See discussions, stats, and author profiles for this publication at: [Link]

net/publication/353395148

Design and Testing of a Medium Frequency Transformer

Conference Paper · June 2021

DOI: 10.1109/APEC42165.2021.9487310

CITATIONS READS

11 308

5 authors, including:

Xufu Ren Weng Haoyuan

University of Cambridge Zhejiang University
15 PUBLICATIONS 41 CITATIONS 14 PUBLICATIONS 52 CITATIONS

SEE PROFILE SEE PROFILE

Ruizhe Wang Dehong Xu

Zhejiang University Zhejiang University
4 PUBLICATIONS 22 CITATIONS 377 PUBLICATIONS 7,780 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Xufu Ren on 16 May 2022.

The user has requested enhancement of the downloaded file.

 Design and Testing of a Medium Frequency
Transformer
Xufu Ren, Jinghang Li, Haoyuan Weng, Ruizhe Wang and Dehong Xu
Institute of Power Electronics
Zhejiang University, Hangzhou 310027, China
nnxfren@[Link]

Abstract—This paper investigates analysis, design and works are distributed in different papers [1-12]. In this paper,
testing of a medium frequency transformer (MFT) for solid- we hope to investigate methodology of the analysis, design
state-transformer (SST) applications. Characterization of the and testing of the MFT, which may be helpful to give readers
transformer is studied and an optimization procedure is more complete view of the state-of-arts of MFT design.
introduced. A 20kW, 50kHz prototype with 7.3kW/L power
density is built, which has been tested to estimate its parameters, This paper is with the following organization. A loss
efficiency and insulation. Experiments show that the MFT analysis of both winding loss and core loss of the MFT is
prototype reaches 99.6% efficiency. introduced in Section II. In Section III, an optimization design
targeting to maximize the efficiency and power density of the
Keywords—Solid-State-Transformer, Medium Frequency MFT is presented. In Section Ⅳ, design of 20kW/50kHz MFT
Transformer, Medium Voltage Insulation prototype and testing are given. Detailed loss/efficiency and
impedance measurement and estimation are discussed.
I. INTRODUCTION Finally, conclusion is given in Section V.
Solid-state transformer (SST) is considered to replace the
conventional line-frequency transformer (LFT) to interface II. LOSS ANALYSIS OF MFT
medium voltage (MV) bus and low voltage (LV) bus. One aim A precise loss calculation is important to high efficiency
of SST is by using power electronics devices and medium and high power density MFT design. This section investigates
frequency transformer (MFT) to improve the power density as winding loss estimation with considering both skin and
well as efficiency. proximity effect, and core loss calculation.
MFT is a complex system consisting of multi-physical A. Winding Loss Calculation
fields including electrical, magnetic, and thermal fields.
As shown in Fig.1, an infinite cylindrical conductor is
Operating at medium frequency can increase the winding loss
supposed to be exposed in a homogeneous alternating
due to skin and proximity effect. Moreover, MFT has to
magnetic field. Its axis is perpendicular to the magnetic field.
provide a high galvanic isolation between MV side and LV
side. The insulation requirement makes it challenging to reach
high power density. z
H
According to the state-of-arts, nanocrystalline [1-3] and
ferrite [4-8] have been widely used for MFT. Due to its eddy I
current loss in the core, the nanocrystalline is usually used in
J
transformers with less than 20kHz operating frequency. For
higher frequency applications, ferrite core is preferred. r ϕ
Several prototypes with ferrite core are developed as listed in
Table I. For MFT, efficiency and power density are always a r0
trade-off. Increasing the frequency can significantly improve
the power density, while it will result in efficiency loss [5,6].
Predecessors have studied analysis and design of the Fig. 1. Schematic for cylindrical conductor loss derivation. The
transformer, which includes winding loss and core loss radius of the conductor is r0. The conductivity and permeability of
models, optimization of transformer, extraction of transformer the conductor are σ and μ respectively. The amplitude of the
parameters by either measurement or finite element analysis, alternating current and alternating magnetic field is I and H
respectively. J represents the current density on the cross section.
evaluation of loss or efficiency through testing etc. These

TABLE I. A BRIEF SUMMARY OF THE MFT USING FERRITE MATERIAL

ETH[4] ETH[5] Virginia Tech[6] Virginia Tech[7] EPFL[8]
Power 166kW 25kW 15kW 15kW 100kW
Terminal voltage 1/0.4kV 3.5/0.4kV 0.8/0.4kV 0.8/0.4kV 0.75kV/0.75kV
Frequency 20kHz 48kHz 500kHz 200kHz 10kHz
Ferrite N87 BFM8 3F36 ML27D N87
Power density 8.2kW/L 7.4kW/L 40.5kW/L N/A N/A
12kV applied voltage test
Insulation 12kV 15kV 30kV basic lightning insulation test, Partial discharge test
partial discharge test
MFT: 99.6% effi. dc/dc: 98.0% dc/dc: 98.7% MFT: 99.3% estimated effi.
Efficiency 99.5% estimated effi.
dc/dc: 99.0% effi. peak effi. peak effi. dc/dc: 97.1% effi.

This work is supported by National Natural Science Foundation of

China (grant number 51777189).
With explained in Appendix A, the loss of the cylindrical
conductor is represented by [13]:
Plitz =  P =nR
n
i dc Fr I
2
+ Rdc Gr H n
i
2

2
P = Rdc ( Fr I 2 + Gr H 2 ) = Rac (
I
2
)2 (1) 
 H 2
i S strand

 (5)
= nRdc Fr I 2 + nRdc Gr  n

where Rdc is the resistance under dc condition, Rac is the  nS strand 
 
equivalent ac resistance associated with high-frequency  
effect. Fr, Gr are coefficients related to skin and proximity where Sstrand is the cross-sectional area of each strand.
effect respectively, they are expressed as follows [13]: Assuming that the number of strands is large enough and the
cross-sectional area of each strand is much smaller than litz
ξ ber0 (ξ )bei1 (ξ ) − bei0 (ξ )ber1 (ξ ) wire cross-sectional area, so that the above summation
Fr = (
4 2 ber1 (ξ ) 2 + bei1 (ξ )2 formula can be approximated by the integral [9]:
ber0 (ξ )ber1 (ξ ) + bei0 (ξ )bei1 (ξ ) 2
− )  H 2 dS 
2
ber1 (ξ ) + bei1 (ξ ) 2
 
ber1 (ξ )bei2 (ξ ) − bei1 (ξ )ber2 (ξ ) Plitz ≈ nRdc Fr I 2 + nRdc Gr  S 
Gr = 2π 2ξ r02 ( (2)  Slitz  (6)
ber0 (ξ ) 2 + bei0 (ξ ) 2  
 
ber1 (ξ )ber2 (ξ ) + bei1 (ξ )bei2 (ξ )
− ) = nRdc Fr I 2 + nRdc Gr H srms
2
ber0 (ξ )2 + bei0 (ξ ) 2
where Slitz is the litz wire cross-sectional area, and Hsrms is the
2r0 1 spatial rms value of magnetic field H on litz wire cross section.
ξ= , δ=
δ π f σµ Define litz wire dc resistance Rdc,litz = nRdc, litz wire dc loss
where f is the frequency of the alternating field. berv and beiv Pdc,litz = Rdc,litz ( ⁄√2)2, then (6) can be expressed as:
are known as Kelvin function, which can be expressed by Plitz = 2 Pdc ,litz ( Fr + Gr ( H srms I ) 2 ) (7)
Bessel function of the first kind and v order:
According to the approaches to obtain Hsrms, there are two
J v ( j 3/2 x) = berv ( x) + jbeiv ( x) (3) methods for litz wire winding loss estimation: hybrid method
Litz wire is very popular for its low loss and flexibility in and analytical method.
the design of high-frequency magnetic components. The 1) Hybrid Method
ideally twisted litz wire enables each strand continuous Hybrid method adopts finite element method (FEM) to
transposition in space. In terms of space symmetry, each obtain Hsrms, which is inserted into equation (7) to get the litz
strand can be assumed to conduct current with same amplitude wire loss. In order to estimate the feasibility of this approach,
I. The strand dimension is generally negligible compared with a litz wire model with 343 strands is constructed as shown in
the winding dimension, therefore it’s assumed that each strand Fig. 3 (a). The FEM software is Ansys Maxwell, and the Hsrms
is exposed in a homogeneous magnetic field. Then each strand can be extracted with the software toolkit field calculator.
loss can be calculated with equation (1). Investigating a litz Fig.3 (b) shows the ratio of ac resistance and dc resistance
wire as shown in Fig. 2, the magnetic field of the ith strand with respect to a frequency range of 0~100kHz calculated by
consists of two parts: the external magnetic field Hext, and the hybrid method, as well as the Ohmic loss obtained by full
internal magnetic field which is the vector sum of the FEM simulation. As it illustrated, the Ohmic loss with FEM
magnetic field generated by other strands such as Hji and Hki, simulation is consistent to the hybrid method accurately. With
etc. Noting the magnetic field of the ith strand is Hi (the vector same derivation of (7), the hybrid method can be easily
sum of the external magnetic field and internal magnetic generalized to winding loss estimation of a transformer:
field), then the loss can be expressed as:

Pi = Rdc ( Fr I 2 + Gr Hi2 ) (4)

The litz wire loss can be derived as the sum loss of all n
Rac/Rdc

strands [9]:
Hext
Hji
Hki j y
I z
(a) x (b) Frequency (kHz)
…
i
I Fig. 3. Verification of hybrid method for litz wire loss estimation.
… (a) The FEM litz wire model with 343 strands. This model is exposed
I k in a y direction homogeneous ac magnetic field with 16000A/m
amplitude. Each strand conducts a z direction ac current with 0.1A
Fig. 2. Magnetic field schematic of a litz wire. The total number of amplitude. The depth (z direction length) of the model is 1m. (b) The
strands is n, and the ith, jth, and kth strand are drawn in this diagram. comparion of hybrid approach and FEM simulation with respect to
The external magnetic field is Hext. Hji, Hki are the internal magnetic a frequency range. The hybrid method enables to depict the losses
field generated by the jth, and kth strand respectively. caused by eddy effect and proximity effect seprately.
 known as a classical equation suitable for sinusoidal voltage
dw excitation [15]:

Pv = Cm f α B β (12)
hw
where Cm,α, β are SE coefficients, f is the frequency, B is the
amplitude of flux density under sinusoidal excitation, and Pv
is the time-average core loss per unit volume. Reference [10]
proposed the improved Generalized Steinmetz Equation
H Hmax (iGSE) and the approaches to deal with minor hysteresis
loops. iGSE is considered to be suitable for arbitrary flux
x density waveforms and with the following form:
Fig. 4. Simpified magnetic field distribution in transformer winding
α
for analytical method derivation. 1 T dB

β −α
Pv = k ∆B dt
Pw = 2 Pdc , w ( Fr + Gr ( H srms , w I ) )2
(8) T 0 dt
Cm (13)
where Pdc,w is the dc Ohmic loss of a transformer winding. k=
1.7061
Hsrms,w is the spatial rms value of the magnetic field of the 2 β +1π α -1 (0.2761 + )
winding, and it also can be extracted by FEM software. α + 1.354
where T is the time period of flux density. ΔB is the peak-to-
2) Analytical Method peak flux density. Coefficients Cm, α and β are generally
Another method for Hsrms calculation is analytical method provided by the core manufacturer. A MATLAB file is
[1]. To simplify the mathematics derivation, the litz wire provided by the authors of [10] for convenient application
winding is taken as foil winding with a uniform current [16].
distribution as illustrated in Fig. 4. Then the magnetic field
distribution in x direction can be simplified to linear. Taking Actually SE coefficients mentioned before vary with
the left foil winding as instance, the spatial rms value of H can temperature, frequency as well as flux density. In order to
be expressed as: improve the calculation accuracy under different operation
conditions, this paper adopts the loss map concept [11].
H max According to loss data provided by the manufacturer, a loss
H srms , w = (9)
3 map consisting of the loss under different temperature,
frequency and flux density can be constructed. Then by linear
where Hmax is the max value of H in x direction. Applying interpolation, the loss data under a certain operation condition
Ampere’s law to the blue loop, Hmax is obtained: are obtained. In our design, we combine iGSE with loss map
to increase accuracy of the core loss estimation.
nNI
H max = (10)
hw III. OPTIMIZATION OF MFT
where n is the strand number of each litz wire, N is the number In this paper, the MFT is used in a 20kW isolated LLC
of turns of this winding, and I is the amplitude current of a dc/dc converter as shown in Fig. 5. This converter operates as
strand. Combining (8)(9) and (10), the analytical formula for a 750V/750V dc transformer. The currents conducted in the
winding loss calculation is as follows: transformer can be approximated as sinusoidal waveform with
42A peak value. The leakage inductance and magnetizing
nN inductance are tightly associated with the optimized frequency
Pw = 2Pdc, w ( Fr + Gr ( )2 ) (11) and modulation scheme. At the beginning, the Lm is designed
3hw as 800uH and Lr is designed with a max value 50uH.
The analytical formula of winding loss calculation is Similar to [4][5][8], EPCOS N87 is selected as the core
suitable for iteration process in software. But due to the material for the consideration of its low core loss density and
simplification, this method causes some error. In contrast, the its availability for large power cores. Choosing shell type
hybrid method features high accuracy. However, since it structure as shown in Fig. 6 (b) for the sake of heat dissipation
needs to run FEM simulation for each calculation, it’s difficult and low leakage inductance. When the core material and
for programming optimization consisting of huge amount of structure are fixed, the free parameters are dimensions (i.e.
iterations. And running FEM simulation with high precision core width tc, core length lc, winding window width dw,
is also time-consuming.
In addition to the approaches mentioned above, the new Iout
edition of Ansys Maxwell software supports litz wire S1 S3 S5 S7
simulation in Eddy Current Solver and Transient Solver [14]. ip Lr is
This provides a convenient way to obtain the winding loss of U1 vp Cr im Lm vs Cd +U2 RL
a transformer. This method is named as FEM simulation
1:1
-
approach in this paper.
S2 S4 MFT S6 S8
B. Core Loss Calculation
Various core loss calculation methods have been studied
by predecessors. Among them, the Steinmetz Equation (SE) is Fig. 5. LLC dc/dc converter topology. The leakage inductance is
intergrated as the resonant inductance in this converter.
 lc
Pareto front with
optimization results

hw

Efficiency (%)
Prototype with
estimated effi.

Selected design
dw tc Prototype with
(a) (b) measured effi.

Fig. 6. Structure and dimensions of the transformer. (a) 2D (b) 3D.

TABLE II. THE RANGE OF FREE PARAMETERS

Parameter Range
Power Density (kW/L)
Frequency f (kHz) [25, 100] Fig. 7. The results of Pareto optimization.
Flux density B (T) [0.05, 0.3]
Core width tc (mm) [10, 100]
Core length lc (mm) [10, 100]
Winding window width dw (mm) [10, 100]
Winding window height hw (mm) [10, 100]

winding window height hw), flux density B and operating

frequency f. As illustrated in appendix B, the geometrical
parameters including core area Ac, winding window area Aw,
core volume Vc, winding volume Vw, heat dissipation area of
core Atc, heat dissipation area of winding Atw, and transformer
volume VTrX can be expressed with tc, lc, dw and hw. Atc and Atw
are the core and winding area which is exposed in air
respectively, and VTrX is the transformer volume including Fig. 8. MFT prototype. According to the power flow direction from
MV side to LV side, the MV winding is named as primary side
core and winding. winding and the LV Winding is named as secondary side winding.
MFT design is taken as a multi-objective optimization used under natural and forced convection conditions. Heat
problem. The objectives are maximizing efficiency and power conduction and radiation processes are not considered in this
density, which can be expressed with the six free parameters model. The convection coefficients in (14) are under natural
according to the loss analysis in Section II. The constraints are convection condition (wind speed 3m/s).
insulation (15kV) and heat dissipation (40℃ temperature
rising for both core and winding), and can be described by This optimization problem has six free parameters. After
following inequality: specifying the parameters range as shown in Table II, the
Pareto front with respect to efficiency and power density can
( N p I p , rms + N s I s , rms ) / J 0 ≤ ku Aw be obtained by genetic algorithm (GA).

( Pw )1 1.09 ≤ ∆T Fig. 7 shows the Pareto front. The optimized parameters
 0.89 max
illustrate that the frequency f converges around 50kHz and
 12 Atw (14)
 P the flux density B converges around 0.1T. For achieving 99%
( c
)1 1.09 ≤ ∆Tmax efficiency of the dc/dc converter, the blue star with 99.7%
 12 Atc0.89 efficiency and 8.2kW/L power density for the transformer is
selected as the targeted design. Three pairs of U93/58/30 core,
where Np is the turn number of the primary winding, Ns is the which are cut from standard U93/76/30 cores, are stacked
turn number of the secondary winding, Ip,rms is the rms current together to form the desired cores. According to the
of the primary side, Is,rms is the rms current of the secondary optimized results, the turn numbers are obtained with an
side. ku is the winding filling factor, ΔTmax is the limited arrangement shown in Fig. 6 (a). After the trade-off of high
temperature rising, and J0 is the current density expressed by: frequency losses and insulation distance, 0.1mm diameter
S with 1800 strands litz wire is selected.
J0 = (15)
K v ku fBAw Ac For insulation design, the MV winding (primary side) is
packed with mica tape and fixed with polyesterimide (PI) film.
where S is the transformer apparent power, Kv is the waveform The winding will be encapsulated with epoxy resin, while at
coefficient, in this case, Kv =4. The first inequality in (14) is present, the clearance between MV winding (primary side)
insulation constraint, which means that the copper area in the and LV winding (secondary side) is separated with NMN
winding cross section is limited by winding filling factor ku. insulation paper for the consideration of adjustable
The insulation thickness (around primary winding) is selected magnetizing inductance.
as 5mm for 15kV insulation. After several iterative attempts,
selecting ku=0.2 can realize the insulation thickness design. IV. EXPERIMENT
The second and third inequality represent required area for A MFT prototype is built according to the optimized
winding heat dissipation and core heat dissipation design as shown in Fig. 8. Table III lists several key
respectively. This thermal model is based on the convection parameters of the prototype. The red star of Fig. 7 illustrates
coefficient and the exposed area of the MFT [17]. It can be
 Fig. 9: Ciruit and waveforms for impedance measurement. (a) Secondary side winding is opened. (b) Secondary side winding is shorted.
(c)Primary side winding is shorted.

TABLE III. PARAMETERS OF THE DESIGNED MFT Lkp RCu,p 1:1

Parameter Value + Lks
Rated power 20kW Full- i Lm
Operating frequency 50kHz bridge vm m
Terminal voltages ±750V/±750V Inverter Rcore RCu,s
Turn ratio 16:16 turns - Cr1 ESR1
Litz wire 0.1mm×1800 for both side
Air gap 2×0.5mm
Fig. 10. The diagram of winding loss measurement.
TABLE IV. MEASURED INDUCTANCE BY DIFFERENT APPROACHES
reduce the sensitivity to phase discrepancy, which is the
Impedance analyzer Current slope FEM
Lkp 20.3 H 18.5 H 21.4 H
dominating error source in the conventional two-winding
Lks 9.1 H 10.7 H 8.5 H approach [12,18]. Apart from the individual measurement of
Lm 805.9 H 732.0 H 845.1 H winding and core loss, the transformer efficiency is also
estimated by measuring efficiency of a dc/dc converter with
that the prototype has 7.3kW/L power density and 99.7% the transformer under test.
estimated efficiency. In order to verify the design, this section
performs the experimental testing of MFT. Tests include 1) Winding loss measurement
impedance measurement, loss and efficiency measurement A diagram to measure the winding loss is shown in Fig.10.
and insulation test. RCu,p and RCu,s are the equivalent resistance that represent the
primary winding loss and secondary winding loss respectively.
A. Impedance Measurement The measured dc resistance of the primary winding and
The impedance measurement is implemented by two secondary winding is 10.6mΩ and 7.1mΩ respectively. They
approaches. The first one is using impedance analyzer will have an increased value when the transformer operates at
‘Keysight E4990A’ to measure the open-circuit and short- a high frequency due to the skin effect and proximity effect.
circuit impedance of each side. Leakage inductance and Rcore represents the core loss and it’s estimated as 3Ω under
magnetizing inductance can be obtained according to the the designed operating condition. Actually, the core loss varies
equivalent circuit. The second one is to measure the current with the frequency, flux density as well as the temperature. For
slope under a square voltage excitation to calculate the this testing circuit, the secondary side winding is shorted, and
inductance. The testing circuit and the measured waveform are the primary side is connected to a full-bridge inverter. Since
shown in Fig. 9. In addition, this paper constructs a 3D FEM Lm>>Lkp and Rcore>>RCu,s, the current flows in Lm branch is
transformer model as shown in Fig.6 (b) to estimate the negligible. The full-bridge inverter produces a 50kHz square
inductance. The inductance measured by different methods waveform as the excitation voltage for the series LC resonant
are listed in Table IV. Lkp, Lks and Lm correspond to the circuit. The leakage inductance resonates with the resonant
equivalent circuit in Fig. 9. capacitor Cr1 to generate the sinusoidal current with 50kHz
frequency. This current amplitude can be adjusted by changing
The impedance analyzer used in this experiment features the dc source amplitude of the full bridge inverter. The
about 0.1% precision when operating at 50kHz. The possible resonant frequency of this circuit is as follows:
error of current slope method results from the precision of the
voltage and current probes and the reading error of the 1
f res = (16)
oscilloscope. The error of simulation relates to the difference 2π ( Lkp + Lks )Cr1
between the FEM model and the transformer prototype. As
illustrated in Table IV, although different approaches have The active power provided by the full-bridge inverter is:
different measured values, their deviation is within an
1
acceptable range. PFB1 =
T 
vm ⋅ im dt
T
(17)
B. Loss and Efficiency Measurement
Two-winding method is widely used to measure the where T is the period of the square waveform. The energy
individual winding loss and core loss. An improved two- provided by the inverter is consumed by the resistance RCu,p,
winding method with capacitor compensating is used to
 vm 5us/div Lkp RCu,p Lks RCu,s
(25V/div) + +
ϕ Full-bridge vm im Rcore
0 Inverter Lm vB
Cr2 ESR2
-
-
vgs im
(20V/div) (20A/div) Fig. 14. The diagram of core loss measurement.

2
0 ∆PFB = (−Vm I m sin(ϕ )∆ϕ + I m cos(ϕ )∆Vm + Vm cos(ϕ )∆I m ) (19)
π
Fig. 11. The experimental waveform of winding loss measurement.
The relative error is represented by:
The vm, im correspond to Fig. 10. vgs is the drive signal of the full-
bridege inverter and φ is the phase discrepancy casused by ∆PFB ∆V ∆I
mismatching of resonant parameters. = − tan(ϕ )∆ϕ + m + m (20)
PFB Vm Im

Thus the relative error resulting from phase discrepancy,

accuracies of voltage and current probes is as follows:
∆ϕ = − tan(ϕ )∆ϕ
Loss (W)


∆ = ∆Vm
 V
 Vm (21)

 ∆ I = ∆I m
 Im

The total relative error is the sum of three items in (21).

Among these three items, Δφ is critical to the total error. Errors
Peak Current (A) ΔLr and ΔCr in resonant inductance and resonant capacitor,
Fig. 12. Measured winding loss as well as the theoretical results with and the ESR of resonant capacitor contribute to phase
respect to different operating current. discrepancy φ. Its resulting error can be calculated as [12]:
Im im v ω ( Lr + ∆Lr ) −
1
Vm m ω (C r + ∆ C r )
ϕ t ∆ϕ = ± tan(ϕ ) ∆ϕ = ± ∆ϕ (22)
RCu , p + RCu , s + ESR1

For the measurement setup, f=49.9kHz, ω=2πf, Lr=Lkp

Fig. 13. The diagram of a typical vm and im waveform for error +Lks=29.3uH, Cr=336.8nF. By considering |ΔLr|≈60nH,
analysis.
typical ΔCr≈±30nF, ESR1≈4.11mΩ, minimum RCu,p+RCu,s
+ESR1≈ 21.8mΩ. Thus tan(φ) can be calculated as about 56.3.
RCu,s and the ESR of the resonant capacitor. Thus, by Typical Δφ=0.00079 (2.5ns time delay between the two
subtracting the ESR part, the remained one can be regarded as probes under 50kHz excitation). The maximum relative error
the winding loss. caused by phase discrepancy is about ±4.5%.
Fig. 11 shows the experimental waveform of winding loss CYBERTEK’s differential voltage probe DP6150B
measurement. In this waveform, im is tuned to achieve 42A (200MHz bandwidth and ±2% accuracy) and CYBERTEK’s
peak to simulate the rated power condition. Fig. 12 shows the current probe CP8150A (12MHz bandwidth and ±1%
measured winding loss with different operating current, which accuracy) are used in this measurement. By considering the
is compared with three different theoretical results as accuracy of voltage and current probes, the possible relative
discussed in Section II. It can be illustrated that the measured error of the winding loss measurement is about ±8%.
winding loss basically accord with the theoretical results. The
experiment shows that the winding loss is 43.9W when 2) Core loss measurement
operating at 41.4A peak current. A diagram to measure the core loss is shown in Fig.14.
Different from the winding loss measurement, the secondary
According to [12,18], the main error of this measurement side winding is opened. The primary side winding is
results from phase discrepancy and errors of probes. For the connected to the full-bridge converter. The full-bridge inverter
waveform shown in Fig. 13, the active power provided by the produces a 50kHz square waveform as the excitation voltage
full-bridge inverter is: for the series LC resonant circuit. The magnetizing inductance
Lm is excited by the voltage vB. By adjusting the dc voltage of
1 2
PFB =
TTvm ⋅ im dt = Vm I m cos(ϕ )
π
(18) the inverter, vB can be tuned to change the flux density. The
resonant capacitor is adjusted to match the resonant frequency
around 50kHz:
where Vm is the amplitude of square waveform vm, and Im is
the amplitude of sinusoidal waveform im. With differentiating 1
(18), the power measurement error can be obtained as: f res = (23)
2π ( Lkp + Lm )Cr 2
 im

Voltage (kV)
5us/div vp
(5A/div)
0
ϕ vs
vm
(25V/div)
0
vB vgs

Current (A)
(1kV/div) (20V/div)
is ip
0
im
Fig. 15. The experimental waveform of core loss measurement. vm,
im, vB correspond to Fig. 14 and vgs is the drive signal of the full-
bridege inverter. φ is the phase discrepancy casused by mismatching
of resonant parameters. Time (us)
Fig. 17. Key waveforms of the dc/dc converter operating at 20kW.
vp and vs are the terminal voltage of the LLC resonant tank, ip, is are
primary side current and secondary side current repectively, and im
is the magnetizing current. The voltage and current are noted in
dc/dc converter topology as shown in Fig. 5.

Efficiency (%)
Loss (W)

(a) Power (kW) (b)

Peak flux density B (mT) Fig. 18. Measured results of dc/dc converter. (a) Measured dc/dc
Fig. 16. Measured core loss as well as the datasheet data with respect converter and MFT efficiency. (b) Loss breakdown of dc/dc
to different peak flux density. converter at rated power.
Typical Δφ=0.00079 (2.5ns time delay between the two
With the similar approach, the core loss can be obtained probes under 50kHz excitation). The maximum relative error
by excluding the extra loss of ESR and RCu,p. Fig.15 shows the caused by phase discrepancy is about ±1.5%.
experimental waveform of core loss measurement. It should
be noted that vB is square waveform with 750V amplitude in CYBERTEK’s differential voltage probe DP6150B
dc/dc operating condition, while in this measurement vB is (200MHz band width and ±2% accuracy) and CYBERTEK’s
sinusoidal waveform. According to the rated condition, vB is current probe CP8150A (12MHz band width and ±1%
tuned to achieve 1025Vpeak (corresponding to 93mT flux accuracy) are used in this measurement. By considering the
density). accuracy of voltage and current probes, the relative error of
the core loss measurement is about ±5%.
It should be mentioned that the core loss is dependent on
the temperature. Fig. 16 gives the core loss of the datasheet 3) Efficiency estimation
under different temperature. When the measurement is A dc/dc converter prototype is built and the efficiency is
performed, the core temperature is between 30℃ and 40℃. measured. Fig. 17 shows the key waveforms of the dc/dc
As shown in Fig. 16, the measured core loss accords with the converter operating at the rated power. The terminal voltage
datasheet data. The measured core loss is under the sinusoidal of the resonant tank is 750V square waveform and the
excitation. In order to reflect the practical operation, iGSE can resonant current flowing in the transformer reaches 42Apeak.
be used for conversion. In this case, the loss under sinusoidal The loss of the dc/dc converter includes four parts: the
excitation is about 1.1 times of the loss under square excitation. transformer loss, the turn-off loss, the conduction loss as well
The measured core loss is about 31.1W under sinusoidal the capacitor loss. The dc/dc converter efficiency as well the
excitation at around 40℃. loss breakdown at rated power is shown in Fig. 18. The
transformer part loss can be extracted according to the
Using (21), the relative error of core loss measurement measured dc/dc converter efficiency and loss breakdown. The
can be analyzed. The relative error Δφ caused by phase dc/dc converter efficiency is measured by ZLG PA6000H
discrepancy φ is calculated as [12]: power analyzer. By experiment, the dc/dc converter has
99.0% efficiency at rated power. Then the efficiency of the
1
ω ( Lr + ∆Lr ) − transformer is estimated to be 99.6% by taking off the turn-off
ω (Cr + ∆Cr ) loss, the conduction loss of the power devices and the
∆ϕ = ± tan(ϕ )∆ϕ = ± ∆ϕ (24)
RCu , p + RCore + ESR2 capacitor loss.
For the measurement setup, f=47.5kHz, ω=2πf, Lr=Lk,p C. Insulation Test
+Lm=826.2uH, Cr=12.3nF. By considering |ΔLr|≈0.9uH, For the transformer insulation, a test is performed with the
typical ΔCr≈±1.2nF, ESR2≈97.8mΩ, minimum RCu,p+RCore circuit shown in Fig. 19. The MV winding (primary side) is
+ESR2≈ 3.11Ω. Thus tan(φ) can be calculated as about 17.9.
  S
J s J p dS = 0 (27)

Then equation (26) can be derived as:

Transformer
under test
P=
1
2σ (  J dS +  J dS )
S
2
s
S
2
p (28)
MV generator
Equation (28) illustrates that the loss of the conductor
consists of two parts: the loss only caused by skin effect and
the loss only caused by proximity effect. Each part can be
calculated independently. This appendix gives a derivation of
1:1 loss caused by skin effect (H = 0). In a cylindrical conductor,
Line
Frequency
Lkp Lks there is no static charge and the displacement current can be
MV Lm ignored. According to Maxwell equation, we can get:

Fig. 19. Insulation test setup and the connection circuit. ∇2 J&s = jk 2 J&s , k = ωµσ (29)

shorted and then applied with 15kV rms line frequency Unfold the Laplacian operator in the cylindrical coordinate
voltage. The transformer passes the 1 minute test successfully. system. Due to the point symmetry, the partial derivative of
During the 1 minute test, the insulation is stable and no current density with respect to φ is zero. Thus, the following
insulation breakdown happens. equation can be obtained:

V. CONCLUSION ∂ 2 J& s 1 ∂J&

2
+ ⋅ s − jkJ& s = 0 (30)
The MFT is the key element of an SST. It needs high ∂r r ∂r
efficiency, high power density and to satisfy insulation Let x=αr, α=j3/2k, we can get the following equation:
requirement. For this purpose, the design and testing of a
20kW MFT are investigated in this paper. A detailed design x 2 J&s′′ + xJ&s′ + x 2 J&s = 0 (31)
procedure is discussed and an MFT prototype with 7.3kW/L
power density is built. Secondarily the measurement of the (31) is a Bessel function with first order, and its solution
transformer parameters is introduced. Then winding loss and is:
core loss testing are discussed. Finally, a DC-DC converter
with the MFT is built to further estimate the transformer J&s = CJ 0 (α r ) + C1Y0 (α r ) (32)
efficiency. For experimental testing, its impedance parameters
accord with the design, and it passes 15kV line frequency Where J0 is Bessel function of the first kind and zero order,
insulation test successfully. The loss and efficiency of the Y0 is Bessel function of the second kind and zero order. Since
prototype are measured and the results show that this MFT J is finite at r = 0, C1 = 0. Because the current passing through
features 99.6% efficiency. Actually, the insulation discussion the conductor is I, thus:
in this paper is limited, and more detailed and comprehensive r0
research about insulation will be studied in the future. I& = 
S
J& s ( r , ϕ ) dS = 2πC  0
rJ 0 (α r ) dr (33)

APPENDIX A According to the integral recurrence property of Bessel

This derivation is based on the cylindrical conductor function, the value of constant C can be calculated. In this
shown in Fig.1. Assuming that the current density distribution way, the current density distribution in the conductor is as
in the conductor is Js(r,φ) when only skin effect is considered follows:
(H = 0), and Jp(r,φ) when only proximity effect is considered & (α r )
(I = 0). Then the actual current distribution in the conductor is α IJ
J& s = 0
(34)
as follows: 2πr0 J1 (α r0 )
J ( r,ϕ ) = J s ( r ,ϕ ) + J p ( r ,ϕ ) (25) The loss caused by skin effect can be calculated as:
According to Ohm’s law, the loss of the conductor is: 1 π r0
 J& J& σ
J& s J& s* rdr
*
Ps = s s dS = (35)
2σ S 0
1 1 2
P=
2σ  S
J 2 dS =
2σ  ( J
S
s + Jp ) dS
(26)
Using Lommel's integral, we can get the following results:

=
1
2σ (  J dS +  J dS +2 J J dS )
S
2
s
S
2
p
S
s p Ps =
kI 2
4 2πσ r0
ber bei − bei0ber1 − ber0ber1 − bei0bei1
⋅ 0 1
ber12 + bei12
(36)

When only skin effect is considered, the distribution of

current density is point symmetric, so the current density is an Where berv represents berv(kr0), and beiv represents beiv(kr0).
even function about the polar axis. When only considering (36) is the original form of loss caused by skin effect in a
proximity effect, the current density is an odd function with cylindrical conductor. Using the same concept, the loss caused
respect to the polar axis. Therefore, the current density by proximity effect can be derived. Here we only give the
distribution caused by skin effect and proximity effect is current density distribution in the conductor and the loss
orthogonal to each other, then the following equation is expression:
obtained [9]:
 [5] T. Guillod, “Modeling and Design of Medium-Frequency
 ( )
J j 3/2 kr
 J& ( r , ϕ ) = 2 j 3/2 kH& ⋅ 1 sin ϕ
Transformers for Future Medium-Voltage Power Electronics
Interfaces,” Ph.D. dissertation, ETH Zürich, 2018.
 p

( )
J 0 j 3/2 kr0
(37) [6] S. Zhao, Q. Li, F. C. Lee and B. Li, "High-Frequency Transformer
 2πkr0 H 2 ber1bei2 − bei1ber2 − ber1ber2 − bei1bei2 Design for Modular Power Conversion from Medium-Voltage AC to
 Pp = ⋅ 400 VDC," in IEEE Transactions on Power Electronics, vol. 33, no. 9,
 σ ber0 2 + bei0 2 pp. 7545-7557, Sept. 2018.
[7] Z. Li, Y-H. Hsieh, Q. Li, F. C. Lee and M. H. Ahmed, "High-Frequency
APPENDIX B Transformer Design with High-Voltage Insulation for Modular Power
TABLE V. THE EXPRESSIONS OF GEOMETRICAL PARAMETERS Conversion from Medium-Voltage AC to 400-V DC," 2020 IEEE
Energy Conversion Congress and Exposition (ECCE), Detroit, MI,
Parameters Expressions USA, 2020, pp. 5053-5060.
Core area Ac tclc [8] M. Mogorovic and D. Dujic, "100 kW, 10 kHz Medium-Frequency
Winding window Transformer Design Optimization and Experimental Verification," in
dwhw
area Aw IEEE Transactions on Power Electronics, vol. 34, no. 2, pp. 1696-
Core volume Vc 2tclc(hw+2tc)+2dwlctc 1708, 2019.
Wingding volume [9] J. A. Ferreira, “Analytical computation of AC resistance of round and
(pidw+2tc+2lc)dwhw=MLT Aw
Vw rectangular litz wire windings,” in IEE Proceedings B (Electric Power
Heat dissipation (2(dw+2tc)+hw+2tc)lc+2(2(dw+2tc)tc+hwtc) Applications), vol. 139, pp. 21-25, Jan. 1992.
area of core Atc +2tclc
[10] K. Venkatachalam, C. R. Sullivan, T. Abdallah and H. Tacca,
Heat dissipation
(2pidw+2tc+lc)hw+2dw(pidw+2tc+lc) "Accurate prediction of ferrite core loss with nonsinusoidal waveforms
area of winding Atw using only Steinmetz parameters," 2002 IEEE Workshop on Computers
Transformer in Power Electronics, 2002. Proceedings., Mayaguez, Puerto Rico,
2(tc+dw)(lc+2dw)(hw+2tc)
volume VTrX USA, 2002, pp. 36-41.
[11] J. Mühlethaler, J. W. Kolar, and A. Ecklebe, "Loss modeling of
REFERENCES inductive components employed in power electronic systems," in 8th
[1] M. Leibl, G. Ortiz, and J. W. Kolar, "Design and Experimental International Conference on Power Electronics - ECCE Asia, 2011,
Analysis of a Medium-Frequency Transformer for Solid-State pp. 945-952.
Transformer Applications," IEEE Journal of Emerging and Selected [12] M. Mu, Q. Li, D. J. Gilham, F. C. Lee and K. D. T. Ngo, "New Core
Topics in Power Electronics, vol. 5, no. 1, pp. 110-123, 2017. Loss Measurement Method for High-Frequency Magnetic Materials,"
[2] Z. Guo, R. Yu, W. Xu, X. Feng and A. Q. Huang, "Design and in IEEE Transactions on Power Electronics, vol. 29, no. 8, pp. 4374-
Optimization of a 200-kW Medium-Frequency Transformer for 4381, Aug. 2014.
Medium Voltage SiC PV Inverters," in IEEE Transactions on Power [13] J. Lammeraner and M. Stafl, Eddy Currents. Iliffe Books Ltd., 1966.
Electronics.
[14] [Link]
[3] C. Zhao et al., "Power Electronic Traction Transformer—Medium
[15] E. C. Snelling, Soft Ferrites, Properties and Applications, 2nd ed.
Voltage Prototype," IEEE Transactions on Industrial Electronics, vol.
London, U.K.: Butterworths, 1988.
61, no. 7, pp. 3257-3268, 2014.
[16] [Link]
[4] G. Ortiz, M. Leibl, J. W. Kolar, and O. Apeldoorn, "Medium frequency
transformers for solid-state-transformer applications — Design and [17] V. C. Valchev and A. Van den Bossche, Inductors and Transformers
experimental verification," in 2013 IEEE 10th International for Power Electronics. Taylor & Francis, 2005.
Conference on Power Electronics and Drive Systems (PEDS), 2013, [18] N. He, M. Chen, J. Wu, N. Zhu and D. Xu, "20-kW Zero-Voltage-
pp. 1285-1290. Switching SiC-mosfet Grid Inverter with 300 kHz Switching
Frequency," in IEEE Transactions on Power Electronics, vol. 34, no.
6, pp. 5175-5190, June. 2019.

View publication stats