Author Topic: Transformer design: how to size a core?  (Read 7498 times)

0 Members and 1 Guest are viewing this topic.

Offline BukTopic starter

  • Contributor
  • Posts: 35
  • Country: gb
Re: Transformer design: how to size a core?
« Reply #25 on: October 15, 2021, 10:52:01 am »
Truly it's amazing stuff, amorphous/nanocrystalline materials; not only low loss, but some offer permeability rivaling supermalloy. 
The now-obsolete ISDN transformers, using the stuff, are just about unmatched in bandwidth. 
The cores are still around fortunately, so you can still make your own, just not off-the-shelf ready made transformers.

Your mention of "ISDN transformers" led me here. They have cores in the size range I'm looking for. Of course, they don't list prices, which is never a good sign :'(

That's quite extreme flux density for ferrite, are you sure your loss model is correct?

Not entirely. I'm talking (waiting for) a magneto-harmonic FEA guy to sanity check my model; but I have some confidence it is correct.

Also, how exactly are you going to get 150 total amp-turns through a 1/4" hole?

I have a whole 8mm to play with not 6.35mm :)

The picture of the model above shows 315 x 0.25mm turns in the secondary in a 10mm ID, with room to spare. Whether I can physically wind it is another matter; but I do have some history rewinding motor cores. It is slow, painstaking work, but 0.25mm is easy to deal with.

I'd like to see core loss more like 0.5W for a core that size, at least for continuous operation, let alone copper losses as well.

This is (maybe) where experience trumps best endevours. My target is burst operation;  say max. 30 seconds at a time. The body of the device can be a heatsink; though interior airflow could be a problem.

Never heard of FeNiZnV, what's that?

It is a material specification "soft iron ferrite" that comes pre-installed in the materials library of the FEA software. (Best reference I have is this. An Iron Nickel Zinc Vanadium nanoparticle ferrite produced using electric arc deposition.

I believe it may once have been market under the brand name 'Sinimax Softiron Softferrite". Very little reference to it.

I used it for the sim because the material specs were available. Trying to get the specs for commerical cores is hard because they tend to be proprietary mixes.

« Last Edit: October 15, 2021, 10:58:16 am by Buk »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 21606
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Transformer design: how to size a core?
« Reply #26 on: October 15, 2021, 07:16:25 pm »
Your mention of "ISDN transformers" led me here. They have cores in the size range I'm looking for. Of course, they don't list prices, which is never a good sign :'(

Well, they're also just a Chinese distributor (if not just window dressing for some bigger no-name distributor, who knows how many companies are at the same address?).  AFAIK, HMG and VAC are the only two sources of the stuff, though it may be China is making it now, I don't know.


Quote
I have a whole 8mm to play with not 6.35mm :)

The picture of the model above shows 315 x 0.25mm turns in the secondary in a 10mm ID, with room to spare. Whether I can physically wind it is another matter; but I do have some history rewinding motor cores. It is slow, painstaking work, but 0.25mm is easy to deal with.

The primary is the stinker here: if it's solid, it's all surface.  If it's stranded (litz) it'll take up way more space (about double?).

And you don't have much space to spare: at a typical 8 A/mm^2, the winding factor is a whopping 24%.  10% is recommended for toroids.  50% would be pushing it for a bobbin style winding, but you get more core.  I don't know how high you want to push the current density.  Maybe more is acceptable given the size and duty cycle (as below).

Also -- if your core losses actually are that low, it would be a valuable trade to reduce winding length, trading copper for core loss.

The secondary, bunched up like that, will also suffer from proximity effect; easy enough to reduce by interleaving the windings.


Quote
This is (maybe) where experience trumps best endevours. My target is burst operation;  say max. 30 seconds at a time. The body of the device can be a heatsink; though interior airflow could be a problem.

Ah, if the duty cycle is enforced to be low, that might not be so bad.  Put a thermistor in the middle?

Potting with a nice moderately conductive compound would also help for internal temperature, considering you'll have basically no airflow inside there.


Quote
I used it for the sim because the material specs were available. Trying to get the specs for commerical cores is hard because they tend to be proprietary mixes.

Any way to craft your own from curves?

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline BukTopic starter

  • Contributor
  • Posts: 35
  • Country: gb
Re: Transformer design: how to size a core?
« Reply #27 on: October 16, 2021, 07:30:20 am »
Well, they're also just a Chinese distributor (if not just window dressing for some bigger no-name distributor, who knows how many companies are at the same address?). 

Frankly I don't care where the stuff comes from if it does what its meant to do.  I read somewhere about testing cores by winding a couple identical coils, a few turns each of a known wire; feeding in AC sweep into one and comparing the output from the other on a scope. I don't remember the details; and I don't have a scope.

Quote
AFAIK, HMG and VAC are the only two sources of the stuff, though it may be China is making it now, I don't know.

I don't know those initials? 

The only companies that I know of producing metallic glass are MetGlas in the US and Hitachi in Japan and Hebei Shenke in China.

My understanding is that the first two 'merged' (Hitachi took MetGlas over) to counter the growing threat from the latter once it became clear that what they produce is of very high quality.
(

Quote
The primary is the stinker here: if it's solid, it's all surface.  If it's stranded (litz) it'll take up way more space (about double?).

And you don't have much space to spare: at a typical 8 A/mm^2, the winding factor is a whopping 24%.  10% is recommended for toroids.  50% would be pushing it for a bobbin style winding, but you get more core.  I don't know how high you want to push the current density.  Maybe more is acceptable given the size and duty cycle (as below).

For my own reassurance, I clipped a 1/2"OD x 1/4"ID x 3/16" core of unknown type from a piece of scrap electronics; dug out a some scraps of used 0.5mm and 0.25mm wire and attempted a wind.
This is what I got in about 20 minutes:


The primary is 5 turns of 4 parallel 0.5mm to get 1.0mmØ equivalent. The secondary is 120 turns (ish, I may have lost count a couple of times) of 0.25mm.

Not the neatest coil you'll ever see, but given new rather than pre-kinked wire; a somewhat larger core; and the incentive to spend more time to get it neater, and I think I can make a better job.

And with a little adaption,

Quote
Quote
This is (maybe) where experience trumps best endevours.

Also -- if your core losses actually are that low, it would be a valuable trade to reduce winding length, trading copper for core loss.
The secondary, bunched up like that, will also suffer from proximity effect; easy enough to reduce by interleaving the windings.

Go find your 'smug face'... I'll wait.

You were right!

Whilst my model is correct, the material definition has a subtle error that means my simulation runs are effectively producing garbage numbers. It'll take me a while to sort that out.

Quote
Quote
My target is burst operation;  say max. 30 seconds at a time. The body of the device can be a heatsink; though interior airflow could be a problem.
Ah, if the duty cycle is enforced to be low, that might not be so bad.  Put a thermistor in the middle?
Potting with a nice moderately conductive compound would also help for internal temperature, considering you'll have basically no airflow inside there.

Both good ideas. Thanks.

Quote
Quote
I used it for the sim because the material specs were available. Trying to get the specs for commerical cores is hard because they tend to be proprietary mixes.
Any way to craft your own from curves?

Yes, but extracting good data from those stupidly small graphs in the datasheets is very hard.

When I was creating material definitions for MetGlas & Hitachi foils, I had to be very pursuasive before they sent me the raw data produced by these guys who do their certification testing for them.
« Last Edit: October 16, 2021, 07:55:14 am by Buk »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 21606
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Transformer design: how to size a core?
« Reply #28 on: October 16, 2021, 08:35:07 am »
I don't know those initials? 

Sorry, Hitachi Metals Group and Vacuumschmelze.  The distributors are... neither here nor there, probably, just that the way Chinese distributors tend to work in general: there might be dozens of seemingly the same or very similar company(s), and you have to wonder which ones are representing prices fairly for example, let alone the providence of their products.  Maybe that's overly paranoid, I don't know.

Heh, on that note, at least amorphous type stuff is very hard to fake.  You won't get very far into production before finding out you've got nothing but transformer iron in there!


Quote
The only companies that I know of producing metallic glass are MetGlas in the US and Hitachi in Japan and Hebei Shenke in China.

Huh, searching on a couple keywords, this only turns up under another domain (which might not mean anything, or even that Google is being terrible again at finding real results). Not really much data there, at a glance, but I also don't read Chinese and the auto translation isn't impressive, so, eh.

If true, that's cool -- always good to have more sources of things!


Quote
For my own reassurance, I clipped a 1/2"OD x 1/4"ID x 3/16" core of unknown type from a piece of scrap electronics; dug out a some scraps of used 0.5mm and 0.25mm wire and attempted a wind.
This is what I got in about 20 minutes:
[...]

The primary is 5 turns of 4 parallel 0.5mm to get 1.0mmØ equivalent. The secondary is 120 turns (ish, I may have lost count a couple of times) of 0.25mm.

Not the neatest coil you'll ever see, but given new rather than pre-kinked wire; a somewhat larger core; and the incentive to spend more time to get it neater, and I think I can make a better job.

Yeah I mean it fits, you can pull off a winding factor of 50% -- but it's going to look like, well, that; and all that crowding increases proximity effect, while the turn length greatly increases as well.  So the efficiency falls; the price of compactness.

Quote
Whilst my model is correct, the material definition has a subtle error that means my simulation runs are effectively producing garbage numbers. It'll take me a while to sort that out.

I mean, no need for smugness, just didn't pass the sniff test -- compare with typical power and B-H curves, ferrites just don't go that high for one and if they do it's going to be many watts (I think?), you're deep into saturation there.  Checking models is a key part of simulation!

Uh, maybe saturation anyway?  Is that Bpk, or like, p-p or RMS or something?  It's usually Bpk, but better to ask to be sure.


Quote
Yes, but extracting good data from those stupidly small graphs in the datasheets is very hard.

When I was creating material definitions for MetGlas & Hitachi foils, I had to be very pursuasive before they sent me the raw data produced by these guys who do their certification testing for them.



But that aside, that's cool!

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline BukTopic starter

  • Contributor
  • Posts: 35
  • Country: gb
Re: Transformer design: how to size a core?
« Reply #29 on: October 16, 2021, 09:46:07 am »
Any way to craft your own from curves?

Turns out that TDK make the raw data available for a whole bunch of their core materials.

That makes life *much* easier.
 
The following users thanked this post: T3sl4co1l

Offline BukTopic starter

  • Contributor
  • Posts: 35
  • Country: gb
Re: Transformer design: how to size a core?
« Reply #30 on: October 16, 2021, 07:48:57 pm »
Yeah I mean it fits, you can pull off a winding factor of 50% -- but it's going to look like, well, that; and all that crowding increases proximity effect, while the turn length greatly increases as well.  So the efficiency falls; the price of compactness.

You might find this: Designing Low-Power Flyback Inductors Using Tiny Toroids (Part 1): A Boost Converter Application

and this: Designing Low-Power Flyback Inductors Using Tiny Toroids (Part 2): Calculating Losses

interesting; or not.
« Last Edit: October 17, 2021, 07:22:48 am by Buk »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 21606
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Transformer design: how to size a core?
« Reply #31 on: October 16, 2021, 08:49:58 pm »
What a strange article; lambda is used for flux, the application is not introduced so there's no motivation or direction behind the subject, and no mention is made of a suitable controller -- the most particular it gets is naming a 2N7000, a TO-92 device of all things.  And he goes directly to #26 powdered iron, literally the worst choice available, without explanation.  This feels so... vintage?  Yet the copyright is proudly 2019.

I think I get the idea of what's being done there -- it's hard to review because it's all over the place, the core selection is unexplained, then the core loss flux density formula is given, then the power loss is given -- it's like it's completely in reverse logical order.

So in the end he's got 8uH Lp, rated for 210mA.  Sounds very much like one of these:
https://www.coilcraft.com/en-us/products/power/coupled-inductors/1-n-shielded-coupled/lpr/lpr4012/lpr4012-103d/
which is a smaller outline at a tiny fraction of total height.  They also have core loss calculations on site; though I'm not sure if the coupled inductors are supported?

Inputting similar numbers, Micrometals recommends very far from #26; T37-6 and T30-2 are typical examples, giving excellent low losses (43, 35mW), though mind that these are single-winding inductor figures, and the copper losses would be about 40% higher in a flyback transformer.

To his credit, the core loss calculations seem consistent with what Micrometals' calculator says.  It would seem the inexplicable core choice, or lack of motivation, is the only real problem; the calculated results are sound. :-+

Tim
« Last Edit: October 16, 2021, 08:52:56 pm by T3sl4co1l »
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline BukTopic starter

  • Contributor
  • Posts: 35
  • Country: gb
Re: Transformer design: how to size a core?
« Reply #32 on: October 17, 2021, 07:22:01 am »
What a strange article; lambda is used for flux, the application is not introduced so there's no motivation or direction behind the subject, and no mention is made of a suitable controller -- the most particular it gets is naming a 2N7000, a TO-92 device of all things.  And he goes directly to #26 powdered iron, literally the worst choice available, without explanation.  This feels so... vintage?  Yet the copyright is proudly 2019.

I don't see any reference to "2N7000"  or "TO-92" in either of the articles I linked? (Sorry one of the links was broken, I'll correct it shortly.)

Quote
I think I get the idea of what's being done there -- it's hard to review because it's all over the place, the core selection is unexplained, then the core loss flux density formula is given, then the power loss is given -- it's like it's completely in reverse logical order.

So in the end he's got 8uH Lp, rated for 210mA.  Sounds very much like one of these:
https://www.coilcraft.com/en-us/products/power/coupled-inductors/1-n-shielded-coupled/lpr/lpr4012/lpr4012-103d/
which is a smaller outline at a tiny fraction of total height.  They also have core loss calculations on site; though I'm not sure if the coupled inductors are supported?

Inputting similar numbers, Micrometals recommends very far from #26; T37-6 and T30-2 are typical examples, giving excellent low losses (43, 35mW), though mind that these are single-winding inductor figures, and the copper losses would be about 40% higher in a flyback transformer.

I can't argue with your logic; but 3 observations:
  • I suspect the choice of core for the example was deliberately obscure by way of demonstrating that it could be done, not necessarily that is was a good solution to anything in particular

    In support of that suggestion, I offer another article of his which might clarify some of the motivation:
    Single-Bundle Windings Make It Easier To Build Custom Magnetics In-House
  • I haven't seen the references to TO-92 packaged transistors you saw; but when it comes to electronics, I like vintage. Components big enough for me to see with my tired old eyes; recognisable  component values; no need to for an STM, nor the dexterity of an eye surgeon, to fit (and remove) from boards. You can  keep your smt ...  :P
  • I think the guy likes to explore the obscure and forgotten topologies; and to prototype. (Which makes him useful to me for this.)

(BTW: I got really excited about this, before I realised I'd need a stack of 20 in parallel :palm: )

Quote
To his credit, the core loss calculations seem consistent with what Micrometals' calculator says.  It would seem the inexplicable core choice, or lack of motivation, is the only real problem; the calculated results are sound. :-+

That's good to hear.
« Last Edit: October 17, 2021, 08:30:25 am by Buk »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 21606
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Transformer design: how to size a core?
« Reply #33 on: October 17, 2021, 09:16:34 am »
The paragraph on page 3, "For a small converter, a small MOSFET switch such as a 2N7000 is used. ..."

It's interesting, subtle; merely change a single digit and it seems current enough (2N7002 is the SOT-23 version, still very much in use for jellybean purposes, alongside BSS138, BSS84 and others).  Just as well, it could mean the "original", generic part, and there's not actually anything to read into.  But so too, the choice of #26 (and yeah, a possible motivation is to "make do" with it).  We can only speculate, short of e-mailing the guy, heh.

I saw the reference to, but didn't look at, the "single bundle" article.  I don't have a problem with that, at least from what I gather from its use in the first document.  Indeed, I've done that plenty of times myself, and actually I'd recommend you do it as well, if you can.  The biggest benefit is: getting excellent transmission line transformer type coupling between windings.  Especially so, if you pair primary and secondary wires (as pairs or quads), and then bunch up however many it is you need to wire up in series and parallel, for primary and secondary.  The biggest downside is: you have to painstakingly ohm out every single goddamned wire, to make sure you're getting them all properly in series and parallel..!  (Even with color coding, you'll only have so many colors to choose from, but for your case, dozens of strands to cover, so...)  But yeah, fine for one-offs.

BTW, if you're curious about off-the-shelf possibilities, shop around for inverter transformers.  Nothing wrong with using a, say, 240V to 5V, flyback transformer in reverse!  Or maybe a forward type would be more suitable (do you really know what inductance you need, as yet?), or multi-secondary so you can wire them for a CT'd primary, etc.  They'll likely be bulkier than what you may be looking for, on account of the squarer shapes (mostly ferrite E-core styles), bobbins with rows of pins, and reinforced insulation, and they tend to be expensive (in part just because you're buying tiny quantities through a distributor, prices are kind of irrelevant -- but anyway, the alternative is your labor, potentially a big savings?).

CCFL transformers are also interesting, but probably much too high a ratio for what you want, and they only come in so many sizes.  I guess if you wanted to rewind one you could.  They tend to have a low profile and high aspect ratio, so might still fit inside your cell-sized enclosure, and I think the power level is about right, give or take size selection that is.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf