Inductor Design Methodology for Power Electronics

Applications
P. H. J. Vilkn, L. M. F. Morais, R. A. S. Santana, P. C. Cortizo, P. F. Seixas
Graduate Program inElectrical EngineeringUniversidade Federal de Minas Gerais
Av. Antônio Carlos 6627, 31270-901, Belo Horizonte, MG, Brazil
[email protected],[email protected], [email protected],
[email protected], [email protected]

Abstract—In this work, a methodology is presented

that allows the automatization of the inductor design,
estimating copper losses, considering skin and prox-
imity effects and also core losses through improved
generalized Steinmetz equation. The inductors de-
signed with this methodology were tested in a Buck
converter, comparing the calculation of losses with
measurements through a calorimeter.
Index Terms—Inductor Design, Core Losses, IGSE,
Calorimeter, Loss measurement

I. Introduction
With the increment of the switching frequency, made
possible by the development of the wide band gap semi-
conductors, the balance between power density, eficiency
and cost of static converters was severely altered due to
the reduction of volume and change in the loss profile of
the devices.
The amount of losses represented by the inductances
becomes more significant due to the reduction of the
losses in the wide band gap semiconductors and also to
the increase of the losses by skin and proximity effects in
the inductor.
This work presents a methodology for the design of
inductors used in switched static converters that allows
the obtaining of projects with losses, volume or minimum
costs.
The proposed methodology performs projects for the
combination of provided parameters for the cores, fre-
quency, current density and inductance, identifying the
possible projects and comparing the results to obtain the
most suitable inductor.
The inductor design is independent of the type of con-
verter used, as long as it is possible to obtain the voltage
and current vectors at the terminals of the inductor and
to identify the harmonic components of the current, it is
then possible to perform all calculations present in the Fig. 1. Flowchart of the inductor design methodology
design proposed.
Section II explains the design of the inductor, section
III describes the system created for loss measurement the generation of a large number of projects, verifying
and section IV describes the experiments performed and the possibility of realization and then identifying the best
presents the results obtained. results.
A design attempt is made for each combination of
II. Inductor project in non sinusoidal inductance, frequency, current density and core inserted
conditions in the input vectors.
The flowchart in Fig. 1 demonstrates the project Boundaries for magnetic flux density B, temperature
methodology proposed in this work, which is based on T , core window fill and inductance tolerance are set to
This work has been supported by the Brazilian agency CAPES ensure that the design is achievable and meets minimum
and Engetron Engenharia Eletrônica Ind e Com LTDA. operating limits.
 When one of the boundaries is not respected in a
project attempt, it is then discarded and the next attempt
is initiated until all combinations are performed.
The core data is loaded from a library that was created
to standardize information of the cores from different
manufacturers.
At the end of the calculation of all combinations the
saved projects are sorted according to the priority of
parameters (volume, losses, temperature, cost) defined by
the user.
A. calculation of the number of turns
The calculation of the number of turns, modified from Fig. 2. Saturated inductance model for simulation
the proposed one in [1] and [2], is an iterative process
since the magnetic permeability µ is affected by the
frequency f , temperature T , density of flux BAC and The temperature value can be estimated using basic
magnetic field HDC , which depends on the number of loss calculations with (17), (27) and (31), but in general
turns used. this effect is not significant except in projects specific for
These relationships between frequency, temperature, high temperatures.
flux density a.c., magnetic field d.c. and magnetic perme- One can then use the relations µ × HDC , µ × BAC ,
ability are given in curves or equations by manufacturers, µ × T , and µ × f to obtain µ% which is the reduction of
however it is unusual to find all curves available. the initial permeability given by (5).
The loop of each iteration starts with (1) obtaining the
µinitial − µef f ective
new number of turns, where AL is the specific inductance µ% = (5)
given by the manufacturer or obtained from the reluc- µinitial
tance of the magnetic path R per (2), with Ac the cross- With this we calculate the effective AL,ef and Lef in
sectional area and le the average length of the magnetic (6) and (7) respectively.
path.
r AL,ef = AL µ% (6)
L
N= (1)
AL
Lef = N 2 AL,ef (7)
1 µe Ac If Lef is not within the tolerance limits we return to
AL = = (2)
R le (1) using AL = AL,ef , until Lef is within tolerance or the
Using air gap adds a term in the reluctance that is maximum number of iterations is reached.
independent of the permeability of the material, which Finding the number of turns the vector obtained in (4)
facilitates the definition of the number of turns but gen- is checked for saturation limits. Alternatively a better
erates greater emission of electromagnetic radiation and result is obtained by simulating the converter using a
additional terms of losses. These effects are intensified saturable inductance model which includes the B × H
in applications of higher power and frequency, and are curve of the material as in Fig. 2, best explained in [3].
generally avoided in switched sources [1].
B. Winding design
While f is known and can be used directly HDC , BAC
and T require additional calculations. In the design of the winding the wires used and the
To obtain HDC is used (3), where IDC in a switched way they are coiled are defined, obtaining the filling of
inverter context is the peak current of the low frequency the window and the resulting number of layers, as well as
sinusoidal component, in this case is calculated HDC the length and d.c. resistance of the conductor. For this
maximum and its instantaneous value will vary, as will it is necessary to calculate the conductor cross-sectional
the final inductance. area Acopper , obtained by (8).

N IDC Ief
HDC = (3) Acopper = (8)
le j
The conductor is a set of Nwires parallel insuled con-
It is possible to obtain BAC using (3) for the maximum
ductors of diameter dwire twisted to reduce skin and
and minimum values of the ripple current and using the
proximity effect. Using dwire ≤ 2δ being δ the depth
B × H curve of the material when available.
of penetration, wepreduce the skin effect considerably.
However, an independent method of the B × H curve
On copper
√ δ = ρ/(πµf ) can be approximated by
requires only the voltage vector in the inductor VL (t), N
δ = 65/ f at a temperature of 25o C [1].
and Ac , obtaining the vector B(t) from (4) [3].
The equations (9) and (10) relate the diameter of the
wires and the number of parallel wires required. High
R
VL (t) dt
B(t) = (4) frequencies require a large number of very thin wires in
N Ac
parallel, which leads to problems in the winding construc- C. Copper losses estimation
tion. In these cases the use of Litz wires is recommended
The power dissipated in the winding due to the low
[4].
frequencies can be calculated using (17).
r
Acopper
dwire = 2 (9) 2
Nwires π P = Rwire Ief (17)
( )
Acopper Pelicular and proximity effects can be calculated by
Nwires = N
2 (10) obtaining the effective conductor cross-sectional area for
π dwire
2 each frequency component of the current signal.
The fill of the core window area AW is calculated by Considering only the film effect, one can calculate the
(11). If the fill is above the operational boundary defined penetration depth δk by (18), where k is the index of the
the project is discarded. A reference for the filling value harmonic component.
expected for each core format can be found in [1].
65
N Acopper δk = √ (18)
f illing = (11) kf
Aw
To estimate the number of layers NC it is necessary The equations of (19) to (22) [5] are approximations to
topalso estimate the diameter of the loop by De = obtain the effective area for each δ.
2 Acopper /π and consider a coil spacing de which will
dwire
vary with the technique used to braid the wires and p0k = δk (1 − e
− 2 δk
) (19)
to make the winding, being an estimated value between
approximately 0.5mm and 3mm.
Considering the spacing between adjacent turns and dwire
between the turns of the first layer and the core equal zk = 0.62006 (20)
2 δk
to de , for toroidal cores can be calculated the internal
diameter of each layer with (12), where ID is the inner
core diameter and kc is the layer number. 0.189774
yk = (21)
(1 + 0.272481 (zk1.82938 − zk−0.99457 )2 )1.0941
Dc (kc ) = ID − (2 (kc − 1) + 1) De − 2 kc de (12)
The inner perimeter of each layer is calculated directly Aef,k = π (dwire p0k − (p0k )2 ) (1 + yk ) (22)
by Pc (k) = πDc (kc ) and the number of wires in the layer
is obtained in (13). The winding resistance is then calculated for each
  harmonic component with (23).
Pc (k)
nwires (k) = N (13)
De + de Cwire ρ
Rk = (23)
The number of layers NC is found when satisfying (14) nwires Aef,k
by comparing the number of turns with the number of
Alternatively the skin and proximity effects can be
wires accumulated in each layer
calculated collectively by using (24) and (25) [6].
C −1
NX NC
X 3
nwires (kc ) ≤ N < nwires (kc ) (14)  π  43 d2
kc =0 kc =0 Ak = √ wire (24)
4 δk dwire + de
This process is simplified in non-toroidal cores, like E
and U cores, since the diameter is replaced by the length
of the winding region which, like the number of wires per 
e2Ak − e−2Ak + 2sen(2Ak )
layer, is constant. Rk = Rwire Ak + ...
e2Ak + e−2Ak − 2cos(2Ak )
The length of each wire Cwire is calculated in (15)
2 NC2 − 1 e2Ak − e−2Ak − 2sen(2Ak )

using the mean length per turn M LT generally given by ... (25)
the core manufacturers, but can also be found by simple 3 e2Ak + e−2Ak + 2cos(2Ak )
geometrical calculations.
With the final winding resistance Rk for each fre-
quency, found in (25), the power is obtained by summing
Cwire = M LT ∗ N (15)
the product of each harmonic component of current Ik
The DC resistance of the wire is calculated from (16), squared by the corresponding resistance Rk , as in (26):
where ρ is the resistivity of the conductor used.
k
ρ
X
Rwire = Cwire (16) P = Rk Ik2 (26)
nwire π (dwire /2)2 0
D. Core loss estimation
The equation (27) presents the Steinmetz, the basic loss
estimation formula for the core. The Steinmetz equation
which is a modified form of the one proposed originally
in [7], whose parameters are always supplied by the
manufacturers of magnetic materials, is a function of the
frequency f and the amplitude of the oscillation of the
magnetic flux density Bpk .

β
P = K f α Bpk (27)
This equation is valid only for sinusoidal conditions.
Different methods were created to obtain equations for
non-sinusoidal estimations using the same parameters, K,
α and β of the Steinmetz equation.
The improved generalized steinmetz equation (IGSE)
developed in [8], an improved form of the GSE method
[9], enables losses estimation using the parameters of Fig. 3. Front view of the calorimeter built
the Steinmetz equation and the density vector of time
flow. Considering smaller loops within the hysteresis loop,
III. Loss measurement System
identifying, separating and arranging the magnetic flow
vector as explained in [10]. Calculating the volumetric A calorimeter developed for the purpose of loss mea-
density of energy dissipated between each point of vector surement in switched inverters and their components,
B(t) for each of the smaller and larger cycles separately shown in Fig. 3, was used to obtain the loss results of the
by (28) with Bpk being the amplitude of the loop in next section. Figure 4 show the supervisory diagram used
question. with the calorimeter. The functioning of the calorimeter
is explained in detail in [11].
 dB α β−α
 
The calorimeter has a topology similar to that devel-
E = Ki   B ∆t (28)
dt  pk oped in [11]. It has two thermally insulated chambers, an
internal chamber where the test object is placed. And an
Where the K constant must be changed according to
external chamber that has its temperature controlled by
(29) to get Ki :
resistive loads to be equal to that of the inner chamber,
K reducing the flow of heat through the wall of the inner
Ki = R 2π (29) chamber. Ventilation systems are present in both cham-
(2π)α+1 2β−α |cos(θ)|α dθ
0 bers to homogenize the temperature.
The power is then calculated by the total energy The main temperature change of the internal chamber
dissipated in a period T : is made by a closed thermal system of water, being possi-
ble to reach an equilibrium where all internally generated
T
Vc X power is transferred to water and the internal chamber
P = E(t) (30)
T t=0

The use of (28) in place of the Steinmetz equation

makes the estimation less dependent on the switching
frequency, but since faithful representation of the flux
density is required for at least an entire fundamental
cycle, it is important to note that with increasing of the
switching frequency the calculation step must be decre-
saed to obtain a good representation of the instantaneous
flux density, which progressively increases the time of
simulation and size of the vectors.

E. Temperature rise estimation

The expected temperature rise ∆T can be calculated
using (31), obtaining the inductor temperature by T =
Tamb + ∆T [1] [2]. Asurf ace refers to the surface area
of the inductor including the coil, wich is usualy an
aproximation.
 0.826
Ptotal
∆T = 450 (31) Fig. 4. LabView supervisory diagram of the calorimeter
Asurf ace
temperature is constant. The thermal system exchanges TABLE I
heat with the inner chamber and the environment by Inductor design data
radiators and also with a commercial refrigeration system Inductor A B C D
for conventional refrigerators. The water flow is regulated Core T520 T400 58090 78090
by a hydraulic pump controlled by PWM. Manufacturer Micrometals Magnetics
Material 34D 14D High Xflux
At the inlet and outlet of water from the inner chamber Flux
are measured the inlet temperature Tin and outlet tem- Lproj 1,06 0,230 0,155 0,097 mH
perature Tout of the water and the flow V̇ . The equation Linitial 1,077 0,236 0,162 0,098 mH
Lf inal 1,038 0,231 0,141 0,096 mH
(32) calculates the power, being the first term referring Lleak 0,128 0,065 0,021 0,014 mH
to the one transferred to the water and the second term Ltotal 1,165 0,296 0,163 0,110 mH
being the sum of the power that passes through the N 102 80 49 34
insulating walls. Where cp is the thermal capacity of the winding 24,81 28,82 40,16 27,87 %
NC 2 2 3 2
water, ρH2 O the volumetric density of the water, Tint and Pcopper,dc 14,70 12,00 6,872 4,440 W
Text the temperatures on the inner and outer chambers Pcopper,ac 0,0281 0,462 0,292 0,484 W
and Rth is the thermal resistance of the insulation wall. Pcopper 14,72 12,46 7,164 4,923 W
Bpk 0,0228 0,058 0,372 0,536 T
Pcore 0,586 1,961 2,659 18,46 W
X Tint − Text Ptotal 15,31 14,43 9.823 23,38 W
P = cp ρH2 O V̇ (Tout − Tin ) + (32) ∆T 17,10 24,11 67,31 137,1 C
Rth Rdc 23,5 19,2 11,0 7,1 mΩ
The temperature and flow sensors were individually Rac 216,1 157,3 44,9 29,0 mΩ
Volume 1336 637 108 88 cm3
calibrated and the power measure was also calibrated
to remove the error caused by the internal ventilation
system of the chambers required to homogenize the tem- TABLE II
Converter parameters
perature. but slow oscillations in the power dissipated by
the ventilation system will cause an error in the power VDC 150 V
measurement of the device under test. Imed 25 A
f 15360 Hz
d 0.5
IV. Results Vload 75 V
In order to evaluate the proposed methodology, four
inductors, identified in table I by A, B, C and D, were de-
signed using different materials and different inductance and the values projected for inductance, Lproject and its
values to be used in a buck converter, shown in Fig. 5. expected final value in the project, Lend .
The converter parameters are present in the table II Looking at the table IV, which shows the inductance
and the design data of the designed inductors in the table error Lest in relation to Lproj , we realize that the pro-
I. jected value for the inductance lies within the expected
error limits , considering the tolerance of 8% in the
A. Inductance measurement permeability µ of the materials.
Using a constant duty cycle d = 0.5, one can estimate B. Loss Masurement
the inductance, when in saturation, by the slope of the
current in the inductor ∆I by (33). The measurement of losses by the calorimeter was
carried out to obtain the total losses Ptotal , the loss in
Vdc the copper Pcopper approximated by the d.c. was obtained
L= (33) by measuring the average voltage and current in the
2 ∆I f
inductor. For a switching frequency of 15360Hz, the
The table III shows the inductance values obtained expected high-frequency effects are significantly smaller
by the measurement through an RLC bridge LRLC , by than the measurement errors encountered. And the loss
estimating (33) through the voltage and current measure- in the core Pcore was estimated by the difference between
ment Lest , by the ideal inductance value Lideal = AL∗N 2 Ptotal and Pcopper . The values of losses can be observed
in the table V.
The calorimeter developed has two error components,
one relative to the measured power of ±1% and another

TABLE III
Comparison of inductance values obtained

Inductor A B C D
Lproj 1,06 0,23 0,155 0,097 mH
Lf inal 1,038 0,231 0,141 0,096 mH
LRLC 1,434 0,269 0,211 0,108 mH
Lest 1,092 0,249 0,153 0,104 mH
Fig. 5. Diagram of the buck converter used to test the inductances Lideal 1,379 0,299 0,214 0,109 mH
 TABLE IV
Inductance error Lest − Lproj

Inductor A B C D
Error L 0,032 0,019 -0,002 0,007 mH
3,02 8,26 -1,29 7,22 %

TABLE V
Measured and calculated losses in the inductors

Inductor A B C D
Pcore,calc 0,56 2,0 2,7 18,5 W
Pcore,est 3,5 2,5 3,7 16,3 W
Error Pcore 2,86 0,51 1,06 -2,14 W Fig. 7. Comparison of losses in the copper
Pcopper,calc 14,7 12,5 7,2 4,9 W
Pcopper,meas 13,3 13,3 10,1 6,5 W
Error Pcopper -1,38 0,86 2,91 1,55 W
Ptotal,calc 15,3 14,4 9,8 23,4 W
Ptotal,meas 16,8 15,8 13,8 22,8 W
Error Ptotal 1,48 1,36 3,97 -0,59 W

independent of the measured device load of absolute

±3W , coming from the calorimeter system itself, making
measurements of small powers inaccurate.
Figures 6 to 8 compares the values: Calculated, wich
are the values obtained using the equations proposed
Fig. 8. Comparison of total losses
in this text; Traditional, wich are the values obtained
using only the Steinmetz equation for Pcore and effective
current for Pcopper ; Measured, wich are the experimental References
values obtained, by calorimeter for Ptotal , by I × V
[1] Colonel Wm. T. McLyman, Transformer and Inductor Design
obtained by osciloscope for Pcopper and by the energy on Handbook, 3rd ed., Marcel Dekker, Inc., 2004.
the B × H loops as in [12] for Pcore ; And the estimated [2] J. R. Zientarski, “Análise modelagem e validação experimental
is the diference between the measured total and core loss de uma metodologia de projeto do indutor em conversores
Boost PFC,” Universidade Federal de Santa Catarina, 2009.
Pcore,est = Ptotal,meas − Pcopper,meas . [3] J. Koscelnik, J. Sedo and B. Dobrucky, “Modeling of resonant
converter with nonlinear inductance,” 2014 International Con-
V. Conclusions ference on Applied Electronics, pp. 153–156 september 2014.
In order to demonstrate the applicability of the pro- [4] S. Hiruma, Y. Otomo and H. Igarashi, “Eddy current analysis
of Litz wire using homogenization-based FEM in conjunction
posed methodology, four different inductors were de- with integral equation,” IEEE Transactions on Magnetics, vol.
signed, achieving adequate inductance values when in 54, pp. 1–4, march 2018.
operation. [5] Knight, David, “Practical continuous functions for the internal
impedance of solid cylindrical conductors.” 2016.
The values of losses estimated by the proposed project [6] M. Bartoli, A. Reatti and M. K. Kazimierczuk, “Modelling
methodology can be confirmed by being consistent with iron-powder inductors at high frequencies,” Proceedings of
the power values obtained by the traditional method- 1994 IEEE Industry Applications Society Annual Meeting, vol.
2. pp. 1225–1232, 1994.
ology. However, the great uncertainty found in the loss [7] C. P. Steinmetz, “On the law of hysteresis,” Transactions of the
measurement system prevents a better analysis of losses. American Institute of Electrical Engineers, pp. 1–64, vol. IX.
A calorimeter better suited to the inductor power levels January 1892.
should be used, or another method to measure losses [8] K. Venkatachalam, C. R. Sullivan, T. Abdallah and H. Tacca,
“Accurate prediction of ferrite core loss with nonsinusoidal
together with the calorimeter should be used to assure waveforms using only Steinmetz parameters,” 2002 IEEE
the accuracy of the proposed losses and temperature rise Workshop on Computers in Power Electronics, 2002. Proceed-
estimates in the methodology. ings. pp. 36–41, June 2002.
[9] Jieli Li, T. Abdallah and C. R. Sullivan, “Improved calculation
of core loss with nonsinusoidal waveforms,” Conference Record
of the 2001 IEEE Industry Applications Conference. 36th IAS
Annual Meeting (Cat. No.01CH37248), vol. 4, pp. 2203–2210,
september 2001.
[10] M. J. Jacoboski, A. de Bastiani Lange and M. L. Heldwein,
“Closed-form solution for core loss calculation in single-phase
bridgeless PFC rectifiers Based on the iGSE method,” IEEE
Transactions on Power Electronics, vol. 33, pp. 4599–4604,
2018.
[11] D. Christen, U. Badstuebner, J. Biela and J. W. Kolar, “Calori-
metric power loss measurement for highly efficient converters,”
The 2010 International Power Electronics Conference - ECCE
ASIA -, pp. 1438–1445, june 2010.
[12] IEC 60404-6, Magnetic materials - Part 6: Methods of measure-
ment of the magnetic properties of magnetically soft metallic
Fig. 6. Comparison of losses in the core and powder materials at frequencies in the range 20 Hz to 200
kHz by the use of ring specimens. [S.l.], 2018.