Anda di halaman 1dari 7

DESIGNING INDUCTORS AT DEVICE AND CONVERTER LEVEL

R.Prieto, O.Garca, J.A.Cobos and J.Uceda


Universidad Politcnica de Madrid (UPM)
Divisin de Ingeniera Electrnica (DIE)
Jos Gutirrez Abascal, 2
28006 Madrid SPAIN
Tel: 34-91-411 75 17 Fax: 34-91-564 59 66
email: rprieto@upmdie.upm.es

them is many times replaced by the construction and


ABSTRACT measurement of the actual sample. However, the
modeling stage is a terrific tool in order to validate the
Commonly, the magnetic component designer focuses
design at device and converter levels. An accurate
the design on the performance of the magnetic
model avoids the construction of the magnetic
component without a detailed analysis of the whole
component until the expected results are obtained. The
circuit. However, a complete design process should
combination of the model with the circuit simulator
cover different stages, from the analytical design to the
provides the whole solution to complete a virtual
circuit simulation using accurate models. This paper
prototyping of the magnetic component based on
presents a design example where it is highlighted the
computer simulation. Obviously, this procedure
importance of designing inductors not just from the
becomes efficient if the designer is able to apply a set of
point of view of the device itself but from the point of
tools easy to use. In other words, if the transition
view of the circuit where the component is working. It is
between design, model and simulation tools is easy, and
shown that the optimum design depends on the working
there is not a need to learn different and complicate
mode of the converter and therefore, the design should
tools in order to generate models valid for circuit
be analyzed applying circuit simulations in order to
simulators.
verify the selected configuration.
This work presents a design case where all these issues
are illustrated. The design example is the inductor of a
I. INTRODUCTION buck converter for an automotive application.
There are many topologies of power converter
Section II presents the different design stages that are
containing inductors (Buck, Boost, Buck-Boost,
proposed to be completed in any magnetic component
Forward, Push-Pull, Half-Bridge, Full-Bridge,...). The
design.
efficiency, size and behavior of the converter depend
drastically on the design of this inductor. The inductor Section III describes the application of the previously
design is commonly oriented to optimize the device mentioned design stages to one specific design case.
using the voltage and current waveforms as design input The difference among the different design stages will be
data. However, the real design goal should be oriented highlighted. The importance of cover all the stages will
to the converter behavior, and not the inductor behavior. be shown with this specific case.
Furthermore, the design of the inductor must be
Finally, a summary of the work will be presented in
carefully selected depending on the working points of
section IV.
the converter, because the optimum design could be
different for each working point.
II. DESIGN STAGES
A complete design process should cover different
stages. The first and common one is the application of The design of a magnetic component is a task that
analytical expressions in order to obtain the optimum should be completed in different stages. However, most
design in terms of losses. This design should be oriented of the magnetic component designers only cover part of
to the worse operating point of the inductor in terms the whole process. A complete design process should
of saturation. The output data of this stage is usually the cover the stages described in the following sections.
core shape and size, the core material, the conductor
size, the number of turns and the gap length. Commonly, 2.1. Analytical design
this is the only stage covers by many designers. The This stage covers the selection of:
generation of a model of the selected design is
something that usually is difficult for many designers, Core shape (EE, POT, RM, EP,), size (RM8,
and the information that is supposed to extract from RM14,) and material (3F3, N87,)
Conductor type (Solid, foil, litz,) difficulties to extract a model become a reason to skip
Number of turns and number of parallel wires this stage. However, for most of the cases, the use of the
Gap length model in the circuit simulator will help the designer to
be able to design at converter level, and not just at
The selection of all those constructive parameters is
device level. If the model is accurate enough, the
usually performed applying analytical expressions using
simulation of the whole circuit with different designs of
as input data several of the following parameters:
magnetic components, will be a very useful tool in order
Switching frequency to show the impact of the magnetic component design
Inductance value or current ripple on the circuit behavior. It is important to highlight that
Voltage and current waveforms the model should be accurate enough. This means that
Maximum flux density it should not be more or less accurate than needed. For
example, if the designer is generating the model of a two
The quality of the design depends on the expressions winding transformer composed by two layers of round
that have been applied for the calculations, especially conductors filling the whole window height, an
the modeling of the losses. There is a trade-off between analytical model based on 1D field distribution
accuracy of the results and time to design. It should be assumption is more than enough. It has no sense to
highlighted that in terms of design purposes it is more apply Finite Element Techniques in order to generate a
important to obtain better qualitative results than model for these simple components because the
quantitative results. In other words, it is more results are approximately the same than the ones that
important to know that one design is better than another can be obtained applying analytical methods, but the
one (for example in term of losses), even if the absolute time to generate the model will be several magnitude
value of the losses for both designs are not extremely orders higher. Figure 1 shows a comparison between a
accurate. Therefore, the design process and the model based on Finite Elements and an analytical one
expressions applied should guarantee that the obtained based on transmission lines for a simple transformer.
results yield to these conclusions. It can be seen that the results are approximately the
There are two main difficulties to get the optimum same.
design for a magnetic component:
Resistance
1. There are many degrees of freedom to be considered
(core size, core shape, core material, conductor type,
conductor dimensions, number of parallel wires, )
2. The wire losses and core losses are geometry and
Reactance

frequency dependent. Therefore, their modeling is a


1D model
complicate task that could make ineffective
(time/results ratio) the design process. Therefore, 2D model
commonly, the designer uses simplified expressions
in order to make more efficient the design process in
terms of time. However, the designer should be
caution about the expressions because the Frequency
simplifications could make useless the design
Figure 1. Comparison of 1D and FEA based models
process. Therefore, a computer aided design is a
good alternative for designing magnetic components.
However, if there are two dimensional effects that could
In the following section it will be highlighted that a
affect on the design, the analytical approaches could
wrong design process could yield to select a wrong
become insufficient. For example, if the magnetic
design.
component contains an air gap, the effect of the fringing
flux around the gap could be a critical design parameter.
2.2. Modeling In this case, the application of Finite Elements
The modeling stage provides a model of the previously Techniques is a very good alternative to take into
designed magnetic component in order to be used for account this effect. Figure 2 shows a comparison
device and system optimization. The combination of the between a model based on Finite Elements and an
model with the circuit simulator is a powerful tool in analytical one based on transmission lines for a gapped
order to optimize the constructive parameters of the component. It can be seen that the differences between
component. Many times, with a wrong criterion, the the analytical approaches, where the fringing flux effect
designer assumes that the model stage does not provide is not considered, and the Finite Element based model is
useful information for the design. Other times, the now very important.
Resistance particular example. The lower the ripple, the higher the
inductance. Therefore, applying the expressions that
2D model command the buck converter behavior, the waveform
applied to the buck inductor presents the specifications
shown in table 2.
Reactance

Average current Current ripple (72.7%) L Duty cycle


25 A 4.29 A 10 H 28.6 %
1D model Table 2. Input variables for the inductor

3.1. Design stage


As mentioned before, the design stage is usually based
Frequency on analytical expressions in order to make it effective
Figure 2. Comparison of 1D and FEA based models from the point of view of the time/accuracy trade-off.
Simple analytical expressions will be applied in this
In the next section, a practical example will be used in example in order to illustrate the results of the design
order to illustrate how the model can be useful in order stage. Alternatively, a CAD tool specially tailored for
to select the proper design. magnetic component designs [1] will also be applied in
this specific example in order to compare these two
2.3. Circuit simulation design approaches.
Finally, the magnetic component design should be There are several steps that should be covered during
oriented to the converter level. In other words, it is the design stage:
convenient to simulate the whole circuit with the model
of the magnetic component in order to validate the
design. At this point it is possible to modify the working 3.1.1. Core selection
conditions of the converter in order to validate the Among the multiple design variables that should be
design under different working conditions. The selected in order to design a magnetic component, the
implementation of the model in a circuit simulator core is one of the key elements. The efficiency and size
netlist, is a very efficient way to carry out this stage. of the design depends strongly on the core selection.
Once again, the use of a CAD tool [1] that aids the
Usually, the core shape is selected accordingly to the
designer to create the model netlist from the design
application (power inductor, current transformer, RF
stage is the easier and more effective way to finish the
filter,). For this application, a POT shape seems
design loop.
reasonable.
The core size selection is based on the power that
III. DESIGN EXAMPLE
should be handled by the component. The area
In order to illustrate different alternatives designing a product (product of the window area by the effective
magnetic component, an inductor for a buck converter area of the core) is a very common rule applied in order
has been selected as design example. The specifications to select the core size. This parameter takes into account
of the converter correspond to a typical automotive the maximum apparent power that the device is able to
application. The specifications of the selected buck handle. Following the data-book recommendations, four
converter are shown in table 1. sizes will be considered for this design example:
POT26/16, POT30/19, POT36/22 and POT42/29.
The core material selection is usually based on the
operating frequency. However, the operating
frequency should be carefully selected accounting for
the harmonic content of the waveform, and not only
based on the fundamental harmonic. It is very different a
current waveform operating at 200 kHz with a large
ripple than another one with a small ripple, because the
Input Voltage(Vi) Output Voltage (Vo) Frequency Output Power last one could be assumed as a constant current. In other
42 V 14 V 200 kHz 300 W
Table 1. Buck converter specifications
words, the DC/AC ratio of the waveforms should be the
real parameter that should be used in order to select the
The converter specifications can be translated into material for the core. In this particular example the
waveform specifications. The current ripple is another material that has been selected is Ferroxcube 3F3.
degree of freedom that has been selected as 72% for this
3.1.2. Wire Selection 2. The expression (2) is only valid if the skin and
proximity effects are negligible. This could be
The diameter of the wires should be selected in order to
commonly true for the first harmonics, but not for
minimize the losses as much as possible. The diameter
the high order harmonics. Therefore, it is convenient
should be selected accounting for the harmonic content
to take into account the AC resistance in order to
of the current waveform. As mentioned above, the
calculate accurately the copper losses. The problem
DC/AC ratio of the current is the parameter that should
is that there are not simple equations to calculate the
be used as reference for the diameter selection. For
resistance considering the skin and proximity effects.
example, if the waveform applied to the inductor is a
Probably, the most commonly applied equations to
sinusoidal one operating at f Hz, the maximum diameter
consider the AC resistance are the Dowell [2]
should be two times the skin depth () in order to take
equations. These equations are not very complex to
advantage of the whole wire section. The equation (1)
be solved with the aid of a computer software, but it
shows the value of the skin depth, where f is the
is not very efficient (in terms of time) to solve them
frequency, 0 is the permeability of the air and is the
by hand for each design iteration. Dowell
conductivity of the copper.
equations take into account the skin and proximity
effects, but they are only applicable for 1D field
1
= distributions. If Dowell equations are applied to
f 0 (1) calculate the resistance at each harmonic frequency,
the total copper losses can be calculated as shown in
However, if the waveform is a typical saw-tooth shape, equation (3).
with an important DC level, it is not convenient to select
the diameter accounting only for the skin depth of the
fundamental harmonic. The DC losses could become P = I2DC*RDC + I2rms_1*RAC_1 +
+ I2 rms_2*RAC_2 + I2 rms_3*RAC_3 + (3)
more important than the AC losses contribution.
Obviously, the larger the diameter, the lower the DC
Where Irms_i is the rms value of the harmonic i and
losses. The ratio between AC and DC losses should be
RAC_i is the resistance at the frequency of the
considered for the diameter selection. Commonly, for
harmonic i. It can be demonstrated that the total
saw-tooth waveforms, the diameter should be something
copper losses can be calculated as the addition of
above two times the skin depth in order to reduce total
these terms.
losses (AC and DC).
In this particular example, and considering the skin 3.1.4. Core losses
depth at the fundamental harmonic (200 kHz, = 0.15
mm), a 0.3 mm diameter wire should be selected There are different approaches valid to calculate the
(AWG30). However, since there is a DC level, a 1 mm core losses with different accuracy levels [5][6]. In the
diameter wire (AWG19) has been selected for the design stage it is convenient to use a simple enough core
analysis. model appropriate to calculate the core losses without
the application of very complex expressions. There are
Once the diameter has been fixed with the AC/DC models that account for Foucault, hysteresis and residual
losses criteria, the way to limit the maximum losses is looses, but they are very complicate to be applied in a
increasing the total wire section adding parallel wires. A design iterative process. One of the most common
rule of thumb that is commonly applied in order to limit models applied to calculate the core losses is based on
the copper losses and temperature rise is limiting the the application of Steinmetz equation (4), where Ve is
current density. In this example, the current density has the core volume, f is the frequency, B is the variation
been limited to 10 A/mm2 adding 10 parallel wires. of flux density and k, and are material parameters.


3.1.3. Copper losses B
P = Ve k f (4)
Copper losses can be also calculated using expressions 2
with different accuracy levels.
Table 3 shows three different designs for several core
1. The simplest expression to calculate copper losses is
sizes in order to compare the losses that have been
considering the DC resistance (RDC) of the wire and
obtained using the analytical design. The core losses
the RMS value of the current (Irms).
have been calculated applying Steinmetz expression.
The copper losses are calculated using the RMS current
P = I2rms * RDC (2)
value and the DC resistance value.
Number Parallel Gap Copper Core Total
Core Size Material Wire
of Turns Turns Length Losses Losses Losses
POT 26/16 3F3 AWG 19 12 2 1.93 mm 4.6 W 8.2 mW 4.6 W
POT 30/19 3F3 AWG 19 8 3 1.13 mm 2.23 W 15.5 mW 2.25 W
POT 36/22 3F3 AWG 19 6 6 912 m 1W 21 mW 1.02 W
POT 42/29 3F3 AWG 19 4 10 444 m 439 mW 49.2 mW 488 mW
Table 3. Comparison of four designs using simple analytical expressions.

In order to illustrate that the application of this design It can be seen in table 4, that using all the parameters as
approach does not provide the optimum design, a CAD degrees of freedom, the losses can be drastically
tool tailored for magnetic component designs [1] has reduced (from 520 mW to 126 mW). Obviously, the use
been used in order to design the same component. The of these degrees of freedom is not a feasible option
main differences between the design procedure applied using hand calculations. Another point to consider is
in table 3 and the one using the CAD tool are as the global design, and not only the losses. As shown in
follows: table 4, the air gap and the core size of the lower losses
design are larger. The designer should introduce in the
The CAD tool is able to calculate many
design process constrains (core size, gap length,) that
combinations of core size, wire diameter and core
should be considered in the design flow. The use of the
material in order to find the lower losses design
CAD tool simplifies these design iterations accounting
Core losses are calculated applying Steinmetz
for those constrains.
expression in both approaches, but copper losses
are calculated in the CAD tool by means of Dowell
3.2. Modeling
expressions and FFT decomposition, as shown in
(3). Therefore, copper losses should be more In order to illustrate the benefits of the modeling task
accurate with the CAD tool application. for magnetic component designing, an inductor design
Since the CAD tool is able to calculate any combination will be modeled and simulated. It will be shown the
of core and wire sizes, three different design processes impact of the winding strategy in the component
have been carried out in order to illustrate the approach. behavior. An alternative design, different than the
ones shown in table 4, has been selected for this
1. Design 1 has been obtained with core shape (POT), purpose. This design has been selected for three
core size (POT42/29), core material (3F3) and wire reasons:
diameter (AWG 19) fixed as design constrains. It
can be seen that the result is the same as obtained A simple winding strategy with two layers of
before. turns in parallel
The whole window height is filled with turns
2. Design 2 has been obtained using wire diameter as
Enough room available in the window in order to
degree of freedom. Core shape (POT), core size
be able to move the windings across the window
(POT42/29) and core material (3F3) have been fixed
width.
as design constrains.
The inductor design characteristics are shown in the
3. Design 3 has been obtained using both, the core size table 5.
and wire diameter as degrees of freedom. Core shape
(POT) and core material (3F3) have been fixed as Core Size POT42/29
design constrains. Core Material 3F3
Wire Diameter AWG10 (1.56 mm)
Design # Design 1 Design 2 Design 3 Number of Turns 5
Core Size POT42/29 POT42/29 POT66/56 Number of parallel turns 2
Core Material 3F3 3F3 3F3
Wire Diameter AWG19 AWG15 AWG10
Gap 737.3 m
(1 mm) (1.56 mm) (2.73 mm) Table 5. Characteristics of the modeled inductor
Number of 4 4 3
Turns Two winding strategies have been modeled in order to
Number of 10 8 10
parallel turns illustrate the effect of the position of the conductors in
Gap 444 m 444 m 685 m the window from a point of view of losses. The situation
Winding 471 mW 243.4 mW 83 mW of the winding in the window is shown in figure 3.
Losses
Core Losses 49.2 mW 49.2 mW 43 mW Design A is the one with the winding close to the air gap
Total Losses 520.2 mW 292.6 mW 126 mW and Design B presents the wires separated 1.33 mm
Table 4. Results using CAD design. from the air gap.
At this point, only the device level design and
modeling have been covered. However, the selection of
the optimum design should be carried out in
combination with the circuit level. It is not possible to
determine which design, between A and B, is better if
the circuit is not considered in the final decision.
Depending on the magnetic component working
"Design A" "Design B" conditions, that are determined by the circuit, the design
Figure 3. Winding strategies that have been modeled selection could be different. In the next section this
point is highlighted.
The modeling task is not a simple stage, because an
accurate model should be applied in order to extract 3.3. Circuit simulation
useful conclusions. There are many proposals to model Both inductor models have been included in the buck
magnetic components [3][4]. Since the effect that is converter schematic and looses have been obtained
pretended to analyze in this case is bi-dimensional (air using Simplorer [1], as it is shown in figure 5.
gap fringing flux), a model based on Finite Elements
Analysis (FEA) has been selected [7]. Therefore, the
two situations shown in figure 3 have been modeled
ref

Type 3 Controller
WP_LIN1

using a FEA based model with PEmag [1]. Figure 4


tY t
vout

shows the resistance as a function of the frequency for Macro21


V
VM2
ET1

+
both inductors. EMWorkShop
A
+ +
Macro1
EMWorkShop1
V

E2 VM1
AM1 C2 R1
Macro11

Model 1: Model 2:
no with
margin margin
tape tape

Figure 5.Circuit simulation with both models

The circuit has been simulated under two load


conditions. The high load is 50 A and the low load is
5A. The current waveform for both load conditions is
shown in figure 6. The simulation results are shown in
table 7 for two load conditions.

Figure 4. Resistance as a function of the frequency of 60.00 A

designs A (up) and B (down) 50.00

Iout = 50 A 40.00

30.00

It can be seen that the resistance at 1 MHz is 2 times 20.00

Iout = 5 A
higher for the case with the conductors close to the air
10.00

gap. However, the DC resistance, as shown in table 6 is


1.800m 1.850m 1.875m 1.900m 1.925m 1.950m 2.000mT

lower placing the conductors close to the air gap, Figure 6.Two load conditions have been simulated
because the length of the wires is smaller.
Case "Design A" "Design B"
Design DC Resistance Power losses Power losses
A 0.734 m High Load 2W 3.5 W
B 0.663 m Low Load 450 mW 350 mW
Table 6. DC Resistance of both inductors. Table 7. Simulation results.
As shown in table 7, depending on the load condition, These results confirm the conclusions extracted in the
one or the other design could be better. If the DC level previous point. The efficiency of the converter can be
of the current is very important compared with the AC improved one point if the appropriate design is selected.
level (high load), the design with lower DC resistance
should present lower loses. This is the reason to obtain "Design A" "Design B"
Load
1.5 W lower losses in Design A than in Design B Efficiency Eficiency
"Design A" is better
for the high load case. However, if the AC level of the High 89 % 87 %
than "Design B"
current presents the same magnitude order than the DC "Design B" is better
level (low load), the air gap effect should be reduced as Low 91 % 92 %
than "Design A"
much as possible in order to reduce the AC resistance. Table 8. Measured efficiency with both designs and
Therefore, the Design B is better for the low load case both load conditions
because the conductors are separated from the gap, and
the AC resistance is reduced.
V. CONCLUSIONS
This example illustrates that the magnetic component
This work describes a procedure to design magnetic
design should be oriented to optimize the whole circuit,
components based on three main steps: analytical
and the final design depends on the circuit working
design, modeling and simulation. A design example has
conditions.
been presented in order to highlight that the use of CAD
tools is a recommended option for magnetic component
IV. VALIDATION design. Since the number of degrees of freedom is very
In order to validate the previous results, a 120W and high, the calculation power of the CAD tool provides a
150 kHz version of the buck converter was built and great help for the designer. It has been shown that the
measured. The same converter has been operated with modeling task provide a deep insight into the design.
two inductors with different separation of the Using accurate models it is possible to optimize the final
conductors to the air gap. Figure 6 shows a photograph design from the point of view of the winding strategy.
of the inductors and the actual prototype. Finally, the simulation stage is very useful to optimize
the design at converter level. The simulation stage
provides information about the impact of the design on
the converter behavior.

VI. REFERENCES
[1] PEmag and PExprt. Ansoft Corporation Web Site,
http://www.ansoft.com
[2] Effect of eddy currents in transformer
windings. P. L. Dowell. Proceedings of the IEE.
[3] Additional losses in high frequency magnetics
due to non ideal field distributions Severns, R.
Applied Power Electronics Conference and
Exposition, 1992. APEC '92.
[4] Improved analytical modeling of conductive
losses in magnetic components Ferreira, J.A.
IEEE Transactions on Power Electronics.
[5] Theory of Ferromagnetic Hysteresis Jiles, D.C.,
Atherton, D. L. Journal of magnetism and
magnetic materials. N 61, p.48-60, 1986
[6] Generalized Preisach model of hysteresis
Mayergoyz, I.D.; Friedman, G. IEEE Transactions
on Magnetics, Volume: 24 Issue: 1 , Jan. 1988
Figure 6.Photograph of the inductors (up) and [7] A full procedure to model high frequency
converter (down) transformer windings Asensi, R.; Cobos, J.A.;
Garcia, O.; Prieto, R.; Uceda, J.. Power
Electronics Specialists Conference, PESC '94
The efficiency under two load conditions was measured. Record., 25th Annual IEEE , 1994.
The results are summarized in table 8.