Help - Ferrite core loss density chart interpretation? Losses with DC+AC core flux?

[Tony]

I'm trying to minimize losses in a high-power power factor controller driving a
continuous conduction flyback power switch circuit (actually it's a hybrid, with
resonant attributes as well, but that doesn't affect this question). As I have a
very limited power dissipation budget, I need to optimize every part of the
project, and as the PSU necessarily comes first, that's my current target.

Problem 1: In the absence of any info to the contrary I have been interpreting
the "B" parameter in the core loss density curves (mW/cc vs frequency, with B
and temperature as parameters) as p-p flux density excursion regardless of the
mean flux density (presumably as long as the flux never approaches saturation).
I have just realized from an obscure downloaded Philips document that the "B"
parameter seems to mean half the p-p flux (good for me, since I can now use
double the flux excursion), but although not it's certainly still not clear,
there seems to be an inference that the curves only hold if the flux varies from
-B to +B, ie, the average flux is zero, and I can't use any of this data to
guess the losses when B varies between, say, +300mT and +400mT. I'm further
guessing that the curves also assume sine wave excitation? Can anyone confirm or
correct all this conjecture?

Problem 2: whatever the answer to problem 1, is there any "rule of thumb", or
mathematical approximation that might help me to estimate quantitatively the
trend in losses if not the actual losses as the mean flux changes (I can't
realistically test this, at least until after I get starting design point)?

All comments will be welcome.



Tony (remove the "_" to reply by email)

 

[CBarn24050]

Hi, the core loss is approxamatly proportional the the product of the flux
sweep and frequency. The "DC" flux has no effect on loss other than putting the
flux sweep on a different part of the BH curve.

 

[John Woodgate]

(in <[email protected]>) about 'Help - Ferrite
core loss density chart interpretation? Losses with DC+AC core flux?',

Problem 1: In the absence of any info to the contrary I have been
interpreting the "B" parameter in the core loss density curves (mW/cc vs
frequency, with B and temperature as parameters) as p-p flux density
excursion regardless of the mean flux density (presumably as long as the
flux never approaches saturation). I have just realized from an obscure
downloaded Philips document that the "B" parameter seems to mean half
the p-p flux (good for me, since I can now use double the flux
excursion), but although not it's certainly still not clear, there seems
to be an inference that the curves only hold if the flux varies from -B
to +B, ie, the average flux is zero, and I can't use any of this data to
guess the losses when B varies between, say, +300mT and +400mT. I'm
further guessing that the curves also assume sine wave excitation? Can
anyone confirm or correct all this conjecture?

Well, not with 100% certainty, but with over 90%. The core losses are
represented by the area inside the hysteresis loop shown by the B-H
curve. This area varies with mean induction in a way peculiar to the
material

Problem 2: whatever the answer to problem 1, is there any "rule of
thumb", or mathematical approximation that might help me to estimate
quantitatively the trend in losses if not the actual losses as the mean
flux changes (I can't realistically test this, at least until after I
get starting design point)?

Unless you can find 'Hanna Curves' for the material, you probably need
to wind a test inductor and plot them yourself. A Google search for
Hanna Curves might let me get away without further explanation.

 

[John Woodgate]

I read in sci.electronics.design that CBarn24050 <[email protected]>
wrote (in <[email protected]>) about 'Help -
Ferrite core loss density chart interpretation? Losses with DC+AC core
flux?', on Sun, 29 Feb 2004:

Hi, the core loss is approxamatly proportional the the product of the flux
sweep and frequency.

Proportional to a *power* of frequency that depends on the material.

The "DC" flux has no effect on loss other than putting the
flux sweep on a different part of the BH curve.

But in doing that, it forces the B-H curve to enclose a different area,
which represents a different value of core loss.

 

[R.Legg]

I have just realized from an obscure downloaded Philips document that

the "B" parameter seems to mean half the p-p flux (good for me, since I can
now use double the flux excursion), but although not it's certainly still not > clear, there seems to be an inference that the curves only hold if the flux
varies from -B to +B, ie, the average flux is zero, and I can't use any of
this data to guess the losses when B varies between, say, +300mT and +400mT.
I'm further guessing that the curves also assume sine wave excitation? Can
anyone confirm or correct all this conjecture?

Although the charts do reference Bpk (not Bppk) and they are compiled
from sinusoidal data, there is no correction factor for unipolar flux
excursions. You still use the Bpk of your circuit for direct reference
and assume it is rest to zero on each cycle, with an equal duty of
rising and falling flux.

This seems to ignore the fact that a sinusoid produces Bpk at twice
the operating frequency - unipolar stress should then intuitively
allow a frequency multiplier of 0.5 when using the charts - but it
isn't done. If anything, there is a tendency in the literature that
suggest that biased excitation actually needs a punitive correction
factor.

Problem 2: whatever the answer to problem 1, is there any "rule of thumb", or
mathematical approximation that might help me to estimate quantitatively the
trend in losses if not the actual losses as the mean flux changes (I can't
realistically test this, at least until after I get starting design point)?

The trend is illustrated in the core loss formulas and coefficients
published by the mfrs; these are far from uniform and are seldom
interchangable. Their margin on worst-case performance will also vary
arbitrarily.

Because the losses generated in two identical core parts, with the
same listed material, can vary by a factor of two, it is actually
dangerous to use measured performance of a test sample as a guideline,
particularly if core loss forms a significant portion of total loss at
the component's intended temperature limit, (where this loss will
exibit a PTC). I expect that published performance 'improvements' of
recent core materials is more a function of improved process control
of this variation.


A more upbeat article by Ridley

http://www.switchingpowermagazine.com/downloads/Jan 02 designer.pdf

Thermal effects examined by Lu

http://www.itee.uq.edu.au/~aupec/aupec99/lu99.pdf

Non-Sinusoidal excitation by Sullivan

http://thayer.dartmouth.edu/other/inductor/papers/gse.pdf

RL

 

[Tony]

First, thank you all for confirming the meaning of the data.

John, that was my feeling too, but based on cbarn24050's post I'm going to
assume that the hysteresis loop doesn't change dramatically at low flux levels,
so I think I'll design for peak flux = half the minimum saturation flux over
temperature, and flux excursion = 75% of what the core loss density curves allow
for my dissipation budget, and see what happens.

I read in sci.electronics.design that CBarn24050 <[email protected]>
wrote (in <[email protected]>) about 'Help -
Ferrite core loss density chart interpretation? Losses with DC+AC core
flux?', on Sun, 29 Feb 2004:


Proportional to a *power* of frequency that depends on the material.


But in doing that, it forces the B-H curve to enclose a different area,
which represents a different value of core loss.

Tony (remove the "_" to reply by email)

 

[Tony]

<snip> when B varies between, say, +300mT and +400mT.

I just noticed this - I don't know what I was thinking. It was supposed to be
200mT to 300mT, but even that was "too close to the wind", so I'm thinking I'll
keep a bit further away from saturation.


Tony (remove the "_" to reply by email)

 

[Tony]

Thank you. Lots of food for thought. Please see inline comments below...

Although the charts do reference Bpk (not Bppk) and they are compiled
from sinusoidal data, there is no correction factor for unipolar flux
excursions. You still use the Bpk of your circuit for direct reference
and assume it is rest to zero on each cycle, with an equal duty of
rising and falling flux.

I can appreciate that it's done, but this still seems to me to be an
unnecessarily pessimistic approach, and illogical - the area in the hysteresis
curve from -B to +B seems like it must be at least double the area in the curve
from 0 to +B (although still less than the area between 0 and +2B).

This seems to ignore the fact that a sinusoid produces Bpk at twice
the operating frequency - unipolar stress should then intuitively
allow a frequency multiplier of 0.5 when using the charts - but it
isn't done.

I can understand why it isn't done. If you continue ramping H down past zero, B
is dragged along without complaining. It's only when you reverse H that B finds
it hard you get hysteresis, so my intuition says that applying an asymmetric the
absolute value wave (at 2f) instead of the original symmetric wave (at f) would
indeed double the number of reversals, and hence increase losses.

If anything, there is a tendency in the literature that
suggest that biased excitation actually needs a punitive correction
factor.

Yes, that's the factor I'm after.

The trend is illustrated in the core loss formulas and coefficients
published by the mfrs; these are far from uniform and are seldom
interchangable. Their margin on worst-case performance will also vary
arbitrarily.

Because the losses generated in two identical core parts, with the
same listed material, can vary by a factor of two, it is actually
dangerous to use measured performance of a test sample as a guideline,
particularly if core loss forms a significant portion of total loss at
the component's intended temperature limit, (where this loss will
exibit a PTC). I expect that published performance 'improvements' of
recent core materials is more a function of improved process control
of this variation.

OK, this is all new to me. I had no idea on the tolerances involved, and that's
all the more reason to get a proper handle on the mechanisms, even if only
empirical, before prototyping..

A more upbeat article by Ridley

http://www.switchingpowermagazine.com/downloads/Jan 02 designer.pdf

Thermal effects examined by Lu

http://www.itee.uq.edu.au/~aupec/aupec99/lu99.pdf

Non-Sinusoidal excitation by Sullivan

http://thayer.dartmouth.edu/other/inductor/papers/gse.pdf

All interesting (and enlightening) reading. And this last reference also quotes
these other references for the effect of DC pre-magnetization on core loss...

A. Brockmeyer, Experimental evaluation of the influence of dc pre-magnetization
on the properties of power electronic ferrites, in APEC ’96. Eleventh Annual
Applied Power Electronics Conference, 1996, pp. 454.60.

A. Brockmeyer and J. Paulus-Neues, Frequency dependence of the
ferrite-loss increase caused by pre-magnetization, in Twelfth Annual
Applied Power Electronics Conference and Exposition, 1997, pp. 375.80.

Wai Keung Mo, D.K.W. Cheng, and Y.S. Lee, Simple approximations of the dc flux
influence on the core loss power electronic ferrites and their use in design of
magnetic components, IEEE Transactions on Industrial Electronics, vol. 44, no.
6, pp. 788.99, 1997.

But I've only so far been able to get the abstracts. Can anyone out there
re-point me? also, since A. Brockmeyer seems to be a guiding light on this
subject, I added his name to my Google search string, and came up with this
interesting document...

http://www.twobarkingdogs.com/ron/Documents/apec02_core_loss.PDF
Tony (remove the "_" to reply by email)

 

[Tony]

<snip>Non-Sinusoidal excitation by Sullivan

http://thayer.dartmouth.edu/other/inductor/papers/gse.pdf

RL

The document linked above is very interesting (now that I've had time to read
it), as it separates minor loops from major loops in an arbitrary waveform. The
minor loops are modeled the same regardless of whether they're centred on zero
or up near saturation; eg, using a, b, T for alpha, beta, theta, the
time-averaged core loss density (after Steinmetz)
= k.f^a.B^b (frequency, peak flux).
And instantaneously,
Pv(t) = k1 . (dB/dt)^a . (Bmax-Bmin)^(b-a)
where k1 = k / ((2Pi)^(a-1) . IntegralOver0To2Pi(cos(T)^a.Sin(T)^(b-a)dT))
(sorry - it's hard to write readable maths in ASCII).

Considering Sullivan is seeking ever better accuracy in this work, it's hard to
imagine him ignoring the effect of DC bias if it was very significant. For my
original question, it's possible I can ignore it too, and just use half the p-p
flux excursion. But reading on this subject has shown me the apparently even
more significant effect of the waveform, so back to the drawing board.

Tony (remove the "_" to reply by email)

 

[John Woodgate]

(in <[email protected]>) about 'Help - Ferrite
core loss density chart interpretation? Losses with DC+AC core flux?',

John, that was my feeling too, but based on cbarn24050's post I'm going
to assume that the hysteresis loop doesn't change dramatically at low
flux levels, so I think I'll design for peak flux = half the minimum
saturation flux over temperature, and flux excursion = 75% of what the
core loss density curves allow for my dissipation budget, and see what
happens.

You didn't say how much asymmetry there is in your induction excursion.
If it is small, the effect on hysteresis loss will be small. But what is
'small' depends on the type of ferrite you are using and may not be very
much.

 

[John Woodgate]

(in <[email protected]>) about 'Help - Ferrite
core loss density chart interpretation? Losses with DC+AC core flux?',

And instantaneously,
Pv(t) = k1 . (dB/dt)^a . (Bmax-Bmin)^(b-a)
where k1 = k / ((2Pi)^(a-1) . IntegralOver0To2Pi(cos(T)^a.Sin(T)^(b-
a)dT)) (sorry - it's hard to write readable maths in ASCII).

You are using too many spaces, but you DO need spaces around -, + and x.
It's possible to keep everything on one line, which helps a lot:

Pv(t) = k1.(dB/dt)^a.(Bmax-Bmin)^(b - a)

where

k1 = k/((2Pi)^(a - 1).Int[0:2Pi](cos(T)^a.sin(T)^(b - a)dT))

If the expression must split over two lines, end the first line with -,
+ or x. If there is a long denominator, as there is for k1, consider
defining 1/k1 instead.

 

[Tony]

John,

Thanks - all points taken. Actually I didn't realize my news client had split
the line when I pasted it in.

(in <[email protected]>) about 'Help - Ferrite
core loss density chart interpretation? Losses with DC+AC core flux?',

And instantaneously,
Pv(t) = k1 . (dB/dt)^a . (Bmax-Bmin)^(b-a)
where k1 = k / ((2Pi)^(a-1) . IntegralOver0To2Pi(cos(T)^a.Sin(T)^(b-
a)dT)) (sorry - it's hard to write readable maths in ASCII).

You are using too many spaces, but you DO need spaces around -, + and x.
It's possible to keep everything on one line, which helps a lot:

Pv(t) = k1.(dB/dt)^a.(Bmax-Bmin)^(b - a)

where

k1 = k/((2Pi)^(a - 1).Int[0:2Pi](cos(T)^a.sin(T)^(b - a)dT))

If the expression must split over two lines, end the first line with -,
+ or x. If there is a long denominator, as there is for k1, consider
defining 1/k1 instead.

Tony (remove the "_" to reply by email)

 

[Tony]

(in <[email protected]>) about 'Help - Ferrite
core loss density chart interpretation? Losses with DC+AC core flux?',


You didn't say how much asymmetry there is in your induction excursion.
If it is small, the effect on hysteresis loss will be small. But what is
'small' depends on the type of ferrite you are using and may not be very
much.

Originally the range was 200-300mT, but it's all a big juggle, and seems like it
might now end up at maybe 150-250mT.
Tony (remove the "_" to reply by email)

 

[R.Legg]

You didn't say how much asymmetry there is in your induction excursion.

Originally the range was 200-300mT, but it's all a big juggle, and seems like it
might now end up at maybe 150-250mT.

You refer to the Continuous Conduction Mode boost/flyback PFC
topology; does this mean you are trying to get an isolated output off
the PFC switch, or are you simply applying a flyback isolation circuit
after a conventional PFC stage? Your reference to high power is
curious, in any event, as the flyback topology is seldom used above
150W, for various reasons. One of these reasons is the inefficient use
of the isolation transformer imposed by the topology.

If this is a single stage PFC flyback isolation circuit, then the flux
levels and conduction mode will be all over the map, during a single
low-frequency input half-period. An average would have to be
calculated.

If this is a two stage converter, and you are only concerned with the
second stage flyback isolation transformer, then a static worst case
full-load operating condition can be characterized by a peak deltaB
level. This is imposed upon the residual level presented by the
incomplete energy transfer technique (continuous conduction mode)
mentioned.

RL

 

[John Woodgate]

(in <[email protected]>) about 'Help - Ferrite
core loss density chart interpretation? Losses with DC+AC core flux?',

Originally the range was 200-300mT, but it's all a big juggle, and seems
like it
might now end up at maybe 150-250mT.

So your d.c. induction is 150 mT? That's BIG, for any ferrite.

 

[Tony]

You refer to the Continuous Conduction Mode boost/flyback PFC
topology; does this mean you are trying to get an isolated output off
the PFC switch, or are you simply applying a flyback isolation circuit
after a conventional PFC stage? Your reference to high power is
curious, in any event, as the flyback topology is seldom used above
150W, for various reasons. One of these reasons is the inefficient use
of the isolation transformer imposed by the topology.

I'm looking at a few options right now, even ones that are considered to make
poor use of materials (an EE65 core set in N87 material is not really much more
expensive than an ETD49 in a lower grade). But I'm curious too - what topology
would you recommend for a high power (>1kW) PFC front end?

I'm working on both the cascaded flyback PFC front end + push-pull isolated
down-converter, and a single stage isolated stage which has a resonant forward
path as well as the usual flyback path. Both have big tradeoffs. Which ever I
use will be implemented as a set of parallel converters driven from a single
controller in staggered phase (vastly lower cap ripple current).

If this is a single stage PFC flyback isolation circuit, then the flux
levels and conduction mode will be all over the map, during a single
low-frequency input half-period. An average would have to be
calculated.

Yes I intended to do that, but the dominant condition will always be max power
transfer at the peak of the AC input, so I was starting there.

If this is a two stage converter, and you are only concerned with the
second stage flyback isolation transformer, then a static worst case
full-load operating condition can be characterized by a peak deltaB
level. This is imposed upon the residual level presented by the
incomplete energy transfer technique (continuous conduction mode)
mentioned.

RL

Tony (remove the "_" to reply by email)

 

[R.Legg]

I'm working on both the cascaded flyback PFC front end + push-pull isolated

down-converter, and a single stage isolated stage which has a resonant forward
path as well as the usual flyback path. Both have big tradeoffs. Which ever I
use will be implemented as a set of parallel converters driven from a single
controller in staggered phase (vastly lower cap ripple current).

So you are actually concerned about core loss in a gapped ferrite
boost inductor used in a non-isolated PFC circuit.

At 1KW, this will be a large core with a large gap, and the flux
excursion will be superimposed upon a low frequency flux that is
itself equal to or larger than the ppk AC flux. The LF flux, in
itself, does not contribute significantly to the loss. Because of the
large gap required, significant local copper winding loss can be
generated in the fringing fields present.

The worst case condition that determines the thermal and electrical
energy storage limitations of your circuit will occur at full load and
low line input. Under normal line conditions, you should see no
apparent problems with saturation or temperature rise, if your initial
part design is in the ballpark.
Both core loss and copper loss will double as input line voltage falls
from 120 to 85 VAC, where the inductor will tend to saturate at the
peak input, simply due to the LF current contribution.

Commercial applications will typically adopt a core with distributed
gap characteristics to avoid the low saturation flux density limit of
ferrites, and the winding problems presented by the concentrated gap
(MPP, cool-mu powdered composite materials). Other materials that can
handle the higher flux may require gapping just to get a usable L
value, when turns are sufficient to maintain core loss at reasonable
levels (micro-lite, metglass and finemet amorphous materials).

http://www.grouparnold.com/products/powder/powder_catalogs.htm
http://www.mag-inc.com/
http://www.dextermag.com/
http://www.kgmagnetics.com/software.htm
http://www.micrometals.com/
http://www.eastern-components.com/honeywellframe-elib-microlite.htm
http://www.hitachi-metals.co.jp/e/prod/prod13/p13_02.html

Application notes for PFC boost inductors tend to concentrate on this
alternate material use and are supplied by the manufacturers of that
material.

If your application does not address a wide-range input, then you
should be able to get a ferrite part to work fairly well, but you will
still be using gap sizes that are off most mfr's charts, and you
should also expect to see discontinuous operation at the higher input
voltages and lighter loads.

RL

 

[Tony]

Thanks once again for the tips and links. I hadn't considered the copper losses
from the fringing (more homework needed on this and on the powdered core
materials). I obviously have quite a few steps to go, but on the surface it
seems that even though the materials you mentioned can take higher DC flux, they
can't take as much flux excursion at 50kHz as the better ferrites, so I need to
rethink a lot of things to make a comparison.

Re the wide range input, I was prepared to compromise as required, the control
loops being implemented in a combination of software and hardware, so the
intention was to impose restraints where needed to compromise between power
factor and efficiency / throughput where appropriate.

So you are actually concerned about core loss in a gapped ferrite
boost inductor used in a non-isolated PFC circuit.

At 1KW, this will be a large core with a large gap, and the flux
excursion will be superimposed upon a low frequency flux that is
itself equal to or larger than the ppk AC flux. The LF flux, in
itself, does not contribute significantly to the loss. Because of the
large gap required, significant local copper winding loss can be
generated in the fringing fields present.

The worst case condition that determines the thermal and electrical
energy storage limitations of your circuit will occur at full load and
low line input. Under normal line conditions, you should see no
apparent problems with saturation or temperature rise, if your initial
part design is in the ballpark.
Both core loss and copper loss will double as input line voltage falls
from 120 to 85 VAC, where the inductor will tend to saturate at the
peak input, simply due to the LF current contribution.

Commercial applications will typically adopt a core with distributed
gap characteristics to avoid the low saturation flux density limit of
ferrites, and the winding problems presented by the concentrated gap
(MPP, cool-mu powdered composite materials). Other materials that can
handle the higher flux may require gapping just to get a usable L
value, when turns are sufficient to maintain core loss at reasonable
levels (micro-lite, metglass and finemet amorphous materials).

http://www.grouparnold.com/products/powder/powder_catalogs.htm
http://www.mag-inc.com/
http://www.dextermag.com/
http://www.kgmagnetics.com/software.htm
http://www.micrometals.com/
http://www.eastern-components.com/honeywellframe-elib-microlite.htm
http://www.hitachi-metals.co.jp/e/prod/prod13/p13_02.html

Application notes for PFC boost inductors tend to concentrate on this
alternate material use and are supplied by the manufacturers of that
material.

If your application does not address a wide-range input, then you
should be able to get a ferrite part to work fairly well, but you will
still be using gap sizes that are off most mfr's charts, and you
should also expect to see discontinuous operation at the higher input
voltages and lighter loads.

RL

Tony (remove the "_" to reply by email)

 

[legg]

Thanks once again for the tips and links. I hadn't considered the copper losses
from the fringing (more homework needed on this and on the powdered core
materials). I obviously have quite a few steps to go, but on the surface it
seems that even though the materials you mentioned can take higher DC flux, they
can't take as much flux excursion at 50kHz as the better ferrites, so I need to
rethink a lot of things to make a comparison.

I am not completely sold on the benefits of the currently available
composite or amorphous materials, because of their loss
characteristics, which seldom compare favourably with ferrite.
However, if you can pay more attention to the winding structure so
that it suits the lossy material, the net result is usually
functional.

Re the wide range input, I was prepared to compromise as required, the control
loops being implemented in a combination of software and hardware, so the
intention was to impose restraints where needed to compromise between power
factor and efficiency / throughput where appropriate.

I believe I've heard this prescription before -

Patient, "Doc, it hurts when I do this."
Doc, "Don't do that."

I guess the problem is the interface - how fast/accurate does it have
to be, how is it isolated and on and on......

I've seen PFC stages, highly touted as the final solution to
everyone's problems, go into production completely ignoring the fact
that the ferrite PFC inductors were saturating at low line. It isn't
automatically a bad thing if the peak currents (peak to average gets
nasty) don't harm anything and EMC is maintained.

Swinging cores, with permeabilities determined by gradually saturating
composite or graduated-gap structures are used some times in places
where wide dynamic ranges are required - the unintentionally
saturating ferrite part survived roughly on the same principle. You
could assume that the intentional structures show less fringing field,
as the saturated material still acts as a prefered path for flux, as
opposed to air.

There is a short review of gapped ferrite inductors in the TSC
technical notes.

http://www.tscinternational.com/mainferr.html

RL

 

[legg]

A note of warning when refering to the TSC app notes.

The formula given for surface temperature rise in

"Predicting Temperature Rise of Ferrite Cored Transformers" by
G.Orenchak from PET'03 conference

is corrupt.

deltaT = ( Pmw / Acm^2 ) ^0.833 doesn't even give results that
correspond to their own published example for TSF-7099-41-16-12-0000,
(chart in the powerpoint presentation with the same title).

The exponent selected typically gives rises that are less than half
real measured results.

The sample chart plotted shows a relationship of

deltaT = K ( Pmw / Acm^2 )

where K is 1.2 at low power, increasing to 1.0 at higher power
densities (>4000mW).

Which is close to the general rule of thumb -

1degC/mW/cm^2 +/-20%.

I don't know how they ever worked the exponent into it.

RL