal to the power loss of the CUT if the temperature rise of to phase discrepancy [63]-[66]. In order to solve this problem,
the two tests is equal. However, a longer measurement time [63] proposes the method of parallel resonant capacitor Cr
is required, and extra equipment is required to ensure that the on the side of the excitation winding, so that occur parallel
physical conditions of the two tests are equal (for example resonate between Cr and the excitation inductance Lm to reduce
temperature, humidity, etc.) [55]. The series balance calorimeter the impedance angle φ. However, the assumptions Z1 << XC
was proposed in [53], which is to allow the main test and the in [63] may not be satisfied, and the measured losses include
calibration test to be carried out at the same time. This method winding losses. [64]-[66] proposed a capacitor cancellation
can effectively reduce the measurement time. [54] proposed method, in which a series capacitor is used instead of a parallel
the parallel balance calorimeter, which not only reduces the capacitor, and the port voltage v3 is measured instead of the port
measurement time, but also can control the physical conditions of voltage v2. The schematic is shown in Fig. 15. When the series
the two chambers to remain the same by introducing a blower. resonant capacitor Cr and the magnetizing inductance Lm are in
The double-jacket calorimeter based on the closed-type series resonance, the measured values v3 and iR are the voltage
calorimeter is the second improved method, which was first and current of Rcore respectively, and φ3 (phase difference
proposed in [60], as shown in Fig. 13(b). This method adds a between v3 and iR) is zero degrees at this time. The core loss
second insulated enclosure to prevent heat leakage. In theory, can be calculated by (28). In order to extend the excitation to
if the internal temperature of the calorimeter is the same as the arbitrary waveforms, an inductive cancellation method was
temperature between the two insulation layers, the heat leakage proposed in [67], which uses an air-core transformer instead of
can be eliminated. In [61], the author analyzed various heat loss a capacitor to cancel the reactive power under the fundamental
sources in detail, and took measures to reduce errors in every wave and various harmonics.
link as much as possible. A real-time calibration system was The capacitor cancellation method can effectively reduce
introduced in [62] by combining the double-jacket calorimeter the error caused by the phase discrepancy. However, this
and the balance calorimeter. This improvement eliminates the method cannot exclude the influence of the equivalent series
need for coolant flow measurement, which improves sensitivity resistance (ESR) of the resonant capacitor. In addition, [68]
and reduces complexity and system cost. pointed out that it is necessary to accurately resonate between
Cr and Lm, otherwise the error caused by the phase discrepancy
B. Two-Winding Method cannot be eliminated. In order to avoid the problem, a partial
cancellation method was proposed in [68], which introduced
The schematic of the two-winding method is shown in Fig.
the cancellation factor k, and the basic principle is to make
14. The CUT carries two windings: the excitation winding and
φ3 close to zero degrees by introducing different k under
the sensing winding. The current of excitation winding can be
measured through current sensing resistance and the voltage of different Cr conditions. By this improvement, the effect of
sensing winding is measured under the open-circuit condition. phase discrepancy is effectively reduced even if mismatch
Integrating the product of the two measured waveforms in one between the resonant capacitor Cr and excitation inductance
cycle can obtain the core loss Lm. In addition, [69] compares the measurement accuracy and
applicability of each method under arbitrary excitation, and
analyzes the error caused by the limitation of the oscilloscope
1
Pcore =
TRref ∫
T
v2 vR dt (28) sampling period. [70] proposed an initial compensation method
based on the partial cancellation method, which makes two-
Z. LI et al.: A REVIEW OF MAGNETIC CORE MATERIALS, CORE LOSS MODELING AND MEASUREMENTS 369
ii the core loss resistor Rcore is measured, the core loss can be
Ii_dc calculated by
Magnetizing current
iL (30)
Q1 Q3 CUT
Vin
C
where Irms is the effective value of the current through the
Q2 Q4
Loss current inductor, and Vrms is the effective value of the voltage through
the inductor.
Several impedance-based methods are proposed in literature
[65], [74], [75], such as bridge method, resonant method, I-V
Fig. 16. The basic schematic of direct current power meter law. method and the auto-balancing method. The Agilent impedance
measurement handbook [75] detailed the advantages, disa-
L Rcore RCu
dvantages and frequency application range of various im-
pedance-based methods. However, similar to two-winding
RS + +
L IL CUT
RC method, almost all of these methods are sensitive to phase
Rcore L
Power
Vin Resonant Vout discrepancy. In order to solve the problem, [76] proposed the
capacitor C
Rcore source - - resonance capacitance method. Fig. 17(c) shows the principle of
the resonant capacitor method. The CUT is modelled by a series
(a) (b) (c) of L, Rcore, and RCu where L is its inductance, Rcore represents its
core loss at a single frequency, and RCu represents its copper
Fig. 17. (a) Series connection. (b) Parallel connection. (c) The basic principle of loss. The resonant capacitor is modelled as a series of RC and C
the resonant capacitor method.
resonated with inductor at the desired frequency, where C is its
capacitance and RC represents its equivalent series resistance.
winding method suitable for higher frequencies. When L and C are resonant, the ratio of the output voltage
amplitude Vout-pk to the input voltage amplitude Vin-pk is
C. Direct Current Power Meter Law
Direct current power meter law is another electrical method 1
Vout pk RC + jω C
[71]-[73], and its basic principle is shown in Fig. 16. The main = r
(31)
V in pk R core + R Cu + RC
idea of this method is to convert the input dc voltage into a
PWM voltage excited on the magnetic core through a dc/ac
circuit. The core loss is equal to the total input power minus where ωr represents the resonant angular frequency of the
other losses in the converter except for the core loss inductor L and capacitor C.
Since the value of RC is very small compared to 1/ωrC,
Pcore = Pin − Pother (29) therefore, RC can be omitted, and we can derive
where Pcore is core loss, Pin is the total input power, and Pother is
other losses in the converter except the core loss. Other losses 1
V out pk RC + jω C ω rL
include MOSFET switching losses, MOSFET conduction = r
≈ (32)
losses, capacitor ESR losses, MOSFET reversal diode losses, V in pk Rcore + R Cu + RC R core + RCu + RC
other resistor losses of PCB and winding loss, etc.
The direct calculation of Pother may produce large errors. In
[72], a method of calibration with air-core inductor is proposed Then, transform (32) to get
to calculate the Pother, which improves the calculation speed
and accuracy of direct current power meter law. However, 2πfrLV in pk
R core RC R Cu (33)
this improvement is only applicable small inductance value. V out pk
[73] proposed a differential method based on the direct current
power meter law to calculate the large inductance value, and where ωr = 2πfr , fr is the resonant frequency.
further improved the measurement accuracy.
Core loss can be calculated by the (30) after obtainment
Rcore. Since only the voltage amplitude is used in the process
D. Impedance-Based Methods of calculating the loss and the phase is not used, this method
The basic idea of impedance-based methods is to equate the excludes the sensitivity to the phase discrepancy. However, it is
core as a resistor connected in series or in parallel with an ideal indispensable to fine-tuning the resonant capacitance C so that
inductor. The schematic is shown in Fig. 17(a) and (b). Once happen resonance with inductor L.
370 CPSS TRANSACTIONS ON POWER ELECTRONICS AND APPLICATIONS, VOL. 7, NO. 4, DECEMBER 2022
TABLE III properties, and nanocrystalline alloys can achieve higher power
Summary of Characteristics of Various Measurement Methods
density levels.
Then, this paper points out the defects of using the SE to
Whether the
Is it sensitive calculate the core loss under nonsinusoidal excitation, and
Measurement Excitation winding Measuring
to phase summarizes in detail various improved methods based on
method type loss can be speed
discrepancy? SE. In addition, this paper lists several literatures verifying
separated?
improved methods under two-level excitation. From these
Calorimeter Arbitrary
methods excitation
No No slow documents, it can be concluded that with the change of the
duty cycle, compared to MSE, GSE and WcSE, the calculation
Two-winding Arbitrary results of iGSE(NSE) can follow the measured value well.
Yes YesNo fast
method excitation However, when the duty cycle is very low or very high, the
Direct current square
accuracy of iGSE will decrease. At this time, ISE may be a
power meter wave or No No fast good method. RESE has higher accuracy, but the application
law PWM range of this method is limited and new parameters have
Impedance-
Sinusoidal
been added. Although i2GSE and FGSE take into account
based No Most Yes fast the additional loss caused by the relaxation effect, five new
wave
method
parameters have been added, which increases the complexity
Note: YesNo means that it will be affected at the beginning, and the impact of the calculation. In addition, the accuracy of some methods
will be eliminated after improvement still lacks experimental verification, which is the direction of
our future efforts.
E. Comparison Finally, this paper divides the methods of core loss measuring
Table III summarizes the characteristics of the above four into four categories, and introduces the principles, advantages,
methods from four aspects. It can be seen: disadvantages and improvements method. Among them, the
1) Calorimeter and two-winding methods can measure calorimeter methods and the two-winding method are the most
arbitrary waves, while direct current power meter law can commonly used methods, but the calorimeter methods are
only measure PWM or square waves, and impedance- time-consuming and requires complicated equipment, which
based methods can only measure sine waves; cannot be satisfied in some laboratories. The two-winding
2) Two-winding method can eliminate winding loss, but method is the most popular in the laboratory, but when the
other methods cannot; frequency is high, the effect of phase discrepancy needs to be
3) Calorimeter and direct current power meter law are considered.
not affected by phase discrepancy, while two-winding
method and impedance-based methods are sensitive to References
phase discrepancy which needs additional measures to
[1] J. Feng, W. Q. Chu, Z. Zhang, and Z. Q. Zhu, “Power electronic
compensate. transformer-based railway traction systems: Challenges and
In short, the calorimeter method and the two-winding opportunities,” in IEEE Journal of Emerging and Selected Topics in
method are two commonly used methods. Calorimeter has high Power Electronics, vol. 5, no. 3, pp. 1237–1253, Sept. 2017.
measurement accuracy, but requires complex equipment and [2] N. Kimura, T. Morizane, I. Iyoda, K. Nakao, and T. Yokoyama,
additional devices to ensure stable physical conditions, which “Application of solid-state transformer for HVDC transmission
from offshore windfarm,” in Proceedings of 2018 7th International
is impractical in some laboratory, and this method is very time- Conference on Renewable Energy Research and Applications, Paris,
consuming. The two-winding method is the most commonly France, 2018, pp. 902–907.
used method in the laboratory. This method is simple to [3] M. A. Bahmani, “Design and optimization considerations of medium-
operate, time-saving, and can eliminate winding losses, but it is frequency power transformers in high-power DC-DC applications,”
necessary to pay attention to the effect of phase discrepancy in Ph.D. dissertation, Dept. Energy Environ., Chalmers Univ. Tech.,
Gothenburg, Sweden, 2016.
high-frequency. [4] A. Inoue and F. Kong, “Soft magnetic materials,” in Proceedings of
Reference Module in Materials Science and Materials Engineering,
Elsevier, 2020.
IV. Discussion and Conclusion [5] A. Krings, A. Boglietti, A. Cavagnino, and S. Sprague, “Soft magnetic
In this paper, the magnetic core of HFTs is reviewed from the material status and trends in electric machines,” in IEEE Transactions on
perspective of: materials, loss modeling and loss measurement Industrial Electronics, vol. 64, no. 3, pp. 2405–2414, Mar. 2017.
[6] F. Fiorillo, G. Bertotti, C. Appino, and M. Pasquale, Soft Magnetic
techniques. Materials. Hoboken, NJ, USA: John Wiley & Sons, 1999.
First, this paper introduces magnetic properties of Fe-Si [7] S. Tumanski, Handbook of Magnetic Measurements, Warsaw, POL:
alloys, soft ferrites, amorphous and nanocrystalline alloys. CRC Press, 2016.
These SMMs are compared in detail from the perspective [8] W. G. Hurley and W. H. Wölfle, Transformers and Inductors for Power
Electronics: Theory, Design and Applications. Chichester, UK: John
of magnetic parameters and core loss, and their advantages,
Wiley & Sons, Ltd., 2013.
disadvantages and operating frequency range are summarized. [9] M. K. Kazimierczuk, High-Frequency Magnetic Components. Dayton,
Compared with Fe-Si alloys and soft ferrites, amorphous and OH, USA: Wiley, 2014.
nanocrystalline alloys have better comprehensive magnetic [10] K. Weyand, Y. Zhang, and H. Bousack, “Compound material for
Z. LI et al.: A REVIEW OF MAGNETIC CORE MATERIALS, CORE LOSS MODELING AND MEASUREMENTS 371
electromagnetic shielding in electrical and cryoelectronical applications,” based on Preisach model,” in IEEE Transactions on Magnetics, vol. 38,
in Proceedings of Conference on Precision Electromagnetic Measurements, no. 2, pp. 885–888, Mar. 2002.
Boulder, CO, USA, 1994, pp. 140–141. [30] C. R. Sullivan, J. H. Harris, and E. Herbert, “Core loss predictions for
[11] D. Neumayr, D. Bortis, J. W. Kolar, S. Hoffmann, and E. Hoene, general PWM waveforms from a simplified set of measured data,”
“Origin and quantification of increased core loss in MnZn ferrite plates in Proceedings of 2010 Twenty-Fifth Annual IEEE Applied Power
of a multi-gap inductor,” in CPSS Transactions on Power Electronics Electronics Conference and Exposition, Palm Springs, CA, USA, 2010,
Applications, vol. 4, no. 1, pp. 72–93, Mar. 2019. pp. 1048–1055.
[12] D. Bortis, L. Fässler, A. Looser, and J. W. Kolar, “Analysis of rotary [31] H. Matsumori, T. Shimizu, K. Takano, and H. Ishii, “Evaluation of iron
transformer concepts for high-speed applications,” in Proceedings loss of AC filter inductor used in three-phase PWM inverters based on
of 2013 Twenty-Eighth Annual IEEE Applied Power Electronics an iron loss analyzer,” in IEEE Transactions on Power Electronics, vol.
Conference and Exposition, Long Beach, CA, USA, 2013, pp. 17–21. 31, no. 4, pp. 3080–3095, Apr. 2016.
[13] W. K. Jun, R. H. Willens, and P. Duwez, “Non-crystalline structure in [32] C. P. Steinmetz, “On the law of hysteresis,” in Transactions of the
solidified gold–silicon alloys,” in Nature, vol. 187, no. 4740, pp. 869– American Institute of Mining and Metallurgical Engineers, vol. 9, no. 1,
870, 1960. pp. 1–64, Jan. 1892.
[14] Y. Yoshizawa, S. Oguma, and K. Yamauchi, “New Fe-based soft [33] J. Reinert, A. Brockmeyer, and R. W. A. A. De Doncker, “Calculation
magnetic alloys composed of ultrafine grain structure,” in Journal of of losses in ferro- and ferrimagnetic materials based on the modified
Applied Physics, vol. 64:10, pp. 6044–6046, Nov. 1988. Steinmetz equation,” in IEEE Transactions on Industry Applications,
[15] Hitachi, “Nanocrystalline soft magnetic material,” [Online]. vol. 37, no. 4, pp. 1055–1061, Jul./Aug. 2001.
Available:https://www.hitachi-metals.co.jp/products/elec/tel/pdf/hl- [34] J. Li, T. Abdallah, and C. R. Sullivan, “Improved calculation of core loss
fm9-h.pdf. with nonsinusoidal waveforms,” in Proceedings of Conference Record
[16] T. Gheiratmand and H. R. M. Hosseini, “Finemet nanocrystalline soft of the 2001 IEEE Industry Applications Conference. 36th IAS Annual
magnetic alloy: Investigation of glass forming ability, crystallization Meeting, Chicago, IL, USA, 2001, pp. 2203–2210.
mechanism, production techniques, magnetic softness and the effect [35] K. Chen, “Iron-loss simulation of laminated steels based on expanded
of replacing the main constituents by other elements,” in Journal of generalized Steinmetz equation,” in Proceedings of 2009 Asia-Pacific
Magnetism and Magnetic Materials, vol. 408, pp. 177–192, Jun. 2016. Power and Energy Engineering Conference, Wuhan, China, 2009, pp.
[17] G. Herzer, “Modern soft magnets: Amorphous and nanocrystalline 1–3.
materials,” in Acta Materialia, vol. 61, no. 3, pp. 718–734, Feb. 2013. [36] K. Venkatachalam, C. R. Sullivan, T. Abdallah, and H. Tacca, “Accurate
[18] L. Varga, and H. Davies, “Challenges in optimizing the magnetic prediction of ferrite core loss with nonsinusoidal waveforms using only
properties of bulk soft magnetic materials,” in Journal of Magnetism and Steinmetz parameters,” in Proceedings of 2002 IEEE Workshop on
Magnetic Materials, vol. 320, no. 20, pp. 2411–2422, Oct. 2008. Computers in Power Electronics, Mayaguez, PR, USA, 2002, pp. 36–41.
[19] C. Jiang, X. Li, S. S. Ghosh, H. Zhao, Y. Shen, and T. Long, [37] A. Van den Bossche, V. C. Valchev, and G. B. Georgiev, “Measurement
“Nanocrystalline powder cores for high-power high-frequency power and loss model of ferrites with non-sinusoidal waveforms,” in
electronics applications,” in IEEE Transactions on Power Electronics, Proceedings of 2004 IEEE 35th Annual Power Electronics Specialists
vol. 35, no. 10, pp. 10821–10830, Oct. 2020. Conference, Aachen, Germany, 2004, pp. 4814–4818.
[20] M. S. Rylko, K. J. Hartnett, J. G. Hayes, and M. G. Egan, “Magnetic [38] A. P. Van den Bossche, D. M. Van de Sype, and V. C. Valchev, “Ferrite
material selection for high power high frequency inductors in dc- loss measurement and models in half bridge and full bridge waveforms,”
dc converters,” in Proceedings of 2009 Twenty-Fourth Annual IEEE in Proceedings of 2005 IEEE 36th Power Electronics Specialists
Applied Power Electronics Conference and Exposition, Washington, DC, Conference, Dresden, Germany, 2005, pp. 1535–1539.
USA, 2009, pp. 2043–2049. [39] D. Lin, P. Zhou, W. N. Fu, Z. Badics, and Z. J. Cendes, “A dynamic core
[21] S. Wei, “Design of high-density transformers for high-frequency high- loss model for soft ferromagnetic and power ferrite materials in transient
power converters,” Ph.D. dissertation, Dept. Elect. Eng., Virginia finite element analysis,” in IEEE Transactions on Magnetics, vol. 40, no.
Polytech. Inst. State Univ., Blacksburg, VA, USA, 2006. 2, pp. 1318–1321, Mar. 2004.
[22] M. A. Bahmani, “Core loss evaluation of high-frequency transformers [40] W. Shen, F. Wang, D. Boroyevich, and C. W. Tipton, “Loss characte-
in high-power DC-DC converters,” in Proceedings of 2018 Thirteenth rization and calculation of nanocrystalline cores for high-frequency
International Conference on Ecological Vehicles and Renewable magnetics applications,” in IEEE Transactions on Power Electronics,
Energies, Monte Carlo, Monaco, 2018, pp. 1–7. vol. 23, no. 1, pp. 475–484, Jan. 2008.
[23] P. L. Dowell, “Effects of eddy currents in transformer windings,” in [41] J. Muhlethaler, J. Biela, J. W. Kolar, and A. Ecklebe, “Improved core-
Proceedings of the Institution of Electrical Engineers, vol. 113, no. 8, pp. loss calculation for magnetic components employed in power electronic
1387–1394, Aug. 1966. systems,” in IEEE Transactions on Power Electronics, vol. 27, no. 2, pp.
[24] J. A. Ferreira, “Improved analytical modeling of conductive losses in 964–973, Feb. 2012.
magnetic components,” in IEEE Transactions on Power Electronics, vol. [42] M. Mu and F. C. Lee, “A new core loss model for rectangular AC
9, no. 1, pp. 127–131, Jan. 1994. voltages,” in Proceedings of 2014 IEEE Energy Conversion Congress
[25] W. A. Roshen, “A practical, accurate and very general core loss model and Exposition, Pittsburgh, PA, USA, 2014, pp. 5214–5220.
for nonsinusoidal waveforms,” in IEEE Transactions on Power [43] S. Barg, K. Ammous, H. Mejbri, and A. Ammous, “An improved
Electronics, vol. 22, no. 1, pp. 30–40, Jan. 2007. empirical formulation for magnetic core losses estimation under
[26] A. Boglietti, A. Cavagnino, M. Lazzari, and M. Pastorelli, “Predicting nonsinusoidal induction,” in IEEE Transactions on Power Electronics,
iron losses in soft magnetic materials with arbitrary voltage supply: An vol. 32, no. 3, pp. 2146–2154, Mar. 2017.
engineering approach,” in IEEE Transactions on Magnetics, vol. 39, no. [44] X. Yu, Y. Li, Q. Yang, S. Yue, and C. Zhang, “Loss characteristics and
2, pp. 981–989, Mar. 2003. model verification of soft magnetic composites under non-sinusoidal
[27] L. Chang, T. M. Jahns, and R. Blissenbach, “Characterization and excitation,” in IEEE Transactions on Magnetics, vol. 55, no. 2, pp. 1–4,
modeling of soft magnetic materials for improved estimation of PWM- Feb. 2019.
induced iron loss,” in IEEE Transactions on Industry Applications, vol. [45] V. C. do Nascimento and S. D. Sudhoff, “Continuous time formulation
56, no. 1, pp. 287–300, Jan. 2020. for magnetic relaxation using the steinmetz equation,” in IEEE
[28] A. Benabou, S. Clénet, and F. Piriou, “Comparison of Preisach and Jiles– Transactions on Energy Conversion, vol. 33, no. 3, pp. 1098–1107, Sept.
Atherton models to take into account hysteresis phenomenon for finite 2018.
element analysis,” in Journal of Magnetism and Magnetic Materials, [46] S. Yue, Y. Li, Q. Yang, X. Yu, and C. Zhang, “Comparative analysis of
vol. 261, no. 1-2, pp. 139–160, Apr. 2003. core loss calculation methods for magnetic materials under nonsinusoidal
[29] Y. Bernard, E. Mendes, and F. Bouillault, “Dynamic hysteresis modeling excitations,” in IEEE Transactions on Magnetics, vol. 54, no. 11, pp.
372 CPSS TRANSACTIONS ON POWER ELECTRONICS AND APPLICATIONS, VOL. 7, NO. 4, DECEMBER 2022
Weijian Han received the B.S., M.S. and Ph.D. nano-grids and are particularly related to digital control of power electronic
degrees in electrical engineering from the School of converters.
Automation, Northwestern Polytechnical University,
Xi’an, China, in 2012, 2015 and 2019, respectively.
From 2016 to 2018, he was a Visiting Ph.D. Student
with the Department of Information Engineering,
University of Padova, Padova, Italy.
Since 2020, he has been a Researcher with the
Department of Electrical Engineering, Hebei Un-
iversity of Technology, China. His research interests Jianliang Chen received the B.S. and Ph.D. degrees
are focused on modeling and digital control of power electronic converters. in electrical engineering from Beijing Institute of
Technology, Beijing, China, in 2013 and 2019,
respectively.
From 2017 to 2018, he was a Visiting Student
Zhen Xin received the B.S. and M.S. degrees from
with CURENT, University of Tennessee, Knoxville,
the College of Information and Control Engineering,
TN, USA. He is currently an Associate Professor
China University of Petroleum, Qingdao, China, in
with the Department of Electrical Engineering at
2011 and 2014, respectively, and the Ph.D. degree
Hebei University of Technology, Tianjin, China. His
from Aalborg University, Aalborg, Denmark, in
research interests include high power density soft
2017.
switching converters and the high-frequency application of wide band-gap
In 2016, he was a Visiting Scholar with the
power devices.
University of Padova, Padova, Italy. From 2017 to
2018, he was a Postdoctoral Research Fellow with
the Chinese University of Hong Kong, Hong Kong.
Since 2018, he has been with the Hebei University of Technology, Tianjin,
China, as a professor. His research interests include condition monitoring of
WBG-based power-electronic systems, modeling and validation of power
electronic component failure mechanisms, and modeling and control of power
converters for renewable energy systems.
Poh Chiang Loh received the B.Eng. (hons.) and
M.Eng. degrees from the National University of
Singapore, Singapore, in 1998 and 2000, respectively,
Qing Liu received the B.S. and M.S. degrees and the Ph.D. degree from Monash University,
in electrical engineering from Northwestern Melbourne, Vic, Australia, in 2002, all in electrical
Polytechnical University, China, in 2012 and 2015, engineering.
respectively, and the Ph.D. degree in information From 2013 to 2015, he was a Professor with
engineering from University of Padova, Italy, Aalborg University, Aalborg, Denmark. Since
in 2020. She is currently a Researcher with the 2015, he has been a Tenured Full Professor with
Department of Electrical Engineering at Hebei the Chinese University of Hong Kong, Hong
University of Technology, China. Her main research Kong, China. His research interests include power converters and their grid
interests are in the fields of low-voltage microgrids, applications.