Modeling and Design of a Medium-Frequency Transformer for High-Power … · 2020. 1. 27. · Modeling and Design of a Medium-Frequency Transformer（Milos Stojadinoviˇ c´ et al.）

IEEJ Journal of Industry ApplicationsVol.8 No.4 pp.685–693 DOI: 10.1541/ieejjia.8.685

Paper

Modeling and Design of a Medium-Frequency Transformer forHigh-Power DC-DC Converters

Milos Stojadinovic∗a)Non-member, Jurgen Biela∗ Non-member

(Manuscript received Aug. 31, 2018, revised Feb. 27, 2019)

Dual Active Bridge (DAB) converters are an interesting solution for the battery interfaces of storage systems usedin traction applications. Due to the environmental conditions and space limitations, the design of the transformer andcooling system is crucial for achieving a high power density. Therefore, in this paper, a detailed design of a transformerwith an integrated liquid cooling structure and high isolation voltage is presented. Analytic models for the design arepresented and verified through FEM simulations and measurements on a prototype system.

Keywords: medium-frequency transformer, modeling, optimization, isolated DC-DC converter, battery storage

1. Introduction

High power DC-DC converters with galvanic isolation area key element of many applications, as for example mediumvoltage DC (MVDC) grids (1)–(4), solid-state transformers andpower supplies for traction (5)–(8), or more-electric ships (9). An-other application are battery interfaces used in electric ve-hicles or traction systems (10) (11). In Fig. 1/Table 1 a possiblesetup and specifications of such a battery interface for a lo-comotive are given. The system enables to store recuperatedenergy during braking, what reduces the total energy con-sumption and enables reuse of the recuperated energy duringthe acceleration phase. Additionally, energy stored in the bat-teries can be used to drive the locomotive on non-electrifiedtracks without a diesel engine, avoiding CO2 emissions (e.g.in shunt yards).

The considered battery interface is based on a dual activebridge (DAB) converter (12). The series connection of modulesat the medium voltage secondary side allows to use switch-ing devices with lower voltage ratings for all switches. It alsomakes the transformer isolation requirements more severe. Ahigh efficiency and high power density is required due to thespace limitations in the locomotive. The high power densityis achieved by optimising the converter design and by push-ing the switching frequency of the semiconductor devicesto higher values under ZVS condition. Several examplesfor high power medium voltage DC-DC converters with gal-vanic isolation can be found in the literature (1) (4)–(8). Figure 2gives a comparison of some of these converter systems in theefficiency-power density plane based on data provided in theliterature. For achieving a high power density, the advancedcooling concepts (13) have to be used in transformer. Manydesigns employ a coaxially wound transformer (5) (14)–(16). Thecooling of such a transformer is usually achieved by usinghollow inner conductors (5) (15) (16) through which the de-ionized

a) Correspondence to: Milos Stojadinovic. E-mail: [email protected]∗ Laboratory for High Power Electronic Systems, ETH Zurich

Physikstrasse 3, 8092 Zurich, Switzerland

Fig. 1. Modular DC-DC converter system based onDAB topology

Table 1. Specifications of the battery interface consist-ing of 4 modules

Parameter ValuePrimary side voltage Vp 518 V..835 VNominal primary voltage 700 VSecondary side voltage Vs 2800 V (4×700 V)Current per battery, continuous 220 ACurrent per battery, peak 280 ASystem power, continuous 200 kWRated nominal withstand voltage 2.8 kVEfficiency > 95 %Power density > 5 kW/dm3

Ambient temperature 75 ◦CWater temperature 60 ◦C

water is pumped. Another investigated transformer structureis the shell type geometry. The cooling of such a structure canbe achieved either with natural or forced convection (17)–(19), us-ing aluminium plates (8) or via heat pipes (20) to conduct theheat from the windings to the core-mounted heat sinks. Dueto the switching frequencies in the kHz range, the eddy cur-rent losses induced in the structures with aluminium coolingplates lead to higher than predicted temperatures in the trans-former. To cope with this issue a thermally conductive coilformers can be used in order to conduct heat from windings to

c© 2019 The Institute of Electrical Engineers of Japan. 685

Modeling and Design of a Medium-Frequency Transformer（Milos Stojadinovic et al.）

Fig. 2. Efficiency - power density comparison of thepresented system to the previous state of the art solu-tions. All the results are given for a single module withfull isolation rating of the transformer. For the designpresented in (8), both calculated and measured efficiency -power density values are given, where the representsthe measured values. The presented design only showsthe calculated efficiency - power density values

the aluminium heat sink mounted on the transformer core (21).Unfortunately, the thermal conductivity of such coil formersis considerably lower than for aluminium, which results inlarger transformer volumes.

In order to increase the power density, a transformer withintegrated liquid cooling is presented in this paper. Thecooling system can be operated with water/glycol as a cool-ing medium, despite the high nominal isolation voltage ofViso = 2.8 kV. For designing the integrated cooling struc-ture, analytical thermal models are presented and verifiedwith FEM simulations and measurements. In section 2 firstthe short overview of the design procedure is outlined, fol-lowed with used models for the transformer design. There,special attention is dedicated to the thermal modeling of theintegrated cooling structure with water channels. Results ofFEM simulations for different design aspects are given in sec-tion 3. Finally, experimental measurements on a prototypesystem are presented in section 4, followed by conclusions.

2. Transformer Design

The system shown in Fig. 1 is used as battery interface inlocomotives. Since the secondary side (DC link side) is con-nected in series, the nominal primary and secondary voltagesare approximately equal (Table 1). This enables to use thesame switching devices for both full bridges. For the pre-sented system 1200 V SiC MOSFET devices are employed.Using SiC MOSFETs under ZVS conditions, enables higherswitching frequencies.

In order to determine the parameters for the highest powerdensity, an optimization of the converter modulation schemeand the transformer (Fig. 3) is performed for worst case in-put conditions. The modulation parameters of the DAB con-verter are the phase shift angle φ, clamping interval of theprimary side bridge δ1, the clamping interval of the sec-ondary side bridge δ2 and frequency fs. Illustrative expla-nation of the modulation parameters is shown in Fig. 4. Moredetails about the used modulation schemes and the descrip-tion of the implementation are given in (22). The optimization

Fig. 3. Simplified flow chart for the optimization proce-dure to find the optimal control variables φ, δ1, δ2, fs andthe optimal transformer design for worst case conditions,by minimizing the DAB converter losses and volume

Fig. 4. Exemplary voltage and current waveforms at thetransformer terminals for a random set of control param-eters [φ2, δ1, δ2]

is multi-objective, i.e. both the system volume and the to-tal system losses are minimized. Starting with the systemspecifications (Table 1) at step 0 in Fig. 3, first the converterelectrical model for the initial control parameters is derivedin step 1, giving at the output the voltage-current waveformsand required leakage inductance (Lσ). The voltage-currentwaveforms (Fig. 4) are used to calculate the losses in theswitching devices in step 2. In step 3, the calculated out-put power Pout (defined by the control parameters) is com-pared to the required power Pnom. In the same step, semi-conductor losses are calculated and verified to be below themaximum specified losses. If any of the constraints is notfulfilled, the control parameters are changed and the proce-dure restarts. In step 4, the transformer optimization loopis executed which determines a suitable core geometry and

686 IEEJ Journal IA, Vol.8, No.4, 2019


Fig. 5. Exploded view CAD drawing of the designedtransformer with integrated cooling structure

winding arrangement that fulfils the following constraints:peak flux density Bm ≤ Bsat, leakage inductance L = Lsigma

and temperature rise T ≤ Tmax. If any of the constraint isnot met the calculation restarts with new control parameters.After all the feasible designs are obtained in step 5, the opti-mal design (step 6) is chosen at the knee point of the paretocurve. For the considered system, major challenges are thedesign of the transformer and the efficient heat removal. Inorder to improve the heat removal, foil conductors are usedfor the transformer windings. Using foil only slightly in-creases the eddy current losses in the windings comparedto the litz wire (23), but due to the large copper filling factorand surface area, foil windings are preferred from a ther-mal and a size point of view. In Fig. 5 the CAD drawingof the transformer with integrated cooling system is shownfor the specifications given in Table 1. One of the specifi-cations for the transformer and semiconductor cooling sys-tem is to use tap water. Therefore, the whole liquid coolingstructure is at the ground potential. The cooling channelsare used for cooling the switches and the transformer. Thechannel structure can be seen in Fig. 12. Because of the highisolation requirement between the transformer windings, thecooling channels are placed only on the outer core legs ofthe transformer. For removing the heat from the transformerprimary winding, first an aluminium bar was considered as apart of the bottom cooling part (Fig. 14). The bar is placedbetween the winding and the transformer middle leg. Dueto the induced eddy current losses, the aluminium bar wasreplaced with an aluminium nitride (AlN) bar (Fig. 15). Alu-minium nitride is an electrical isolator which offers high ther-mal conductivity equivalent to thermal conductivity of alu-minium (Table 2). For cooling the secondary winding andfulfilling the isolation requirement, the transformer is pot-ted using a thermally conductive casting compound WepesilVU 4675, which offers a wide operating temperature rangeand low hardness. For isolating the secondary side switches,an AlN plate is used to separate the secondary side coolingstructure, on which the secondary side switches are mounted,and the grounded bottom cooling part. The fixation of thealuminium and AlN plate to the main structure is achievedwith nylon glass filled bolts. By increasing the outer diam-eter and rounding the edges of the holes in the aluminiumparts (see Fig. 6) it is possible to shape the e-field in order tofulfil the given isolation requirements. Table 2 lists the speci-fications of the designed medium frequency transformer with

Table 2. Specifications of the medium frequency trans-former

Element MaterialThermal

ConductivityCore N97 4 W/(m K)Winding isolation Poly-Pad K10 0.85 W/(m K)Trafo cold plate AlN > 150 W/(m K)Switch cold plate AlN 20..30 W/(m K)Potting Wepesil VU 4675 1.2 W/(m K)

Parameter ValueMaximum frequency 35 kHzLeakage inductance 26.5 µHPrimary turns 20Secondary turns 20Primary foil winding thickness 100 µmSecondary foil winding thickness 100 µmPrimary winding losses 73 WSecondary winding losses 150 WCore losses N97 37 W

Fig. 6. Cut view of the fixation point between bottomcooling part and secondary side cooling structure

important parameters used for thermal modeling.2.1 Loss Modeling In this section, the models used

for modeling the transformer core and winding losses aresummarized.2.1.1 Winding Losses Since foil windings are used

for the design of the transformer, the method outlined in (24) isused for calculating the skin and proximity effect losses. Themethod is based on an 1D approximation of the H-field inthe windings, which results in the resistance factor expressedin the term of hyperbolic trigonometric functions of the skinpenetration depth.2.1.2 Core Losses For calculating the transformer

core losses the Improved Generalized Steinmetz Equation(iGSE), presented in (25), is used. This procedure takes thederivative of the flux waveform, as well as the peak-to-peakvalue of the flux into account in order to calculate the coreloss. The method offers good precision with low complexity,which is advantageous for optimization problems.

Care must be taken when using manufacturer loss measure-ments for core materials. These measurements are usuallyperformed on small toroidal cores and the losses might besignificantly lower (2-3x) compared to the losses in the coreswith different shapes (e.g. lare E-U-core). It is advisable tocompare the losses of the actual used core, usually given justfor a single frequency and flux point, to the same point on the



Fig. 7. Transformer geometry for calculating leakageinductance

loss curves in the material datasheet, and scale the curves ac-cordingly. The core material which is used during the designprocedure is EPCOS N97 (Table 2).2.2 Isolation The required isolation level translates

into a minimum distance diso between the conductors:

diso,min =Viso

kisoEiso· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · (1)

where Eiso is the dielectric strength of the isolation materialextracted from material manufacturer datasheet, Viso is thevoltage required to be isolated, and kiso ∈ [0, 1] is the safetycoefficient that is highly dependent on the transformer con-struction and the used isolating material. There are very fewsources in literature that discuss selection of this parameter.Additionally, the given formula does not take into account theinhomogeneity of the electric field. Because of these reasons,for the presented design a very conservative value of 0.04 isused for the safety coefficient kiso. For the chosen transformerwinding geometry given in Fig. 7, the parameters dL and diso

must have a greater value than the value of diso,min.2.3 Leakage Inductance Modeling The winding

height in transformer design with relatively high isolation re-quirements is typically smaller than the core window height.In order to calculate the leakage inductance in such design, aformula that employs the Rogowski coefficient can be used (26)

Lσ = μ0N2pπDmeanwL√hcuP · hcuS

kσ · · · · · · · · · · · · · · · · · · · · · · · (2)

where Np is the number of primary turns, Dmean is the meandiameter of the reduced leakage channel, wL is the widthof the reduced leakage channel, hcuP is the primary windingheight, hcuS is the secondary winding height, and kσ is theRogowski coefficient. A simplified illustration of the trans-former winding structure is depicted in Fig. 7. The expres-sions for calculating the required parameters in the leakageinductance formula have been adapted for taking into account

the non-circular shape of the windings

wL =dP + dS

3+ dL · · · · · · · · · · · · · · · · · · · · · · · · · · · · · (3)

kσ ≈ 1 − dP + dS + dL

π√

hcuP · hcuS· · · · · · · · · · · · · · · · · · · · · · · · · (4)

Dmean = D + dP + dL + dS − dS − dP

2dP + dS + 4dL

dP + dS + 3dL

· · · · · · · · · · · · · · · · · · · · (5)

where D is the equivalent internal diameter of the inner wind-ing, and can be calculated as

D = 2

√(bc + dP)(dc + dP)

π· · · · · · · · · · · · · · · · · · · · · · (6)

The given formulas for leakage inductance calculation arevalid for cases where the heights of the primary and sec-ondary winding are approximately the same. Larger differ-ences in winding heights lead to larger deviations in the cal-culation, and the given formulas lose on their accuracy.2.4 Thermal Modeling Two main types of heat

transfer are considered in the thermal model - the conduc-tive heat transfer and the convective heat transfer. Due torelatively small differences between surface body tempera-tures and ambient temperatures, the radiative heat transfer isneglected in the considered model. The expressions for cal-culating conductive heat transfer are very simple and can befound for example in (30). In Fig. 8 the internal channel struc-ture of the integrated cooling system is shown. For modelingthe conductive flow in the channel of the integrated coolingstructure, a basic channel model is shown in Fig. 9. There aretwo thermal resistances that are used for modeling the heattransfer through the channel. The conductive thermal resis-tance Rth,r from the hot surface to the channel wall, and theconvective thermal resistance Rth,d from channel wall to thecooling medium temperature. These thermal resistances canbe calculated using

Rth,r =awlλHS

· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · (7)

Rth,d =1

hlπde· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · (8)

where w is the width of the channel, a is the distance fromhot surface to the center of the channel, l is the length ofthe channel, de =

√A is the equivalent channel diameter

that takes into account channels of non-circular shape, A isthe cross-sectional area of the channel, and λHS is the ther-mal conductivity of heat sink material. For the circular shapechannels used in the presented design equivalent diameter de

is equal to the actual channel diameter d. Parameter h is theheat transfer coefficient which can be calculated with

h =Nuλfluid

de· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · (9)

For calculating the heat transfer coefficient, it is necessary toknow the Nusselt number (Nu). The Nusselt number is, ingeneral, a function of the average ducted fluid velocity, ductgeometry, and the fluids Prandtl number (Pr). The authorof (27) has derived an analytical model for the generalized Nus-selt number (Nu

√A) that is suitable for the extruded channel



Fig. 8. Channel structure of the integrated cooling sys-tem

Fig. 9. Basic channel used for modeling the thermal re-sistances Rth,r and Rth,d

model with arbitrary cross-section

Nu√A =

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎣⎛⎜⎜⎜⎜⎜⎜⎜⎜⎝⎧⎪⎪⎨⎪⎪⎩C2C3

(f Re√A

l∗

) 13⎫⎪⎪⎬⎪⎪⎭

5

+

{C1

(f Re√A

8√πεγ

)}5⎞⎟⎟⎟⎟⎟⎠m5

+

(C4

f (Pr)√l∗

)m⎤⎥⎥⎥⎥⎥⎥⎥⎦

1m

· · · · · · · · · · · · · · · · · · (10)

with

f (Pr) =0.564[

1 +(1.664Pr1/6

)9/2]2/9· · · · · · · · · · · · · · · (11)

f Re√A =

[11.8336V

l · ν + ( f Refd)2

]1/2

· · · · · · · · · · · · (12)

f Re f d =12√

ε(1 + ε)[1 − 192

π5 tanh(π2ε

)] · · · · · · · · · · (13)

where ρ is the density of the channel fluid, V is the flow ratein [m3/s], ν is the kinematic viscosity of the channel fluid,l∗ = l · ν/VPr is the dimensionless thermal duct length, ε isthe nominal aspect ratio that is defined as a ratio between ver-tical and horizontal dimension of the channel with arbitraryshape, and the parameters C1,C2,C3,C4, γ are defined in (27)

Fig. 10. Thermal network model of the transformer withintegrated cooling. The model results in a 8x8 matrix sys-tem, that is solved in order to obtain the required nodaltemperatures

for uniform wall temperature problems as

C1 = 3.01 C2 =32

C3 = 0.409 C4 = 2 γ =110

The blending parameter m is defined by

m = 2.27 + 1.65Pr13 · · · · · · · · · · · · · · · · · · · · · · · · · · · · (14)

With the presented models all thermal resistances in the sys-tem can be numerically evaluated. As a result, the ther-mal network of the presented transformer can be derived andused in the optimization procedure given in Fig. 3. For easierrepresentation of the thermal network, the transformer fromFig. 5 is split in two 2D cut planes, each showing a cut viewof the transformer with cooling channels as shown in Fig. 10.

In order to solve the thermal model of the transformer, ad-ditional information on the fluid is required. Due to the mul-tiple parallel pipe structure shown in Fig. 11, the fluid flowin the different channels differs from the input fluid flow. Inorder to calculate the velocity of the fluid in the parallel chan-nels, the following expressions are used (28):

Δptot = Δp1 = Δp2 = ... = Δpi · · · · · · · · · · · · · · · · · (15)

V = V1 + V2 + ... + Vi · · · · · · · · · · · · · · · · · · · · · · · · · · (16)

where Δptot is the total pressure loss of the system, and Δpi



Fig. 11. Flow and pressure losses inside presented inte-grated cooling structure. Top side cooling channel is notshown

(i = 1, 2, ...) are the pressure losses in the individual chan-nels. V and Vi (i = 1, 2, ...) designate the input flow rate andflow rate in the individual channels, respectively. For cal-culating the pressure losses in the individual channels, theDarcy-Weisbach relation can be used (28):

Δp = f · ρ2· l

d· V2 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · (17)

where V is the fluid velocity, l is the channel length, d is thechannel diameter, f is the channel friction factor, and ρ isthe fluid density. Each channel has a quadratic parallel re-sistance, and the pressure loss is related to the total flow rateby

Δp =V2

(∑ √Ki/ fi

)2· · · · · · · · · · · · · · · · · · · · · · · · · · · · · (18)

where

V = V · πd2i

4Ki =

π2 · d5i

8ρli

In the general case, the channel friction factor fi is a func-tion of the Reynolds number and the roughness ratio. Sincethe Reynolds number varies with the fluid velocity, solvingof the set of equations must be done iteratively. In the firststep, an arbitrary values of fi are chosen and with them a firstestimate of hf is calculated. Then, the resulting flow rate es-timate Vi ≈ (KiΔp/ fi)1/2 is obtained for each channel. Usingthese results, a new Reynolds number and a better estimate offi is calculated. Usually, a few iteration steps are sufficient toobtain a satisfactory solution. When the flow in the channel islaminar, a simple expression for f , known as Darcy frictionfactor, can be used

f =64Re· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · (19)

where Re = V · d/ν is the Reynolds number. Finally, the ob-tained fluid flow rate is then used in expression Eq. (11) inorder to calculate the thermal resistance of the water channel(Eqs. (7) and (8). For the channels which have a turbulentflow, Haaland’s equation offers good approximation of theturbulent region of the Moody chart (28)

f =

⎡⎢⎢⎢⎢⎢⎣−1.8 log

⎛⎜⎜⎜⎜⎜⎝6.9Re+

(ζ/d3.7

)1.11⎞⎟⎟⎟⎟⎟⎠⎤⎥⎥⎥⎥⎥⎦−2

· · · · · · · · · · · · · · (20)

Table 3. Comparison between the values of the leakageinductance obtained with the analytical calculation andthe 3D FEM simulation

Calculation Analytical FEMLeakage inductance Lσ 26.5 µH 26.2 µH

where ζ/d is the channel roughness ratio and ζ is the wallroughness height.

For designs with long channels and or slow fluid flows, thetemperature of the fluid can increase along the axial direc-tion. In order to calculate this temperature rise of the fluid,an 1D energy balance expression (29) is used:

q′ = VρCp(T (x) − Tin)

x· · · · · · · · · · · · · · · · · · · · · · · · · (21)

where q′ = q/l are the power losses per unit of length, q isthe total power losses dissipated along the channel of lengthl, ρ is the density of the fluid, Cp is the thermal capacity offluid, and Tin, T (x) are the input temperature and temperatureat point x in axial direction, respectively.

3. FEM Simulations

In this section, results of FEM simulations are presentedfor different design aspects.3.1 Leakage Inductance Simulation For verifying

the analytical model of the leakage inductance from sec-tion 2.3, a 3D magnetic field simulation is performed. For thesimulation, an 1 A current excitation of the primary and sec-ondary windings are assumed with opposite directions, andthe resulting total magnetic energy is obtained. The leakageinductance is then calculated from the energy with

Lσ =2Etot

I2· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · (22)

The comparison of the calculated and simulated leakage in-ductance value is shown in Table 3.3.2 Heat Transfer and CFD Simulations In order to

verify the thermal models presented in section 2.4, combined3D heat transfer and fluid dynamic FEM simulations wereperformed. The simulated velocity field inside the presentedcooling structure is shown in Fig. 12, while the temperaturedistribution is given in Fig. 13. The input flow rate at the in-let is assumed to be 8 L/min and the inlet water temperatureis equal to 20 ◦C, which corresponds to the flow rate and wa-ter temperature used during experimental measurement. InTable 4, a comparison between the analytical calculation andthe FEM simulation for the flow velocity distribution insideindividual channels of the parallel multi-channel structure isgiven. The values of the material parameters considered inthe analytical thermal calculations listed in Table 2 are alsoused for the 3D FEM simulations. The comparison betweenthe calculated temperatures of the transformer obtained withthe thermal model given in Fig. 10 and the results from FEMsimulations are given in Table 5.3.3 Eddy Current Simulations In order to get the

eddy current induced losses in the transformer cooling struc-ture 3D FEM simulations were performed. There, two caseshave been considered: 1) single Al bar located between themiddle transformer leg and the primary winding on the bot-tom, 2) two AlN bars located on top and bottom of the mid-dle transformer leg. The surface current densities induced



Fig. 12. Water velocity field inside the presented con-verter cooling structure

Fig. 13. Temperature distribution of the integrated con-verter structure

Table 4. Calculated velocity distribution in internalchannels of the integrated cooling structure from Fig. 11

Velocity Analytical FEMV1 0.45 m/s 0.47 m/sV2 0.34 m/s 0.35 m/sV3 0.26 m/s 0.34 m/sV4 0.21 m/s 0.29 m/sV5 0.092 m/s 0.086 m/s

Fig. 14. Eddy currents induced in the bottom coolingstructure with Al bar

in the bottom cooling parts in case of the Al and AlN barsare given in Figs. 14 and 15. The peak current density withthe AlN bars is 3 times lower. The total losses induced in thetransformer cooling structure for the two mentioned cases aregiven in Table 6.3.4 Calculation Speed For comparing the calcula-

tion between FEM simulations and analytical calculations, aserver computer with an Intel Xeon E5-2697A processor with32 cores at 2.6 GHz and 512 GB of RAM memory is used. InTable 7 the computation times are given for different modeledparts.

Fig. 15. Eddy currents induced in the bottom coolingstructure with AlN bar

Table 5. Calculated temperatures of the transformerwith presented thermal model (Fig. 10)

Temperature Analytical FEMCore mid leg (TC1) 72 ◦C 70 ◦CCore outer mid leg (TC2) 42 ◦C 40 ◦CCore side leg (TC3) 28 ◦C 29 ◦CAlN inner (TAlN) 68 ◦C 71 ◦CPrimary winding (TWp) 83 ◦C 78 ◦CPotting material (TPot) 80 ◦C 77 ◦CSecondary winding (TWs) 88 ◦C 81 ◦CAl top/bottom cover (TTop) 66 ◦C 43 ◦C

Table 6. Induced eddy current losses in the transformercooling structure

Calculation Al bar AlN barInduced losses 75 W 36 W

Table 7. Comparison of computation times for analyti-cal calculations and FEM simulations

Calculation Analytical FEMLeakage inductance 0.01 s 450 sHeat transfer & CFD 0.12 s 1380 sEddy currents - 320 s

4. Experimental Results

In order to experimentally validate the presented mod-els, a 50 kW module employing SiC MOSFETs as switch-ing devices and medium-frequency transformer was built. InFig. 16 the photo of a single module for a modular DC-DCconverter system (Fig. 1) is given.4.1 Leakage Inductance Measurement The leak-

age inductance is measured with a power choke tester (ED-KDPG10-1500B), that is a pulsed inductance measurement de-vice. The measurement was performed on three built trans-formers and the results are shown in Fig. 17. Comparing themeasurement results with the results of the analytical calcu-lation and FEM simulation, given in Table 3, it can be seenthat these match very well.4.2 Thermal Measurement In order to measure the

temperature of the transformer winding, an NTC temperatureprobe is attached to the secondary winding. The converter isoperated in open loop with nominal voltages of 700 V andcurrent of 60 A. The water cooling system was operatedwith constant flow rate of 8 L/min and a water temperatureof 20 ◦C. The converter was operated until reaching steady



Fig. 16. Photo of the 50 kW DAB converter for the con-sidered modular DC-DC system shown in Fig. 1

Fig. 17. Measured short circuit inductance of the builttransformers

Fig. 18. Steady state converter temperatures measuredwith a thermal camera

state temperatures, i.e. approximately 90 min. The picturefrom the thermal camera at the end of the test is shown inFig. 18. Table 8 gives the comparison between the measuredand calculated temperatures.4.3 Partial Discharge Measurement The final mea-

surement which has been performed is the partial dis-charge measurement. This measurement is performed bya 2.8 kV/50 Hz peak voltage to the secondary side windingwhile having the primary side winding and core grounded.The results of the partial discharge measurement procedureare depicted in Fig. 19. The figure shows the cumulative par-tial discharges during a testing period of 20 min, and theiroccurrence as a function of the supplied voltage. As canbe seen, the highest discharge values are lower than 18 pC,which can, for the given requirements, be regarded as partial

Table 8. Comparison of measured and calculated tem-peratures of the transformer

Temperature Analytical FEM MeasuredCore mid leg (TC1) 72 ◦C 70 ◦C -

Core outer mid leg (TC2) 42 ◦C 40 ◦C 63 ◦CCore side leg (TC3) 28 ◦C 29 ◦C 27 ◦CAlN inner (TAlN) 68 ◦C 71 ◦C -

Primary winding (TWp) 83 ◦C 78 ◦C -

Potting material (TPot) 80 ◦C 77 ◦C -

Secondary winding (TWs) 88 ◦C 81 ◦C 83 ◦CAl top/bottom cover (TTop) 66 ◦C 43 ◦C 66 ◦C

Fig. 19. Results of the partial discharge measurementperformed on the designed transformer

Fig. 20. Voltage and current waveforms of the trans-former during startup phase in SAB mode

discharge free.4.4 Transformer Circuit Operation The designed

transformer has been tested in single active bridge (SAB)mode during startup phase (Fig. 20) at nominal voltage, andin DAB mode during normal operation (Fig. 21) at 700 V and60 A, that corresponds to overload of approximately 10 %.The controller design is presented in (22).

5. Conclusion

In this paper, the design of a medium frequency trans-former with integrated cooling structure is presented. Thesystem integration with advanced cooling design results in ahigher power density value (Fig. 2). Detailed loss, leakage in-ductance and thermal models of the medium frequency trans-former with integrated cooling structure are used to find theoptimal design parameters using the methodology presentedin section 2. The resulting design was validated with exten-sive FEM simulations for various design aspects. The finalvalidation is performed with experimental measurements on



Fig. 21. Voltage and current waveforms of the trans-former during DAB mode of operation

the prototype system.AcknowledgmentThis research is part of the activities of the Swiss Centre

for Competence in Energy Research on Efficient Technolo-gies and Systems for Mobility (SCCER Mobility), which isfinancially supported by the Swiss Innovation Agency (In-nosuisse - SCCER program) and Bombardier TransportationAG Switzerland. CTI funding grant no.: PFIW-IW 18312.1

References

( 1 ) “Uniflex-PM” (2009) [Online]. Available: http://www.eee.nott.ac.uk/uniflex/Documents/Deliverable

( 2 ) G. Reed, G. Kusic, J. Svensson, and Z. Wang: “A case for medium voltagedirect current (MVDC) power for distribution applications”, in IEEE-PESPower Systems Conference and Exposition (2011)

( 3 ) F. Mura and R.W.De Doncker: “Design aspects of a medium-voltage directcurrent (MVDC) grid for a university campus”, in IEEE 8th InternationalConference on Power Electronics and ECCE Asia (2011)

( 4 ) Y. Matsuoka, K. Takao, K. Wada, M. Nakahara, K. Sung, H. Ohashi, andS. Nishizawa: “2.5 kV, 200 kW bi-directional isolated DC/DC converter formedium-voltage applications”, in International Power Electronics Confer-ence (2014)

( 5 ) M. Steiner and H. Reinold: “Medium frequency topology in railway applica-tions”, in European Conf. on Power Electronics and Applications (2007)

( 6 ) J. Weigel, A. Ag, and H. Hoffmann: “High voltage IGBTs in medium fre-quency traction power supply”, in European Conference on Power Electron-ics and Applications (2009)

( 7 ) C. Zhao, S. Lewdeni-Schmid, J. Steinke, M. Weiss, T. Chaudhuri, M.Pellerin, J. Duron, and P. Stefanutti: “Design, implementation and perfor-mance of a modular power electronic transformer (PET) for railway applica-tion”, in European Conference on Power Electronics and Applications (2011)

( 8 ) G. Ortiz: “High-power DC-DC converter technologies for smart grid andtraction applications”, Ph.D. dissertation, ETH Zurich (2014)

( 9 ) C. Chryssakis and B.J. Vartdal: “Ship electrification and alternative fuels”,Motorways of the Seas (2016)

(10) A. Burke: “Batteries and ultracapacitors for electric, hybrid and fuel cell ve-hicles”, in Proceedings of the IEEE (2007)

(11) S. Vasquez, S. Lukic, E. Galvan, L. Franquelo, and J. Carrasco: “Energy stor-age systems for transport and grid applications”, IEEE Trans. on IndustrialElectronics (2010)

(12) R. De Doncker, D. Divan, and M. Kheraluwala: “A three-phase soft-switchedhigh power density DC-DC converter for high power applications”, in IEEEIndustry Applications Society Annual Meeting (1988)

(13) J. Biela, U. Badstubner, and J. Kolar: “Design of a 5 kW, 1-U, 10 kW/ltr.resonant DC-DC converter for telecom applications”, in 29th InternationalTelecommunications Energy Conference, INTELEC (2007)

(14) K.W. Klontz, D.M. Divan, and D.W. Novotny: “An actively cooled 120 kW

coaxial winding transformer for fast charging electric vehicles”, IEEE Trans.on Industry Applications (1995)

(15) L. Heinemann: “An actively cooled high power, high frequency transformerwith high insulation capability”, in IEEE Applied Power Electronics Confer-ence and Exposition (2002)

(16) H. Hoffmann and B. Piepenbreier: “Medium frequency transformer for railapplication using new materials”, in 1st International Electric Drives Pro-duction Conference (2011)

(17) I. Villar: “Multiphysical characterization of medium-frequency power elec-tronic transformers”, Ph.D. dissertation, EPFL, Lausanne (2010)

(18) P. Shuai and J. Biela: “Design and optimization of medium frequency,medium voltage transformers”, in 15th European Conference on Power Elec-tronics and Applications (2013)

(19) M. Mogorovic and D. Dujic: “Thermal modeling and experimental verifi-cation of an air cooled medium frequency transformer”, in 19th EuropeanConference on Power Electronics and Applications (EPE) (2017)

(20) M. Pavlovsky: “Electronic dc transformer with high power density”, Ph.D.dissertation, Technical University Delft (2006)

(21) M.A. Bahmani: “Design and optimization considerations of medium-frequency power transformers in high-power DC-DC applications”, Ph.D.dissertation, Chalmers University of Technology (2016)

(22) M. Stojadinovic, E. Kalkounis, F. Jauch, and J. Biela: “Generalized PWMgenerator with transformer flux balancing for dual active bridge converter”, inProc. 19th European Conf. Power Electronics and Applications (EPE) (2017)

(23) M.H. Kheraluwala, D. Novotny, and D.M. Divan: “Coaxially wound trans-formers for high-power high-frequency applications”, IEEE Transactions onPower Electronics (1992)

(24) E. Bennett and S.C. Larson: “Effective resistance to alternating currents ofmultilayer windings”, Transactions of the American Institute of ElectricalEngineers (1940)

(25) K. Venkatachalam, C. Sullivan, T. Abdallah, and H. Tacca: “Accurate predic-tion of ferrite core loss with nonsinusoidal waveforms using only steinmetzparameters”, in IEEE Workshop on Computers in Power Electronics (2002)

(26) V.V. Kantor: “Methods of calculating leakage inductance of transformerwindings”, Russian Electrical Engineering (2009)

(27) Y.S. Muzychka: “Generalized models for laminar developing flows in heatsinks and heat exchangers”, Heat Transfer Engineering (2013)

(28) F.M. White: Fluid Mechanics: McGraw-Hill (1998)(29) S. Ghiaasiaan: Convective Heat and Mass Transfer. Cambridge University

Press (2011)(30) J.H. Lienhard IV, J.H. Lienhard V: “A Heat Transfer Textbook”, Cambridge

MA (2000)

Milos Stojadinovic (Non-member) studied electrical engineering atthe University of Belgrade, Belgrade, Serbia, focus-ing on electrical drives and converters. He receivedhis M.Sc. degree in 2011. His master thesis dealt withmeans of compensating non-idealities in three-phaseinverters. In January 2013, he joined the High PowerElectronics Laboratory, ETH Zurich, where he is cur-rently working toward the PhD degree. His mainresearch focus is on multi-domain modeling, opti-mization and design of medium-frequency medium-

voltage DC-DC converters for traction applications.

Jurgen Biela (Non-member) received the Diploma (Hons.) de-gree from FriedrichAlexander Universitat Erlangen-Nurnberg, Nuremberg, Germany, in 1999, and thePh.D. degree from the Swiss Federal Institute ofTechnology (ETH) Zurich, Switzerland, in 2006. In2000, he joined the research department, SiemensA&D, Erlangen, Germany, and in 2002, he joined thePower Electronic Systems Laboratory, ETH Zurich,as Ph.D. student focusing on electromagnetically in-tegrated resonant converters. From 2006 to 2010, he

was a Postdoctoral Fellow with the Power Electronic Systems Laboratory.His current research interests include the design, modeling, and optimisa-tion of PFC, dc-dc, and multilevel converters with an emphasis on passivecomponents, the design of pulsed-power systems, and power electronic sys-tems for future energy distribution.
