Connect with us

Core loss article

Discussion in 'Electronic Design' started by Tim Williams, Jan 9, 2013.

Scroll to continue with content
  1. Robert Baer

    Robert Baer Guest

  2. legg

    legg Guest

  3. Fred Abse

    Fred Abse Guest

    Dowloaded instantly using wget.

    Looks good. I'll go away and read it properly,
    now. Thanks, Tim.
  4. Tim Williams

    Tim Williams Guest

    Maybe your computer/ISP can't talk to my server? Or maybe it was clogged
    or down at the moment?

    It's just a file on the http directory, nothing special, no scripts or

  5. Sjouke Burry

    Sjouke Burry Guest

    ??????? It works perfectly here(firefox/foxit reader on XP-SP3.).
  6. legg

    legg Guest

    I finally just saved the blank page that was displayed in IE6, as
    'Core_Loss.pdf'. This proved to be a working pdf document.

    Fiddling with preferences in Seamonkey/firefox, I pointed the
    application helper for Acrobat(application/pdf) to adobe reader7. This
    allowed the pdf to open in the browser.

    Never had either issue previously. Curious.

    Will review, anyways.

  7. Neon John

    Neon John Guest

    Firefox/Linux chugged away for awhile and then gave me a blank screen.
    Opera/Linux retrieved the paper normally. Apparently your server
    doesn't like Firefox, at least under Linux.

    Nicely done paper, BTW. What did you use to typeset it?


    John DeArmond
    Tellico Plains, Occupied TN
    See website for email address
  8. He used "Latex with hyperref package" according to the document
    properties, and pdfTex-1.40.12 to produce the PDF 1.5 file.

    What's that Type-3 Adobe font "F58" for? Oh, I see, the bullets on
    pages 6 and 8.

    Best regards,
    Spehro Pefhany
  9. Guest

    That paper bears all the hallmarks of having been done in LaTeX.

    Tim -- how in heck did you get the text to wrap around your figures? I want that!
  10. Fred Abse

    Fred Abse Guest

    It doesn't mind this Firefox under Linux, I just tried it. No problem.
  11. legg

    legg Guest

    I (sort of) groaned when you adopted MMKS as units. Perhaps you didn't
    realize that KW/m^3 of common core loss graphs can be read directly as
    mW/cm^3? Dropped, missing or mistranslated orders of magnitude can
    play havoc.

    Perhaps you should have titled this article
    Core_Loss_of_Powdered_Toroidal_Inductors? There's nothing wrong with
    adding information to titles, if it cuts down on the end user's time.

    A power loss budget, in preliminary design, is dependant on some major
    indices; surface area (or thermal impedance to ambient); maximum
    permitted spot temperature; and the total power loss.

    In free air, thermal impedance can be rule of thumbed as somewhere
    between 0.8 and 1.2 degreesC rise per mW/cm^2 of surface area. A
    forced air requirement reduces this, but places downstream
    environmental restrictions and costs on the entire system in which the
    component is expected to perform.

    Saying that a T-68-based inductor might dissipate one or two watts
    comfortably is not helping your readers make a decision on this point.

    Maximum permitted spot rise will either be the margined rating of the
    materials used, a similar but safety-agency-dominated limit value, or
    the maximum temperature at which the actual published core loss data
    remains valid. At the limit temperatures, the core loss characteristic
    become increasingly positive in temperature coefficient, encouraging
    thermal runaway. It is seldom characterized or specified for powdered
    cores, probably for this exact reason; and that it would be extremely
    discouraging to sales. For ferrites, manufacturers are more
    forthcoming, as the PTC of loss only starts to kick in above a certain
    grade-type-specific value, in a bathtub curve. By selecting the
    material grade, one also selects the intended end-use environment.

    Other core material features exhibit temperature dependence and should
    be reviewed at the design limits intended.

    Core loss is an AC phenomena. The Bmax in your formula should probably
    be deltaB/2, as this corresponds to the peak flux density for
    sinusoidal data presented by the manufacturer. This correspondence
    isn't guaranteed, as core loss is noted to vary, the farther it strays
    from the zero-flux crossing point, into regions approaching
    saturation. Very poor data from the industry or academia on this.

    As the tolerance of core loss can vary by a wide factor within a
    single material grade, it is unwise to use it as a main limiting
    factor in any design. Copper losses, which predictably increase
    logarithmically with increased current, will produce a much more
    reliable indication of end-use, load-limiting factors in the
    production of spot temperature limits.

    In isolating transformer applications, the amount of copper fill
    obtained can dominate calculations of through-put power capability for
    any core shape and topology. Philips/Ferroxcube and Siemens both
    publish guidelines illustrating expected power transfer through
    characteristic power core shapes, when used in specific topologies,
    over the frequency range of specific material grades.

  12. brent

    brent Guest

    Nice job on writing the paper. I cannot comment on the technical
    stuff becasue I am trying to absorb it. However,

    I would change:
    Seven Transistor Labs takes no responsibility for liability, damage or
    other adverse consequences incurred in the use, handling or operation
    of its

    to :
    the author takes no responsibility for liability, damage or
    other adverse consequences incurred in the use, handling or operation
    of its

    I suspect seven transistor labs does not really exist as any kind of
    legal entity with property to protect, and since STL is you
    ( completely , I assume) I would make it explicit that the author
    assumes no liability. After all, I suspect that protecting the author
    is more important than protecting a shell entity that has no value.
    (Of course I could be wrong on this...just food for thought If you
    want the mumbo jumbo in there)
  13. Tim Williams

    Tim Williams Guest

    Weird. Wonder if the mime type is funky or something.

  14. Tim Williams

    Tim Williams Guest

    Too bad, you'll have to pay :^)

    Kidding, it's the /wrapfig/ package. Can be trouble to make it work right
    unfortunately, but it works most of the time.

  15. Tim Williams

    Tim Williams Guest

    Ah, good point. It would be nice to have an LLC or something some day,
    but you're right it's basically me being silly right now. :)

  16. Tim Williams

    Tim Williams Guest

    I try to work in the handiest orders of magnitude, which for cores are cm
    or mm (mm wins because it's an even 10^3 out), uH or nH (I'll usually
    write mu_0 ~= 1.26 nH/mm, and A_L in nH/t^2 for lower values, but I went
    with uH for consistency here), MHz (problems are usually given in kHz, but
    you've got to admit, "0.2MHz" is a smidge faster on the calculator) and
    uWb (V and us). Adopting

    I should add a note that some manufacturers work in kW/m^3 (Ferroxcube I
    believe usually does), which is equal as you say.

    Further, I should change it for consistency's sake to uW/mm^3 (again
    noting the identity), or change the axis to read 10^3 smaller and go with
    mW/mm^3. Or W/mm^3 and label the axis with multipliers (0.1m, 1m, 10m,

    I did in fact change the horizontal axis on the plot; Micrometals gives
    their graphs in gauss and oersted, which just... oh come now. Yes, gauss
    are just teslas done with cm^-2 instead of m^-2, but that factor of
    it's-not-a-power-of-10^3 bothers me more.
    Especially if they have trouble getting into that candy center, in which
    case a verbose title helps, or an index page.... but not some asshole
    Usenet poster shoving around links with no description. ;o)

    And so on-- hence why I didn't cover that, because it's a whole mess of
    thermal budget and temperatures and airflow and blah blah blah... Highly
    important of course, just beyond scope. The procedure is given that, once
    you know how much power you can withstand in a component or its rough
    outline, you can find the power.
    They do actually give some examples:
    Powdered cores are like global warming: it's fine for a while, then,
    hockey stick!

    They also go into more detail on that "messy stuff" here (and in a few
    other articles), which is very useful. Some manufacturers don't bother
    (or their website is so old that it's impossible to find if they do),
    which is very nice of them.
    Yes, and material, especially ferrite selection, is easily another short
    article's worth. Frequency, power, thermal, geometry, and maybe special
    purpose specs like tempco and linearity, etc.
    Samey samey -- (most?) manufacturers go on Bpk for a sine wave, which of
    course is the only condition under which this is strictly valid (a square
    wave will have, say, 10 or 20% higher losses, which you can calculate
    knowing the harmonics).
    Now, asymmetry and such, I've never seen data regarding that.

    Let's see. Assuming losses are predominantly eddy currents (i.e.,
    "classical Steinmetz model"), as it saturates, permeability drops, so skin
    depth goes up, which makes the particles look bigger, which should reduce
    eddy currents.

    Depending on how you're driving (constant deltaB, H, ???), the amount of
    B-H curve traveled may or may not change. Eyeballing a B-H curve, the
    opening shuts off towards saturation, but that's not necessarily an
    indication that hysteresis is, in fact, smaller in that region, especially
    for a small cycle rather than full loop excitation.

    Ferrites, at least, tend to get toastier towards saturation. I don't know
    if that's a nonlinear effect or just because, yeah, drive it harder and it
    gets hotter. Seems to me, manufacturers rarely provide losses in ferrite
    past 0.2T or so (i.e., not up to 0.3 or 0.4T depending on material and
    Tell that to anyone who's tried using a #26 core in a boost converter. ;)

    But as the point is being able to determine if it's right or wrong, anyone
    looking at a #26 core will be able to make that determination easily now.

    In a properly made component, copper and core are equally important. As I
    observed, that #26 core might only handle a few VA, but the copper might
    handle 80W, nothing to sneeze at. It takes the right application.

    Copper is probably more important than the core in RF chokes. For being a
    pile of dusty iron, mix #2 is surprisingly low loss: most points on the
    graph show a Q over 100. You'll be hard pressed to maintain a Q that high
    once you've put some pesky wire around it, especially the sheer amount you
    need to get a useful inductance from such a low permeability core. But
    then, I've got a power transfer mindset in that statement -- even at a few
    MHz, I need more inductance than an RF final at 20MHz does, which puts
    bigger demands on the copper, which either smothers the core or ruins the
    Q. Simple solution, buy a #8 or something higher permeability like
    that -- losses are higher, but you use less copper.

    Micrometals strikes again on the matter: they published typical curves of
    Q vs. typical cores and materials for actual windings. Hard to beat that.
    Yes, transformers have different limitations -- I went into some of this
    in an older article here,
    if you know how much copper you can get into the thing, and how much core
    you get (i.e., the quad product), and various other application-specific
    assumptions (resistivity, power loss, etc.), you can calculate the power
    handling capacity of that core in a typical application, allowing them to
    be ranked in much the same way as in this article, and a choice (or at
    least initial selection) made.

  17. Tim Williams

    Tim Williams Guest

    I have "pdfTeX, Version 3.1415926-2.3-1.40.12 (MiKTeX 2.9) (preloaded
    format=pdflatex 2012.12.9)" in the log file, to be painfully specific.

    LaTeX has quite a learning curve, but it can't be beaten for geeky, mathy
    publication-grade material, and it's damn powerful.
    Dunno, it's just \item to me :)

    Say, any idea how much the fonts account for in the file size?

    About 340k should be due to images, give or take compression methods.
    Geez, more than 30% of the total is just the choke picture... can hardly
    bear myself to dent its pixel count though, it's a very shiny picture...
    go ahead... zoom in on it... :p

  18. I ran mubusy on the file, and get these.. the pfa are all fonts, and
    png are images, obviously.

    10,501 AMUUOH+CMR17-0147.pfa
    5,341 HZCUSP+CMTT12-0159.pfa
    9,394 IDAFQO+CMR8-0149.pfa
    60,480 img-0027.png
    138,993 img-0051.png
    930,528 img-0064.png
    21,941 img-0075.png
    7,528 IWVHCE+CMMI6-0139.pfa
    7,319 JKSUUE+CMSY6-0155.pfa
    11,137 LEZPSX+CMMI8-0141.pfa
    12,328 OQFKAE+CMSS10-0151.pfa
    8,281 SEIURA+CMSY10-0153.pfa
    12,559 SWZFET+CMBX12-0135.pfa
    9,358 TVGLTB+CMR10-0143.pfa
    11,098 WCSXOI+CMBX10-0133.pfa
    7,409 WQDNZG+CMSY8-0157.pfa
    12,256 ZCLSJF+CMMI12-0137.pfa
    23,116 ZLAVVI+CMR12-0145.pfa

    This adds up to 1.3M vs. the actual 546k of your file, and doesn't
    include the actual text of your paper, but it doesn't look like the
    fonts account for much of the size. Maybe 10-20%.
    Shiny! Dui.

    Best regards,
    Spehro Pefhany
  19. Tim Williams

    Tim Williams Guest

    Ahh, that'd be the choke pic... it should be JPG internally, so your tool
    converted it to PNG somewhere I guess.. which obviously poofs it a bit.

    Good to know, thanks.

Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day