I'm waiting on getting the assembly details from the contract manufacturer because the engineer that designed it <ahem> only specified the number of turns, RMS current for each winding, frequency, core part number, and total gap. He relied heavily on the wisdom/experience of the contract manufacturer to fill in all the pesky details, including what needed to be done to get through UL. Which I am somewhat sympathetic to - it is very difficult keeping track of all the regulations and directives here - but you need to have at least some idea of what is required otherwise you might find out you only have 30mm of winding width available instead of 36mm, etc.
Ah, hm... yes, can be tricky.
Reminded of the one transformer design journey I had -- an oddball transformer, knew it was going to be something like, toroidal, stripwound, GOSS, water cooling an option. 1-4kHz I think, 20kVA, modest ratio like 10 or 20:1, with a 650V primary. Got responses ranging from no-quote to some meter-wide monstrosity to something way too small (~8" across?).
It seems magnetic design knowledge is a bit rarefied, even for those in the business of designing and producing them. A lot of engineers do encounter it (or if not by magnetics directly, then through good old EMC, say), but it seems few have the knack for it.
We ended up going with a place I think in Connecticut, which quoted a very reasonable design, something like a 10" o.d. core, cooling pipe wrapped around it one turn, reasonable Bpk given the frequency, voltage and number of turns, and the overall result was something like 5-6" tall, 12" wide, and cost $750/ea or something like that. Didn't have any problems with it, didn't seem to get hot in operation. As far as I know, it's still cooking away, down in Los Alamos (*what* they're "cooking", I'll uh, leave to your imagination
).
Did notice some quirks with it, which is to be expected from steel, and how we were driving it (H bridge). It seemed to "flux walk" on its own, despite the series coupling capacitor -- it couldn't possibly be walking due to leakage, the capacitor would have to be abysmal. Nor due to duty cycle shifting in operation, it was a digital control. In any case, the behavior was apparent by ear -- you'd turn it on and hear a chirp as it finds its operating frequency, and then the whine intensifies as the waveform goes asymmetrical, just barely saturating to one direction. I think there's some weird spontaneous magnetization stuff going on there.
It was history-dependent, too: once a transformer had been hard saturated (say due to startup transient), it would move more aggressively towards saturation. Also, always the same direction towards saturation. (Perhaps if it were hard-saturated in the opposite direction it would flip, I don't know. Or, heck, maybe I did test that, I don't remember; this was ~7 years ago.)
We ended up with a dummy resistor across the coupling capacitor, to dampen startup transients. The tiny bias (timing imbalance * supply voltage / dampening resistor) didn't affect the core, and a virgin transformer behaved perfectly (quiet operation, no slow walk towards angry whine).
Ah, but I digress!
Nope - it's a big transformer and uses actual copper strip for the windings. Insulation is... aramid tape? And no bobbin of course - planar core - so plenty of issues with clearance and creepage to "dead metal" (ie - the core itself).
Hmm, I'm thinking about a similar design for a new 10kW induction heater design; it would be 3 x E100/60/28 cores, and the windings would be water-jetted from copper sheet. The secondary will be, basically, a stack of split washers, shaped to fit around the core of course, and spaced to fit the primary inbetween. The secondary terminals (which have to carry 1-2kA) will be parallel copper bars with cooling pipes; the bar ends would be notched to accept the turns, which get soldered in.
I don't see any obvious reason why the current would distribute poorly; more that the current gets very crowded at the end of each turn, both because the metal is turning away from the primary (so all the current flow bunches up along the inside corner) and because the current flow is being turned, from a flat-plane geometry, to a perpendicular, parallel-plate geometry. As long as the primary turns are the same between each layer, though, the only secondary currents that should flow are the image currents of the primary. (The primary will have several taps for impedance matching, of course only the active turns carry current and therefore the unused part just idles with whatever residual eddy currents they end up with.)
Another way to think of it is, suppose I started with a super thick plate, and wrapped that around a core. Then I cut slits crosswise into the bend, making a stack of plates somewhat like a
dough cutter. And into those slits, I place the primary winding.
Now, with such a construction, I do still have reasonably smooth current flow, in that it goes from the bar straight into the stack-of-plates section. But in a PCB construction, it may be exacerbated.
In a PCB, you usually make the connections with thru-vias/pads, likely reinforced with solid wire, but still the current crowds up near the connection point, which is off to one side. And if the pads are surrounded by other rigid connections, the current may be forced to flow over each layer, crowding up multiple times, before reaching the circuit.
That could make a difference in EMF as well, as there's more voltage drop (some resistive, hence the power, but mostly inductive) to one side of the stack-of-plates, and there's your eddy currents or however you might like to phrase it. (Really, eddy currents, proximity effect, current crowding, skin effect and etc. all become somewhat ambiguous or equivalent in a complicated structure like this.) In that case, yeah, the only thing you can do is keep the stack low, and make it wider and wider.
Which isn't that bad an idea, just using more transformers overall, in parallel. Depends. It doesn't scale well, is probably the biggest problem. Different mounting methods could be handy (planar transformer sticks up vertically from the backplane; a bunch sit in rows, all tied together?).
Or taking that further and using multiple converters in parallel, which can be advantageous (lower input/output ripple?) but also very annoying (N times more components, assemblies..).
I have had terrible experiences with massive circulating currents when I've tried to parallel strip windings before to get around the width vs. thickness dilemma so I hesitate to go that route again. Also, this is a resonant mode converter so there is more need than usual to control the leakage inductance and distributed capacitance.
(Which I've addressed above, but didn't want to interrupt the segue with this quote.
)
Ah yes, leakage too; although, if leakage is a useful part of the circuit, you can at least handle some, which gives a lot less pressure towards super-low-Z planar designs, and gives you a lot more leeway to use magnetically-spongy litz cable.
You might still be limited to less leakage than a single layer of full-size litz offers, but you could kinda hybridize by using multifilar litz, or multiple windings or layers in parallel.
Also, exiting the bobbin of a more conventional windup ("shell" style, your average ER and such design) is a big pain with wide conductors, foil especially. But, eh, that just happens. The bigger pain I guess is going to be ensuring adequate insulation along its path -- which further reduces the available winding area. (Hm, the winding factor of PCB material doesn't look bad at all when you need big currents like that.)
Or if you were thinking about litz for everything, then, that too; amps are amps, you can just collect cables a bit better than you can fold up a hunk of metal is all.
Oh -- one thing that may be helpful: these guys make a (patented??) flat litz of sorts, that's apparently as good as foil, without the eddy currents / proximity effect.
https://www.wcmagnetics.com/Got prototypes from them once; didn't move into production, but it seems like they knew their stuff.
Would 2 years have been enough time to optimize things???
Heh, well, with competent engineers, perhaps even less than that? But, therein lies the problem, and as I alluded to before, magnetism seems to be enigmatic to many.
I'm getting a crash course in just how miserable your life can become if you stray from using UL Recognized "electrical insulation systems" - that is, an approved combination of wire insulation, optional margin/layer insulation, and varnish and/or potting - but sleeving standard Litz with UL Recognized sleeving (e.g. - Ex-Flex, Silflex, Varglass, etc...) is something I've been considering. It will worsen both leakage inductance and thermal resistance, though, so it's definitely a last resort.
Yeah, I can see that being a big problem... if it's not perfectly to spec, it's basically something like: "*throw hands up*, well, you've got to subject it to this and that and this other test, plus regular inspection / retests, and..."
On the upside, fiberglass sleeving should fill nicely with varnish, for certain values of sleeving. If that's acceptable as a "cemented joint", that should do. Or failing that, potting. Hm, maybe a partial-discharge test, to verify no voids..? (I wonder where those procedures sit, on the fence between UR and "whelp".)
Tim