EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: TimNJ on March 26, 2021, 03:21:06 pm
-
Greetings,
Long post incoming. Even if you just have small bits of feedback, I'd love to hear it.
Many months ago, I designed a flyback transformer based on 100W nominal design, using a POT33 sized core. After finally getting around to testing the total design, the efficiency is very good, over 90% on average for the entire power supply. The control IC is a quasi-resonant current mode type.
But, for other reasons, I need to modify the transformer to a higher turns ratio (from 6:1 to ~7.5:1). Basically, due to certain constraints with the secondary synchronous rectifier I need to reduce the reflected voltage stress on the secondary. (Additionally, I have some radiated EMI issues that I wonder if have any relationship to the turn-on voltage on the primary MOSFET. Higher turns ratio would reduce the turn-on voltage.
1. Interleaving Strategy:
My original transformer design is sandwich/interleaved construction, (P - S - P). Everything is a single layer to keep the leakage inductance low. Due to concerns about proximity effect losses near the gap, I made the primary "half" near the gap litz wire, and the primary "half" farthest from the gap normal enameled wire.
An odd thing I did was to divide the primary into 1/3 (10T litz) and 2/3 (20T normal wire). I'm not sure why I did this!! (I may have run some calculations on copper loss and found this as a good solution.)
My first question is: Is there any obvious disadvantage to splitting the primary in this way? Would you expect a 50/50 split to be better for some reason? What might EMI implications be?
2. Air-Gap Fringing Flux Screen
The new proposed transformer will have a lower magnetic flux density as it will have higher primary turns count. The gap length will be larger, based on new target inductance. If the flux density was held constant, then a large gap would mean more fringing in the windings, but I think since the flux density is lower, that the fringing flux will also be lower, so a larger gap may not be problematic.
Even so, I did some Googling and found some research around an open-circuit "flux screen" that encompasses the gap (first layer in the bobbin). For example: https://pureportal.strath.ac.uk/en/publications/air-gap-fringing-flux-reduction-in-inductors-using-open-circuit-c. They found you needed quite a thick screen (about 1.0mm!) to be effective, but they were also using it on a 100A RMS inductor. Maybe the thickness scales down too. (I have about 20x less current.)
Has anyone used this type of screen with any success? People also suggest that this screen may be beneficial for EMI, specifically radiated EMI. How so? Usually I think of the center gap as well shielded by the rest of the inductor/transformer's windings.
You can see my original transformer design attached. And diagram showing fringing flux with/without flux screen attached.
Thanks!
-
That's a "long" post? You disappoint me... :-DD
I think the asymmetrical primary sections would reduce coupling on the order of 17%, and by percentage, I mean in the progression from symmetrical PSP to fully asymmetrical SP. So, not much, but enough that you may need to review the primary snubber or something.
Note that Litz -- by definition, on account of its transparency to lengthwise magnetic fields (what would otherwise cause eddy currents in solid wire) -- also has higher leakage inductance. This becomes especially significant at high currents, where Litz can have a huge improvement in current density, but also where few sections of wire windings are undesirable in the first place due to the leakage requirements. That is, if you're driving some low voltage and high current, you need much lower leakage than, say, a PSP stackup can deliver -- more sections of thinner wires in parallel, or foil, are the way to go.
This shouldn't be a problem with thin stuff like that, it probably amounts to some percent of total leakage -- but it is something to keep in mind.
Heh, and as weird sectioning schemes go, if you phase the primary sections so that the foil shield faces the common end of the primary winding, you could use the shield itself as one extra turn for the primary. It's being grounded to DC+ after all. :)
As for the flux screen: it doesn't impress me..? It's going to get pretty damn hot in there. It can definitely reduce losses in the windings per se, but I don't know how much it's going to save overall. I'm inclined to suspect it makes it worse overall, because it guarantees intercepting all that divergent flux, whereas the wire winding has a reduced cross section.
Note the test frequency, they had to use a thick shield because of that. At more ordinary switching frequencies, PCB stock actually does a very good job, for example (which is also encouraging when you need shielding between noisy and quiet things in a multilayer PCB design).
If you need absolute lowest eddy current losses, I would suggest a shaped gap core if you can find/make one; or a slightly bigger core and a padded bobbin; or a distributed gap material -- you can get powdered iron shapes, though they aren't common, and you may have to order a lot at once or something. Or actually, they're probably on Ali, everything is... Or just use a toroid, assuming the winding cost is manageable.
Tim
-
That's a "long" post? You disappoint me... :-DD
I think the asymmetrical primary sections would reduce coupling on the order of 17%, and by percentage, I mean in the progression from symmetrical PSP to fully asymmetrical SP. So, not much, but enough that you may need to review the primary snubber or something.
Note that Litz -- by definition, on account of its transparency to lengthwise magnetic fields (what would otherwise cause eddy currents in solid wire) -- also has higher leakage inductance. This becomes especially significant at high currents, where Litz can have a huge improvement in current density, but also where few sections of wire windings are undesirable in the first place due to the leakage requirements. That is, if you're driving some low voltage and high current, you need much lower leakage than, say, a PSP stackup can deliver -- more sections of thinner wires in parallel, or foil, are the way to go.
This shouldn't be a problem with thin stuff like that, it probably amounts to some percent of total leakage -- but it is something to keep in mind.
Heh, and as weird sectioning schemes go, if you phase the primary sections so that the foil shield faces the common end of the primary winding, you could use the shield itself as one extra turn for the primary. It's being grounded to DC+ after all. :)
As for the flux screen: it doesn't impress me..? It's going to get pretty damn hot in there. It can definitely reduce losses in the windings per se, but I don't know how much it's going to save overall. I'm inclined to suspect it makes it worse overall, because it guarantees intercepting all that divergent flux, whereas the wire winding has a reduced cross section.
Note the test frequency, they had to use a thick shield because of that. At more ordinary switching frequencies, PCB stock actually does a very good job, for example (which is also encouraging when you need shielding between noisy and quiet things in a multilayer PCB design).
If you need absolute lowest eddy current losses, I would suggest a shaped gap core if you can find/make one; or a slightly bigger core and a padded bobbin; or a distributed gap material -- you can get powdered iron shapes, though they aren't common, and you may have to order a lot at once or something. Or actually, they're probably on Ali, everything is... Or just use a toroid, assuming the winding cost is manageable.
Tim
Oh no, I think I just disrespected the long post king himself. In my head, it felt like a much longer post than it is.
Thanks for your input. If the assymetrical sections does cause some increase in leakage, maybe I will see if I can make it more symmetrical on the next revision. A couple of sources indicate that leakage inductance increases with total turns, which I intuitively don't really understand. I would figure it is mostly due to layer-to-layer distance and winding breadth. But, as I think I need to increase primary turns on the next revision, perhaps I can counter this increase in leakage inductance by making the P-S-P structure more symmetrical (i.e. 50/50 or close).
What advantage is there in using the foil as an extra turn? Better space utilization, more copper devoted to carrying current?
Regarding the flux screen, thanks. I too wondered what the eddy current situation looked like in the screen. The paper still suggests overall loss improvement. I wonder if the situation is helped if you use multiple (insulated) layers of thinner foil material? Like 4 layers of cuffed shield, open-circuit, un-terminated...eddy current situation better?
I have a little extra bobbin winding height left. I could afford to put some margin tape on the inner surface of the bobbin. Is this generally an okay idea without other drawbacks? Other than theoretical lost bobbin utilization.
Thanks!
-
Thanks for your input. If the assymetrical sections does cause some increase in leakage, maybe I will see if I can make it more symmetrical on the next revision. A couple of sources indicate that leakage inductance increases with total turns, which I intuitively don't really understand. I would figure it is mostly due to layer-to-layer distance and winding breadth. But, as I think I need to increase primary turns on the next revision, perhaps I can counter this increase in leakage inductance by making the P-S-P structure more symmetrical (i.e. 50/50 or close).
Well, width matters in that, a wider winding will have more turns, so has quadratically more magnetizing inductance, but only linearly more leakage. So k goes up, but LL goes up too.
Leakage mainly depends on wire length, and is compounded when a given winding/section is made of several layers.
For single layer sections, of equal turns and wire size, the equivalent geometry is simply a parallel wire transmission line, wrapped up edgewise. The characteristic impedance is some Zo, with inductance and capacitance per length, Lo and Co, such that sqrt(Lo / Co) = Zo. And velocity factor given by dielectric constant (and mu_r = 1 in the space between windings so we don't have to take account of that*), and Lo and Co are relative to mu_0 and e_0 by all these parameters (dimensional analysis suffices to relate them). By taking the wire winding length times these parameters, we arrive very close to the parameters needed to fill out this model: https://www.seventransistorlabs.com/Images/XfmrEquiv.png (https://www.seventransistorlabs.com/Images/XfmrEquiv.png)
There is some coupling from turn to turn (the field from a given turn, intersects adjacent turns), so it's not an ideal (non-dispersive) transmission line; it's a small enough amount that we don't mind it. We're rarely using transformers high enough to be concerned about above-cutoff response -- just the LF equivalent parameters will do.
*I suppose it's interesting to note that a multiwinding molded-powder/composite inductor might have some goo between windings. Not sure, I haven't taken apart very many to see if they fill that well or not.
*Also consequential to this, and something you can experimentally verify: in a well coupled winding, LL has very little if any dependence on the core. Short the secondary and measure Lpri(sc), while moving the core in and out of the bobbin. This proves the leakage flux is almost entirely between the windings themselves.
Which also gives us the insight that, for looser coupled designs (e.g., bank wound / split bobbin), we do expect the core to have some effect; and for very loose (e.g., bobbins on separate legs of the core), we expect a dominant effect! That just means it's harder to estimate leakage of such arrangements; which I guess is relevant when trying to design with CMCs in unintended applications. But they often give LL, or DM impedance or attenuation or whatever, which is even better.
Anyway, if we make a multilayer winding, it's transmission lines with respect to themselves, weird, right? So, the impedance is much higher (intuitively, for a very fast step: at least layer count * Zo), and dispersion is even weirder (depending on how the layers are wound, start/finish to the left/right..), and obviously there's less coupling to the secondary just on the fact that any given wire is surrounded by less of it.
I'm not sure offhand what effect dispersion has; I want to say it's effectively Zo rising with frequency, and whatever weird things that does to group and phase velocity. The helical waveguide (i.e., a single layer solenoid near resonance) has overall Zo much higher than expected from wire diameter and spacing (Zo between turns, as it were), and has non-harmonic resonant modes; I think this is somewhat applicable here. Anyway, we're really uninterested in above-cutoff response of multilayer windings -- it's a mess, as you can see. :) Again, just knowing the LF parameters will suffice, so we only need estimate the Zo and length, or LL and Cp other ways.
What advantage is there in using the foil as an extra turn? Better space utilization, more copper devoted to carrying current?
Just saving one pin and one turn. Not much, hence the "heh". :)
Regarding the flux screen, thanks. I too wondered what the eddy current situation looked like in the screen. The paper still suggests overall loss improvement. I wonder if the situation is helped if you use multiple (insulated) layers of thinner foil material? Like 4 layers of cuffed shield, open-circuit, un-terminated...eddy current situation better?
Alas, I don't have good intuition for what shielding does to losses. Copper is a good reflector (diamagnetic at AC, when thick enough), but not an ideal one. I've seen Q factors in the range of 5-10 when "shorting" a transformer with wire or sheet. The inductance drops significantly (LL dominant), and whatever resistance is left acts in series with LL, along with ordinary coil losses, and that sum gives the overall Q.
And it's not that it's shorting out the whole fracking transformer, of course; only the radial / fringing component. I do have some intuition as far as what amount that is: the cross sectional area should be on the order of the cross section of the core, enlarged by the air gap distance, minus the cross section. So, say it's an EE core, center gapped, 10mm x 10mm, Ae = 100mm^2, and l_g = 1mm. Expand the center core by 1mm (or "on the order of": maybe half this, maybe some irrational ratio) on all sides, now it's a 12x12 mm square with rounded corners, and area near 144 mm^2. Minus the original, 44 mm^2.
And the area intercepted by the shield, drops with distance from the core as well, of course; so maybe at 1mm distance (i.e. thickness of bobbin wall), the effective fringe area is only 20 or 10 mm^2. Basically the expansion factor depends on distance from core to shield, as well as relative core dimensions.
And finally, that area of flux, is shorted out by the shield, so has a low Q (but not zero), and that acts to divert some flux into the main (Ae, or somewhat less because of crowding effects?) cross sectional area, which reduces inductivity (A_L). You end up with an equivalent circuit of magnetizing inductance in parallel with an L+R, where the L is the leakage from primary to shield (as eddy currents) and R is its resistance.
The resistance should be... I think close to the skin surface of the shield, times its dimensions facing the fringing area. So again a fringing area, but this time parallel to the shield, rather than parallel to the core face. It should have similar dimensions (again, give or take some factor). The eddy currents run one way (e.g. right hand rule) around the top of the fringing area, and returning the opposite way around the bottom. I don't think it matters that we can divide that same area into multiple eddies in parallel; it should suffice to take it as one whole? And that should be enough to calculate an actual resistance.
So, that shorted area might be small, and so the coupling factor is small (approximately the ratio of areas), and so we expect the leakage L to be quite large (some times the magnetizing inductance?), and the Q factor is.. whatever all that comes out to.
And we're still not done, because we somehow need to do the same analysis for round cylinders in the same location, and compare results. Good luck with that?...
So, just based on a cross sectional argument, and the potential for the wire to be thinner (or even Litz), it seems like it would be for the worse. But that still depends on a lot of things. Q will definitely be reduced from an ideal case (distributed gap), in both cases; figuring out how much for each, heck, you might just have to simulate or measure it.
One thing is for sure: the less fringe cross section the shield intercepts (or any wires), the higher the unloaded Q can be. The more that gets intercepted, the less pure-magnetizing-inductance there is in the total, and the more lossy-leakage-to-shield there is. We can measure this by the reduction in magnetizing inductance.
And this extends all the way to the entire core cross section, not just fringing -- say if we were using a single E piece as a flux concentrator, for induction heating of flat stock. Granted that the normal magnetizing path changes in this case (because we're not using a second E, or the air gap is variable).
So, hey, if you have some time to try different windings... I'd love to hear the result. :-+
I have a little extra bobbin winding height left. I could afford to put some margin tape on the inner surface of the bobbin. Is this generally an okay idea without other drawbacks? Other than theoretical lost bobbin utilization.
Yeah, that's probably fine. The drawback is, there will be fewer turns that are still on the bottom layer (beside the margin tape).
Though, if the primary is wound up over the margin tape as well as beside it, and the shield and secondary all conform to this uneven base, that will work just fine. Lumpy, hard to wind flat and smooth, but if it works, it works.
Even if the leakage ends up somewhat higher, maybe it's still acceptable -- it's possible the increased loss (in snubbing the leakage) ends up less overall than with the eddy current loss.
Or you can lay down margin over the whole base, effectively increasing the bobbin wall thickness. Mind, you do have to balance against the outer core gap, if this is a shimmed gap; in that case, a somewhat peanut-shaped winding area would be best.
Oh, by the way, speaking of leakage around shields -- note that a given wire over a shield, acts like a round wire microstrip transmission line. And same thing on the opposite side, so that the leakage is given by the sum of both transmission line characteristics. This shouldn't be much different from the parallel-wire case, as the shield is a symmetry plane there. (The fact that the shield has a gap in it, does mean that image currents, under a given turn, are prevented from crossing, and have to loop around; but this is as intended -- as, if not for that, it wouldn't be much of a shield, would it? Also, if one winding doesn't cover the full width of the shield, it will act to spread that current out, with some loss incurred, as that current needs to be re-conducted, if you will -- this may be another tradeoff between LL and copper loss.)
Tim
-
Thanks for your input. If the assymetrical sections does cause some increase in leakage, maybe I will see if I can make it more symmetrical on the next revision. A couple of sources indicate that leakage inductance increases with total turns, which I intuitively don't really understand. I would figure it is mostly due to layer-to-layer distance and winding breadth. But, as I think I need to increase primary turns on the next revision, perhaps I can counter this increase in leakage inductance by making the P-S-P structure more symmetrical (i.e. 50/50 or close).
Well, width matters in that, a wider winding will have more turns, so has quadratically more magnetizing inductance, but only linearly more leakage. So k goes up, but LL goes up too.
Leakage mainly depends on wire length, and is compounded when a given winding/section is made of several layers.
For single layer sections, of equal turns and wire size, the equivalent geometry is simply a parallel wire transmission line, wrapped up edgewise. The characteristic impedance is some Zo, with inductance and capacitance per length, Lo and Co, such that sqrt(Lo / Co) = Zo. And velocity factor given by dielectric constant (and mu_r = 1 in the space between windings so we don't have to take account of that*), and Lo and Co are relative to mu_0 and e_0 by all these parameters (dimensional analysis suffices to relate them). By taking the wire winding length times these parameters, we arrive very close to the parameters needed to fill out this model: https://www.seventransistorlabs.com/Images/XfmrEquiv.png (https://www.seventransistorlabs.com/Images/XfmrEquiv.png)
Interesting. So does leakage inductance actually depend on magnetizing inductance? Let's just say you keep all windings exact the same, and just adust the gap dimension to adjust the magnetizing inductance. Do we expect leakage inductance to also change? My understanding (from what you say) is that Lleak does not directly depend on Lmag, but rather that when you have a higher Lmag, you usually have more turns, and thus the leakage inductance is also higher.
In my case, if I increase the number of primary turns, I expect leakage inductance to increase. But, if I kept the winding structures the same and re-gapped, I'd expect leakage to remain the same.
Regarding the flux screen, thanks. I too wondered what the eddy current situation looked like in the screen. The paper still suggests overall loss improvement. I wonder if the situation is helped if you use multiple (insulated) layers of thinner foil material? Like 4 layers of cuffed shield, open-circuit, un-terminated...eddy current situation better?
Alas, I don't have good intuition for what shielding does to losses. Copper is a good reflector (diamagnetic at AC, when thick enough), but not an ideal one. I've seen Q factors in the range of 5-10 when "shorting" a transformer with wire or sheet. The inductance drops significantly (LL dominant), and whatever resistance is left acts in series with LL, along with ordinary coil losses, and that sum gives the overall Q.
....
So, just based on a cross sectional argument, and the potential for the wire to be thinner (or even Litz), it seems like it would be for the worse. But that still depends on a lot of things. Q will definitely be reduced from an ideal case (distributed gap), in both cases; figuring out how much for each, heck, you might just have to simulate or measure it.
One thing is for sure: the less fringe cross section the shield intercepts (or any wires), the higher the unloaded Q can be. The more that gets intercepted, the less pure-magnetizing-inductance there is in the total, and the more lossy-leakage-to-shield there is. We can measure this by the reduction in magnetizing inductance.
And this extends all the way to the entire core cross section, not just fringing -- say if we were using a single E piece as a flux concentrator, for induction heating of flat stock. Granted that the normal magnetizing path changes in this case (because we're not using a second E, or the air gap is variable).
So, hey, if you have some time to try different windings... I'd love to hear the result. :-+
Tim
You'll have to excuse me. -- My little brain is working hard to try to comprehend what you're saying here. Regarding your line "So, just based on a cross sectional argument, and the potential for the wire to be thinner (or even Litz)...", are you talking about using an open-circuit, unterminated wire-structure shield instead of say, a large copper strip wrapped around the center-leg?
Paper says they reduced losses by increasing what percentage of the bobbin width the shield covered. Though, beyond a certain percentage, the improvement was not much. But, maybe it's a different story for a transformer compared to a plain inductor? You seem to suggest there will be some leakage between primary 1/2 and air-gap shield.
I have a little extra bobbin winding height left. I could afford to put some margin tape on the inner surface of the bobbin. Is this generally an okay idea without other drawbacks? Other than theoretical lost bobbin utilization.
Yeah, that's probably fine. The drawback is, there will be fewer turns that are still on the bottom layer (beside the margin tape).
Though, if the primary is wound up over the margin tape as well as beside it, and the shield and secondary all conform to this uneven base, that will work just fine. Lumpy, hard to wind flat and smooth, but if it works, it works.
Even if the leakage ends up somewhat higher, maybe it's still acceptable -- it's possible the increased loss (in snubbing the leakage) ends up less overall than with the eddy current loss.
Or you can lay down margin over the whole base, effectively increasing the bobbin wall thickness. Mind, you do have to balance against the outer core gap, if this is a shimmed gap; in that case, a somewhat peanut-shaped winding area would be best.
Oh, by the way, speaking of leakage around shields -- note that a given wire over a shield, acts like a round wire microstrip transmission line. And same thing on the opposite side, so that the leakage is given by the sum of both transmission line characteristics. This shouldn't be much different from the parallel-wire case, as the shield is a symmetry plane there. (The fact that the shield has a gap in it, does mean that image currents, under a given turn, are prevented from crossing, and have to loop around; but this is as intended -- as, if not for that, it wouldn't be much of a shield, would it? Also, if one winding doesn't cover the full width of the shield, it will act to spread that current out, with some loss incurred, as that current needs to be re-conducted, if you will -- this may be another tradeoff between LL and copper loss.)
Tim
I'm thinking for this transformer, I may drop the litz on the inner primary 1/2 and go for normal wire to try to counteract the additional copper loss by increasing the turns. I also am not really 100% sure the true effectiveness of using litz near the gap. How to possibly estimate the loss by fringing fields in nearest winding? I have no idea really.
I've seen people/papers suggest the lumpy winding where you stick some margin tape over the center and build up over it, but frankly, if I asked someone to make that, I think the wouldn't be so happy with me. So, if anything I'd just cover the whole surface of the bobbin, I think. Keep it even. Any rules of thumb to understand what kind of thickness would be useful given a certain flux density and gap dimension? One layer of 3M 44T-A is about 0.45mm. Maybe don't think about it too much and try it.
I'd really like to do a side-by-side of with and without margin tape, and no-margin tape with copper gap shield. Hmmm....
By the way, what do you mean by "Mind, you do have to balance against the outer core gap, if this is a shimmed gap; in that case, a somewhat peanut-shaped winding area would be best."? Are you talking about gapped outer legs? I don't have any of those, if that's what you're talking about. Or something else?
As always, thanks so much. It's like throwing mud at a wall (I'm the wall). But, appreciate it nonetheless. Trying to get my brain to think in terms of transmission lines but it just don't wanna!
-
Interesting. So does leakage inductance actually depend on magnetizing inductance? Let's just say you keep all windings exact the same, and just adust the gap dimension to adjust the magnetizing inductance. Do we expect leakage inductance to also change? My understanding (from what you say) is that Lleak does not directly depend on Lmag, but rather that when you have a higher Lmag, you usually have more turns, and thus the leakage inductance is also higher.
In my case, if I increase the number of primary turns, I expect leakage inductance to increase. But, if I kept the winding structures the same and re-gapped, I'd expect leakage to remain the same.
Exactly. So,
- As you vary the core (like I said, you can do this by hand, moving the core pieces in real time), Lmag is varying, LL stays ~constant (again, little change for tightly coupled windings), and k is varying (less with less core, because Lmag is smaller relative to LL).
- As you vary the turns count, layers get wider/taller, and wire length gets ~proportionally longer, while Lmag goes up quadratically. LL goes up ~proportionally (unless multilayer sections, which will cause it to go up faster).
- And also, as you vary section count (interleaving), you effectively get more transmission lines acting in parallel, or more layers of shorter (fewer turns) and fatter ones (the "fatness" can be up to full width foil). So LL is roughly inverse with section count.
Of course, this is most obvious if you're duplicating sections and wiring them all in parallel, but it works out the same for sections in series, too. It's just, at some point, you can't connect more sections in series -- like, you can't have more sections than there are turns, the least you can have is one turn per section. At that point, you have to use a series-parallel combination to make more sections. But, that's just number theory, stacking integers, it doesn't really matter; the fields would be perfectly happy with fractional turns, if there were such a thing.
So, that's how you can get arbitrarily low LL, or k --> 1. It comes at the expense of Cp, of course -- again, it's ultimately a transmission line structure and you need to balance LL and Cp, not just reduce one. Well; reducing both is satisfactory, too!
For which, again, we see how to proceed: use minimal winding length. Which suggests using a relatively large core area, with a round cross section, to maximize flux per turn while minimizing turn length. Which, pot cores are quite good at by the way!
You'll have to excuse me. -- My little brain is working hard to try to comprehend what you're saying here. Regarding your line "So, just based on a cross sectional argument, and the potential for the wire to be thinner (or even Litz)...", are you talking about using an open-circuit, unterminated wire-structure shield instead of say, a large copper strip wrapped around the center-leg?
Yeah, my pseudo-analytical spitballing probably isn't the easiest to follow; there, I'm talking about the cross section of the wire that would otherwise be placed where the shield is, and without using margin tape to block off that region. Simply enough, the wire has less cross section to the fringing field than the shield does -- the shield is wider.
A wire structure... wouldn't do anything; like say, a wire mesh? -- the average effect is increased sheet resistance. As long as the holes are spaced closer than the length scale of the fringe field itself (so for my 1mm gap example, hole spacing << 1mm), it just looks like thinner copper. :-+
Same is true of any other metal, too, and don't forget contact resistance because that'll be, well, it'll be erratic across a wire mesh -- but like, expanded mesh, perforated metal, pattern etched foil, those will act like homogeneous shields of some equivalent sheet resistance and shielding effectiveness.
This is a useful way to think of things, for foil conductors and planar transformers -- whereas bulk conductors have skin effect, sheet conductors have edge effect, and the length scale depends on the sheet resistance (given that the sheet is comparable to, or thinner than, the skin depth -- if it's thicker, it looks like a regular blocky conductor).
This can be quite significant in planar transformers, like say you want to make one super wide single-turn winding, well, all the current will crowd to the inside edge, and somewhat less to the outside edge, because it's just so wide. You can do a sort of planar Litz by segmenting such a winding into thinner traces, and after a half turn, make connections to vias, then swap layers and reverse the conductor order, so inside is outside and vice versa, and complete the winding. (Or for N turns, N halves either side of the swap.) Not sure if this is actually worth it (there's a lot of current crowding by the swap itself), but it's fun to think about, I guess.
It's also why you don't want to make multilayer windings with foil: each turn shields itself and its neighbors, so all the current flows on the edges and the thing just burns itself up. I think it's arguably okay to lap two foils together, when the current draws are exclusive -- like you'd do for a push-pull converter, you want the low voltage primary halves to be closely coupled, and only one or the other carries current at a time. Then the secondary can be kinda wherever, you don't care about secondary leakage as much in that case.
Paper says they reduced losses by increasing what percentage of the bobbin width the shield covered. Though, beyond a certain percentage, the improvement was not much. But, maybe it's a different story for a transformer compared to a plain inductor? You seem to suggest there will be some leakage between primary 1/2 and air-gap shield.
Yeah, that checks out -- the fringing field is dominant within a few gap lengths around the gap, both above/below the gap, and laterally away from it. The shield only needs to be tall enough to cover whatever fraction of that field is desired (70%? 90%? 99%? etc.).
There should be a formula which describes the fringing field... at least for the semi-infinite boundary condition (i.e., infinitely wide and tall core, cylindrical cross section, finite gap in the middle), that would be simply the dipole field equivalent to currents flowing at... oh, is it one ring each at each edge of the gap, or is it more like a hoop current for the entire gap? I'm not sure. Certainly, it's not much difference when gap length is relatively small.
So, the field around a loop, there's all those r^(-3/2)'s in the formula, and uh, whatever the off-axis intensity goes as, I forget... You'd have to integrate over some stupid secant looking path to see how much flux is enclosed by a given shield height, at a given distance from the core. Doable. Probably annoying analytically (integrals, amirite), easy to set up numerically (well, given a field solver).
Then for finite core geometries, who knows, but again as long as the gap is relatively small, the fringe won't be extending too far into the rest of the winding area, so it should be okay to approximate it as above.
I'm thinking for this transformer, I may drop the litz on the inner primary 1/2 and go for normal wire to try to counteract the additional copper loss by increasing the turns. I also am not really 100% sure the true effectiveness of using litz near the gap. How to possibly estimate the loss by fringing fields in nearest winding? I have no idea really.
Oh, it's quite good, at least when you have a bad fringing problem -- solid wire might literally burn up, whereas adequately sized Litz stays completely cool. You love to see it. The real question is, well, exactly that, how much does it really matter? Alas, my advice can do no better; at some point, we have to measure or simulate these things.
Also, don't discount simply using multiple wires in parallel (twist or multifilar) -- poor man's litz, but it gets you an improvement without splurging for the good stuff.
(Also also, regarding my foil edge effect discussion earlier -- you can avoid that too, by braiding multifilar wire. Heh, well, needless to say that isn't going to be any more attractive to coil winders, but it's interesting when it's available as cable. NEWT can make the stuff, and at least one I know of (West Coast Magnetics) claims a proprietary foil construction with the same property.)
I've seen people/papers suggest the lumpy winding where you stick some margin tape over the center and build up over it, but frankly, if I asked someone to make that, I think the wouldn't be so happy with me. So, if anything I'd just cover the whole surface of the bobbin, I think. Keep it even. Any rules of thumb to understand what kind of thickness would be useful given a certain flux density and gap dimension? One layer of 3M 44T-A is about 0.45mm. Maybe don't think about it too much and try it.
Well, you've already wound it once, and found you have so-and-so space left at the outside; clearly you can't get any further than that. That's probably a good starting point. 8)
And (just below this) you said there's no outer limb gap, so no fringe to worry about out there. Beautiful. :)
As always, thanks so much. It's like throwing mud at a wall (I'm the wall). But, appreciate it nonetheless. Trying to get my brain to think in terms of transmission lines but it just don't wanna!
Had a physics prof who preferred the shotgun strategy -- throw everything at them and hope something sticks. I forget anymore what grades I got in his classes (if it was B's or A's), but I think my tests mostly graded as C, and I had some of the better scores in his classes -- but not to sound unfair, he curved quite steeply you see. You're very unlikely to answer everything on his tests, but if you answer a few correctly, well, something must be sticking.
That was fun, heh well, from my perspective at least, I'm sure I can't say as well for a lot of other students -- since he spent relatively little time on core topics and followed related materials and tangents, I got to learn about special functions, and like, the history of weird "that shit will never work" hackery like the uh, somewhere in the 17th century I think it was, someone figured out what "(1/2)!" might mean, and solved for it -- and that led into the analysis of infinite series, and much of calculus as we know it, and finally the gamma function (one of the more common special functions, aha, that's what we're here for!), which is effectively a generalization of the factorial function to complex values.
I should probably figure out some strategy to curb that kind of thing, honestly -- my mind is filled with a multitude of little connections here and there, and I find it's as helpful to remember things by tracing those connections from one subject to the next, as it is powerful to be able to connect all those analytical tools together for a given problem. But a newbie (or even most graduates, to be honest) isn't going to have a clue what any of those topics are, it's just noise on top of an already confusing (or intimidating) subject. What I revel in, others probably cower from... or more likely glaze over and skip paragraphs of.
Anyway, (and here I go again, huh?), regarding transmission lines -- if you perspective is from within the circuit, like, you've got a terminal on this transformer and you wiggle it and so-and-so; yeah, transmission lines may not make much sense that way.
Shift your perspective into the conductors themselves, think of the fields between them -- it is, after all, the space between conductors (and a small skin depth into them) where electromagnetic fields exist. Take a cross section of a parallel conductor line (or, really any two prismatic, non-intersecting conductors, the shapes aren't special), plot the electric field between them and the magnetic field around them. Electric field means there's capacitance, and magnetic field means there's inductance; albeit not much in some differential slice, but over a finite length, both add up to very-much-finite values, and this is how we get capacitance and inductance.
Maybe it's the fact that all those little connections in my mind are so well-worn that I abstract across them effortlessly; not sure anymore. They tell me that, anywhere there are conductors in proximity, there's some kind of transmission line structure; and we can solve the impedance and length from the geometry, and use those parameters to decide whether it's significant or not. :)
Also, symmetry. This kind of cooks your brain the first time you understand it, but it's so powerful after that. Students get a taste of this from solving basic networks (like Thevenin/Norton transforms, series/parallel resistors, delta/wye equivalents, etc.; and some mesh analysis---but usually nodal analysis is emphasized, a notable asymmetry!). But the real crazy thing is general parallel-series transformations: swap V and I, swap L and C, and swap parallel for series. Everything else seems simple enough, but the topological transformation is... hard to explain. There is an algorithmic method to it, though.
And for transmission lines, there's this:
First of all, we can look at a transmission line as two ports connected by a delay. An ideal port, is two pins, through which a current flows, and across which a voltage is dropped. Note, that's no current to anything else, not to the other port, not to ground, just an ideal, isolated element, current in one pin, current out the other. This models one end of the transmission line, not one conductor of it. (It's already a twist from the DC case!)
So, the common-mode stuff is irrelevant when we're just looking at coax cables with grounded shields; signal goes in one end, and some time later, comes out the other, and so on. Simple enough. Well, we don't need to tie the shield to ground, we can put it anywhere we like; in that case, we have a shield current to worry about, but we can choke that current (at AC at least) with a big stack of ferrite beads, or turns around a core, and approximate an ideal port in real life.
So, that's the common mode choke version of a transmission line.
What do we get if we turn it literally sideways? Well, looking at the construction, it's just two windings of equal turns around a core -- an isolation transformer. But it's also well and truly a transmission line! What does that mean? Well, between the primary start and secondary start, we have one port wired in series; and between the primary end and secondary end, the other port. These ends are connected by a delay. If we apply a step voltage to the primary start, say, we get a resistor divider between the "start" port (whose transient impedance is Zo), the load, and the "end" port (phased oppositely). Some time later, those waves propagate through the transmission line and appear at the opposite ports, changing the voltage across the load (and source, if it is a Thevenin equivalent). The frequency corresponding to this delay, is the resonant frequency of the transmission line, or the cutoff frequency of the isolation transformer.
And the inductivity of this line, times its length, is the leakage inductance, and so on.
This also gives us a way to model common mode interference: if we drive the isolation transformer unbalanced (common ground on the "end" side, say), then a step change between ground voltages is propagated through the transmission line and appears at the signal end in full; if the port is terminated (Rs + RL = Zo), that's the end of it, the CM step interference manifests as one DM impulse. If mismatched, it will bounce back and forth, effectively ringing down (the stepwise nature of that ringdown may not be visible in a less carefully constructed transformer -- dispersion smears out harmonics, so you don't get to see a staircase, but some lumpy ringing instead). Whereas if we drive it balanced (both "signal" and "ground" ends are driven by Rs = Zo/2 from the common mode source), a CM step is transformed to a full CM impulse, but DM is untouched -- so long as we have enough CM range in our receiver, say, we can reject noise just fine this way. Or we can follow it up with a CMC to filter off that impulse. This is a common motif in some places, improving an isolation transformer with a common mode choke, both of which are more or less identical components, just wired differently -- Ethernet transformers are a prime example!
Tim
-
Bonjour TimNJ:
Fine topic and BRAVO for rising these issues. But we need more info to comment more easily.
What is switch frequency? Vin/Vout? What is Ferrite POT33, 33 mm pot core? Gap size?
You have a winding sheet or cross section?
For production or just one off project/proto?
Bon Chance,
Jon
-
Yeah, my pseudo-analytical spitballing probably isn't the easiest to follow; there, I'm talking about the cross section of the wire that would otherwise be placed where the shield is, and without using margin tape to block off that region. Simply enough, the wire has less cross section to the fringing field than the shield does -- the shield is wider.
A wire structure... wouldn't do anything; like say, a wire mesh? -- the average effect is increased sheet resistance. As long as the holes are spaced closer than the length scale of the fringe field itself (so for my 1mm gap example, hole spacing << 1mm), it just looks like thinner copper. :-+
Actually by wire structure, I meant something like a litz wire screen, wound like a normal layer, except just leave it floating. For instance, compare a solid piece of copper strip with 0.5mm thickness compared to a multi-turn litz winding with diameter of about 0.5mm. This is with respect to the air-gap screen. Again, not 100% sure what you were saying before, but if there was some concern with eddy currents in the foil/strip shield, would making it litz reduce those induced currents? Or it's not equivalent because now there are multiple turns involved?
Oh, it's quite good, at least when you have a bad fringing problem -- solid wire might literally burn up, whereas adequately sized Litz stays completely cool. You love to see it. The real question is, well, exactly that, how much does it really matter? Alas, my advice can do no better; at some point, we have to measure or simulate these things.
Also, don't discount simply using multiple wires in parallel (twist or multifilar) -- poor man's litz, but it gets you an improvement without splurging for the good stuff.
(Also also, regarding my foil edge effect discussion earlier -- you can avoid that too, by braiding multifilar wire. Heh, well, needless to say that isn't going to be any more attractive to coil winders, but it's interesting when it's available as cable. NEWT can make the stuff, and at least one I know of (West Coast Magnetics) claims a proprietary foil construction with the same property.)
In my experience, litz really isn't a huge cost adder, but it definitely can cut the cross-sectional copper area down considerably (when compared to solid wire). On that note, I think I ought to save the html to this website as I use it so often: https://www.elektrisola.com/en/hf-litz-wire-litz/dimensions.html (https://www.elektrisola.com/en/hf-litz-wire-litz/dimensions.html)
For example, for this particular design I could use (for a single layer 1/2 primary) either 0.45mm solid wire (0.159mm^2) or 10*0.1mm litz (0.0785mm^2). Huge reduction in copper area! The hope of litz is that the Rac losses go down enough to outweigh the "DCR" increase. Of course. Additionally, without directly monitoring inner winding hot-spot temperature, maybe using litz is just the wise thing to do anyway, as a precautionary measure.
Can you link the the NEWT and WCM braids/foils? That sounds neat. I think the "litz" I'm used to is the cheap stuff, no real braiding - wires on the outside are always on the outside.
Had a physics prof who preferred the shotgun strategy -- throw everything at them and hope something sticks. I forget anymore what grades I got in his classes (if it was B's or A's), but I think my tests mostly graded as C, and I had some of the better scores in his classes -- but not to sound unfair, he curved quite steeply you see. You're very unlikely to answer everything on his tests, but if you answer a few correctly, well, something must be sticking.
That was fun, heh well, from my perspective at least, I'm sure I can't say as well for a lot of other students -- since he spent relatively little time on core topics and followed related materials and tangents, I got to learn about special functions, and like, the history of weird "that shit will never work" hackery like the uh, somewhere in the 17th century I think it was, someone figured out what "(1/2)!" might mean, and solved for it -- and that led into the analysis of infinite series, and much of calculus as we know it, and finally the gamma function (one of the more common special functions, aha, that's what we're here for!), which is effectively a generalization of the factorial function to complex values.
I should probably figure out some strategy to curb that kind of thing, honestly -- my mind is filled with a multitude of little connections here and there, and I find it's as helpful to remember things by tracing those connections from one subject to the next, as it is powerful to be able to connect all those analytical tools together for a given problem. But a newbie (or even most graduates, to be honest) isn't going to have a clue what any of those topics are, it's just noise on top of an already confusing (or intimidating) subject. What I revel in, others probably cower from... or more likely glaze over and skip paragraphs of.
Tim
I haven't had too many mentors so far, in my (short) career, but I usually prefer when they don't "hold back", because I'm of the opinion that the brain absorbs more than you may consciously think it does. Yes, your responses are often quite challenging to get through, but I think they (subtly) "inspire" the brain to open up to a new way of thinking. Even if I don't immediately get it, it at least gives my brain something to recall sometime in the future. I don't know - I'm not a psychologist. That said, for the "practicing professional", often immediate needs to finish something or to make something work makes us less patient to explore at great depth. I think most people are interested in gaining a deeper understanding, but just don't have the time or bandwidth to.
Thanks!
-
Bonjour TimNJ:
Fine topic and BRAVO for rising these issues. But we need more info to comment more easily.
What is switch frequency? Vin/Vout? What is Ferrite POT33, 33 mm pot core? Gap size?
You have a winding sheet or cross section?
For production or just one off project/proto?
Bon Chance,
Jon
Greetings Jonpaul. The original transformer design is attached to my first post. The switching frequency at full load is designed to be around 60KHz, based on quasi-resonant (first valley) switching, assuming a Lm/Cd ring frequency in the range of 400-500KHz, so deadtime around 1.5us, maybe. Vin is about 265V for low input line and 395V when high input line. Variable PFC voltage.
The core is actually not exactly POT. it's modified (open end) version "PTS". The exact core I'm using is PTS33-19. (Each core half 9.5mm)
http://www.chinadmegc.com/en/upload/201305/beaa121e9031837133f784a4c72906bc.pdf (http://www.chinadmegc.com/en/upload/201305/beaa121e9031837133f784a4c72906bc.pdf)
The material is DMR44, which is equivalent to PC44. This is indeed for a professional, large(ish) quantity production project. It's a bit of a back-burner project, but professional nonetheless. The company already uses the DMR44 material heavily, in this core shape, so it's preferred to keep using '44 material. But '47 is probably a better choice overall.
The gap size is estimated at about 0.3 - 0.4mm per calculation. I have not much intuition as to whether a 0.4mm gap on a ~13mm center leg height is probable to cause considerable fringing flux. It's only about 3% of the height of the center leg. (Operating at a peak flux density of about ~0.27T.) Tim gave some guidance above, but still trying to make heads or tails of it.
Thanks!
-
drill hole in to the end of the core and move gap to the end of the core away from winding and put ferrite cap to enclose all fringing fields escaping from end of the core
both problems solved
-
drill hole in to the end of the core and move gap to the end of the core away from winding and put ferrite cap to enclose all fringing fields escaping from end of the core
both problems solved
Wait isn’t that already a thing for certain select core geometries? ‘RM’ comes to mind. They make versions with a screw- tunable slug? I’ve never used one myself, but I suppose there is no exposed inner gap for those right?
But, IIRC, these have lower Ae, effective area, which is maybe why they’re not so common?
-
TimNJ: Many thanks for the notes,
I can only recall my past designs in 1970s..1990s. ... forward, flyback, buck and resonant/soft switching for medical, lighting, avionics, audio, high voltage applications
Core shape: We found more open, round center leg modified EE types like EER ETD etc have better heat dissipation and more rugged than the modified pot cores.
Material ...for 60khz kHz you have many choices, PC44 or 47 fine.
Litz: More a headache in termination and procurement. We never had to use it, bunched wire or foil were better and easier to use. The foil windings for shields and low leakage designs were especially effective as radial build is minimized.
Fringing flux: Had many gapped designs with Pout 100..2 kW and never has a Trise issue due to the gaps.
Interleave: We did many P-S-P and many S-P or P-S. Many had foil shields and some with wire shields.
Compliance: Achieving the required SH-P and P-S HyPot was the tricky part, and affected by the shield and winding design as well as core shape and bobbin.
Leakage L: We used perfect layer wind to reduce radial build.
Good luck on your project!
Just the ramblings of an old retired EE.
Bon Chance
Jon
-
Are you able to prepare a few alternative winding builds and swap them in to a converter and accurately measure loss differences ?
Is the coreset gap in the centre leg as well as the outer return, or do you have access to gapped centre leg cores?
I'd prepare some builds with solid wire for both primary layers (ie. don't use litz to allow more available build height; and use same turns for each primary layer to balance the interleaving better), with builds having stepped levels of margin tape for spacing closest wire layer from gap.
I haven't read the linked paper on gap screening (used the link to ask the author for a copy) - is the open screen located between inner primary layer and secondary (as per your diagram), or between inner gap and inner primary (as per your text) ?
-
TimNJ: Many thanks for the notes,
I can only recall my past designs in 1970s..1990s. ... forward, flyback, buck and resonant/soft switching for medical, lighting, avionics, audio, high voltage applications
Core shape: We found more open, round center leg modified EE types like EER ETD etc have better heat dissipation and more rugged than the modified pot cores.
Material ...for 60khz kHz you have many choices, PC44 or 47 fine.
Litz: More a headache in termination and procurement. We never had to use it, bunched wire or foil were better and easier to use. The foil windings for shields and low leakage designs were especially effective as radial build is minimized.
Fringing flux: Had many gapped designs with Pout 100..2 kW and never has a Trise issue due to the gaps.
Interleave: We did many P-S-P and many S-P or P-S. Many had foil shields and some with wire shields.
Compliance: Achieving the required SH-P and P-S HyPot was the tricky part, and affected by the shield and winding design as well as core shape and bobbin.
Leakage L: We used perfect layer wind to reduce radial build.
Good luck on your project!
Just the ramblings of an old retired EE.
Bon Chance
Jon
Thanks for the good tips Jon. Interesting note about the EER vs POT style cores. I've also found that EER gave better (overall) power density when compared to something like PQ or RM. The tricky part is more open design also means its more open for EMI. So far, pretty good luck with EE/EER for LLC resonant transformers which have generally softer waveforms. But, maybe think twice before using it for flyback, et. al.
Good to know about your experience with fringing flux. Clearly there is always some effect due to the fringing flux. The question is "how much?". I wish I had half a clue about how to use field-solvers/FEM. Maybe matters more for some high current, low current inductor with a big ol' gap.
Thank you.
-
Are you able to prepare a few alternative winding builds and swap them in to a converter and accurately measure loss differences ?
Is the coreset gap in the centre leg as well as the outer return, or do you have access to gapped centre leg cores?
I'd prepare some builds with solid wire for both primary layers (ie. don't use litz to allow more available build height; and use same turns for each primary layer to balance the interleaving better), with builds having stepped levels of margin tape for spacing closest wire layer from gap.
I haven't read the linked paper on gap screening (used the link to ask the author for a copy) - is the open screen located between inner primary layer and secondary (as per your diagram), or between inner gap and inner primary (as per your text) ?
As I think it may be a good learning experience, I want to build ~3 revised transformers, fundamentally very similar to each other, but each one with one distinct difference. For example, 1. litz wire on inside layer, 2. solid wire on inside layer, 3. solid wire on inside layer with margin tape applied before inner layer...and in an ideal world with unlimited time, 4. add core gap shield.
Sorry for the confusion - My diagram is for an existing design which is giving 90% + efficiency, fair EMI performance, but am looking to increase turns ratio to address secondary synchronous rectifier voltage stress issue. (Basically, want to switch the synchronous MOSFET to a low-inductance, leadless SMD type instead of TO-252, but no one is making the voltage rating needed in that package.)
The paper places the copper screen as the "first winding layer". i.e. It is directly next to the bobbin wall, closest thing to the gap. The copper foil shield in my diagram is to divert some CM current from flowing between primary and secondary.
The plan is to only gap the center leg, not the outer legs. Overall, it seems most times it's best to leave the outer legs alone unless you have a really particular circumstance. We have many ungapped core halves, plus diamond hand files and lots of elbow grease ready!
-
Whew that's a tough ask to gap the centre leg by personal blood, sweat and tears!
-
Whew that's a tough ask to gap the centre leg by personal blood, sweat and tears!
It's not that bad for <0.5mm. Anything more starts to border on inhumane. >:D
-
Here's a follow-up for anyone interested. To refresh your memory, the fundamental design is a ~100W quasi-resonant DCM flyback with boundary-mode PFC.
Two versions of the revised transformer were made in-house transformer/coil-winding department. The build-ups are identical except one uses 3 layers of 3M 44T-A (T=0.45mm) tape around the inner-wall of the bobbin, nearest the core gap. So, there is 1.35mm build-up before the copper begins. In terms of build consistency, given that these were wound by seasoned professionals on proper machines, I think the two transformers should be quite similar. Something still to be compared is the leakage inductance which may shed light on whether the following results are truly a result of the the additional tape, or if just leakage current deviation between the two. I shall report back on that soon.
For the meantime, here is the "4-point avergge" efficiency data, both taken on the exact same DUT, just swapping transformers. The air-gap size is approximately 0.45mm. The height of the center leg is about 10mm. So, the gap is about 5% of the height of the center leg, for what it's worth.
Without tape:
115V: 89.95%
230V: 90.24%
With tape:
115V: 90.31%
230V: 90.63%
Change:
115V: +0.36%
230V: +0.39%
-
Hey what was the bobbin wall thickness anyway? I don't remember and I can't seem to find a dimension above...
Impressive that the difference is measurable, and seemingly in the right direction, for such a small change!
Tim
-
Hey what was the bobbin wall thickness anyway? I don't remember and I can't seem to find a dimension above...
Impressive that the difference is measurable, and seemingly in the right direction, for such a small change!
Tim
Yeah. I still want to make sure that it's not just leakage inductance/coupling related. As I said, I have faith in the people that made them, but you never know. As you said, it's encouraging that the difference is in the right direction.
The thickness of the inner wall for the POT3319 (PTS3319) bobbin is 1.5mm. There is also about 0.25mm gap between core and bobbin, so about 1.75mm total.
-
Oh well hell, that was a lot of conversation for nothing. :-DD
I think I made the implication above that, for distances comparable to the gap, it's a concern, while for larger distances, who cares... Well, with the bobbin already taking up three times the air gap, that's about what to expect!
Could try it again with, like, a tight fitting, thin cardboard bobbin, I suppose you'll have to do it yourself (and so, give or take skill of the winders who made the above articles), but it should be a quite dramatic difference in the negative direction this time.
Tim
-
Oh well hell, that was a lot of conversation for nothing. :-DD
I think I made the implication above that, for distances comparable to the gap, it's a concern, while for larger distances, who cares... Well, with the bobbin already taking up three times the air gap, that's about what to expect!
Could try it again with, like, a tight fitting, thin cardboard bobbin, I suppose you'll have to do it yourself (and so, give or take skill of the winders who made the above articles), but it should be a quite dramatic difference in the negative direction this time.
Tim
I am confusion. Seems you are saying that the original distance (about 1.75mm) was already large with respect to the gap length (1.75mm : 0.45mm = 3.9). But, after comparing a build with 1.75mm to 3.10mm, we see that the 75% increase appears to be helpful helpful... implying that the original 3.9 ratio was perhaps not so large after all.
What kind of ratio of (core/bobbin separation) : (gap length) do you consider large? 10:1?
Thanks.
-
I'd be worried for < 3:1. It goes as what, distance cubed or so?
The difference you saw is all of, what, 0.3W? Could easily make more by tweaking gate resistors, or snubber stuff -- it's small in the grand scheme of things, and like we said, impressive to measure seemingly correctly. I certainly wouldn't sweat it!
Tim
-
Got it. Yeah somewhere between 0.3 and 0.4W reduction. It's a nice improvement. I would be foolish to say the rest of the design is "maxed out" efficiency-wise, but there aren't too many more levers to pull, given cost and EMI constraints, etc...so I'll take it.
-
Just to round out the 'picture', can you elaborate on some aspects such as:
- the height margin available (to outer leg) with and without tape.
- the accuracy tolerance of the efficiency measurement
- the likely increase in DC resistance loss due to larger turn length with tape (that is then more than compensated for by reduction in loss likely related to fringing effect)
- the likely total loss of the transformer (core plus winding)
- did you use the litz inner and solid wire outer layer plan described before (assuming you used same P-S-P interleaving) or just go for simpler solid wire inner and outer primary layers with same number of turns per inner/outer layer.
-
Just to round out the 'picture', can you elaborate on some aspects such as:
- the height margin available (to outer leg) with and without tape.
- the accuracy tolerance of the efficiency measurement
- the likely increase in DC resistance loss due to larger turn length with tape (that is then more than compensated for by reduction in loss likely related to fringing effect)
- the likely total loss of the transformer (core plus winding)
- did you use the litz inner and solid wire outer layer plan described before (assuming you used same P-S-P interleaving) or just go for simpler solid wire inner and outer primary layers with same number of turns per inner/outer layer.
- the height margin available (to outer leg) with and without tape:
The bobbin supports about 5mm build-up. With the assumption that lower leakage inductance of single layer primaries/secondaries outweighs multi-layer construction with larger wire, the actual build up is not that high. Without margin tape, the height is about 3.3mm. With additional margin tape, the height is about 4.6mm.
- the accuracy tolerance of the efficiency measurement
The efficiency was measured with Yokogawa WT3000, 0.02% basic accuracy. Measured once the rate of change in efficiency is sufficiently low, usually 30 minutes minimum to reach equilibrium.
- the likely increase in DC resistance loss due to larger turn length with tape (that is then more than compensated for by reduction in loss likely related to fringing effect)
Hmm, I can try to re-calculate based on new mean length of turn. Not sure, but really doesn't seem like it will change much
- the likely total loss of the transformer (core plus winding)
For the new version, I've estimated roughly 0.62W of copper loss, and 0.46W of hysteresis core loss. Accuracy of the core loss calculation may be a little sketchy though, but about 1.1W total.
- did you use the litz inner and solid wire outer layer plan described before (assuming you used same P-S-P interleaving) or just go for simpler solid wire inner and outer primary layers with same number of turns per inner/outer layer.
I have switched to symmetrical number of turns on both primary halves, both using same type of solid wire. Interestingly, compared to the original transformer (with 10T (litz) + 20T (solid) primary), the new transformer with (20T (solid) + 20T (solid) primary, no margin tape) had slightly worse efficiency, even though the increased turns (and lower flux density) indicated that the improvement in core loss would outweigh the increase in copper loss.
Without accounting for loss related to fringing flux, I estimated that the 20T + 20T construction would beat the 10T + 20T construction by about 0.8%, but in fact, for the non-margin-tape version it was worse by about 0.1-0.2%. After switching to margin-tape construction, the efficiency improved by about 0.35% (as above) and beat the (old) litz/solid construction by about 0.15-0.25%.
So, it's possible the unaccounted loss factor was the loss related to the fringing flux. The litz may have been doing more than I thought.
-
Regarding changing from (litz + solid) to (solid + solid) construction, I recognize that the turns ratio of the transformer was changed between these two versions...from 6:1 to 8:1. The discrepancy in expected efficiency improvement may be related to secondary effects like higher secondary RMS currents, etc.
Overall not much was ~really~ gained between the old version (10T + 20T, litz/solid) and new version (20T + 20T, solid only)...at least from an efficiency perspective. However, the main goal of the revision was actually to adjust the turns ratio to play better with the MOSFET drain-source voltage ratings on both primary and secondary side. In this sense, it's still a win for me, in that I didn't make the efficiency any worse. :-DD
-
The WT3000 appears to have about a 0.1% tolerance for each input channel when measuring DC (0.05% for reading, and 0.05% for range), so 0.4% for an efficiency measurement, but as you are doing a comparison measurement using the same channel ranges and levels, and mitigating as many other accuracy issues as possible, then I guess the comparison results could have an accuracy tolerance acceptably under 0.1%.
The test winding with tape appears to be pushing the winding close to the max fill height of the bobbin, so that makes sense as a good test case.
Is it practical to do a bifilar winding test case with/without tape, as I could imagine that would be the main other practical configuration that could be manufactured.
A nominal reduction from circa 1.4W to 1.1W dissipation in the transformer makes good sense to me, but I guess that depends a bit on what the 'in situ' core temperature benefit turns out to be.
-
Rebonjour a Tous et Tout:
To TimNJ:
1/ Bravo for the fine thread and responses. I had designed and ran prod for power supplies used in audio, High voltage, avionics, radar, medical, lighting, cinema.
50-750W, topology forward, double forward, resonant (soft switching) at freq 40-500 kHz in the 1970s..1990s.
2/ We used gapped cores but placed extra tape on the bobbin to avoid the issues of gap fringe flux as the OP asked.
3/ CORES The open style cores EE, EI, EER, are most economic, and have best heat dissipation. The shielding of a ferrite core shape is not an effective EMI screen. We used shield layers where required eg between P and S. EMI was always OK if the CM and DM noise filters and components were well engineered. Finally external field can be greatly reduced with an external copper strap shorting band.
Thus, core shape is not a determinant of external field or EMI. The old pot cores have lowest external field but worst heat dissipation, ease of production and fragility. PQ was a compromise of TDK between a pot core and EER. Those were costly and gave little benefit. Vendors were TDK (the best) and NiCera, Nippon Ceramic.
4/WIRE: We never had to use litz, very costly and hard to terminate. We used bunched wire and Cu foils to control skin effect etc.
triple insulated wires for medical and special application ...Magnet Wire Supply in LA, MWS, has everything but very costly.
Litz is also avail triple insul from
https://www.rubadue.com/winding-wire/ (https://www.rubadue.com/winding-wire/)
Today we use Belden Wire from disty butfiner gauges #36-39 are more and more made only in China.
Hope that someone will find this note useful!
Bon Chance,
Jon
-
Forgot: Measurement of efficiency is very difficult taking P1-P2, in the calcs mentioned as you had two numbers very close, and they that have tolerances.
We used a temperature rise with a thermocouple at the core/winding hot spot, embedded during the winding.
Jon
-
Core shape: We found more open, round center leg modified EE types like EER ETD etc have better heat dissipation and more rugged than the modified pot cores.
Thanks for the good tips Jon. Interesting note about the EER vs POT style cores. I've also found that EER gave better (overall) power density when compared to something like PQ or RM. The tricky part is more open design also means its more open for EMI. So far, pretty good luck with EE/EER for LLC resonant transformers which have generally softer waveforms. But, maybe think twice before using it for flyback, et. al.
Any clues on why pot type cores tend to perform worse? I haven't physically built and validated many transformer designs yet but from simulations so far it seems that pot and modified pot type such as RM perform better than open types such as EER or ETD so I'm wondering what's being missed in simulations.
The only case I've noticed open E type cores out performing more enclosed pot type cores is when copper loss is quite high so the more open cores tend better allow heat dissipation from the windings but when magnetic and copper losses are well balanced and optimised it seems pot type cores perform best.
-
There was another thread a few months before your OP where @fourtytwo42 was also looking at the same paper and method to try to deal with fringing flux issues. Apparently he tried that method in the past but didn't go into detail on how well it worked other than it was difficult to fabricate. He ultimately just used a spacer to keep windings away from the gap fringing flux https://www.eevblog.com/forum/projects/reducing-fringing-flux-problems-in-gapped-etd-cores/msg3280672/#msg3280672 (https://www.eevblog.com/forum/projects/reducing-fringing-flux-problems-in-gapped-etd-cores/msg3280672/#msg3280672)
Edit: Since no one else has shared a similar take on why the copper flux screen works, I'll share my take on an "intuitive" picture of why it works. The key point is that the copper flux screen is open-circuit i.e. non-shorting with respect to the main "well behaved" flux passing through the core and the gap parallel to the winding surface (perpendicular to the core gap face). For the fringing flux with vector components perpendicular to the main flux and core surface however this copper screen is not "open circuit" and will have substantial induced current (eddy currents) producing a magnetic field opposing and suppressing that perpendicular component of the fringing flux (Lenz's law). Thanks to this suppression of fringing flux by the copper screen, the actual windings which are carrying much more substantial current coupled to the main flux are spared from additional eddy currents which do not contribute to output power at the terminals.
I'm not 100% certain in my understanding of why this results in lower loss overall being that there are still eddy current present but now in the screening layer instead of the windings but here goes: Without the screening/shielding layer, the windings will have a similar interaction still causing an induced eddy current which suppresses the fringing flux (and helps reduce EMI) but the losses are worse due to I^2R (?) and the windings are also less geometrically "efficient" since the windings are normally long thin wires (particularly in the case of Litz) so the current path confined to higher resistance loops?
Simply using a non-conductive spacer (to keep windings away from the worst of the fringing flux and give the fringing flux a chance to turn back on itself and return to the core) seems to be the easier and perfectly adequate solution in most cases. I guess the flux screen can be helpful when you really want to maximise winding window utilisation?
-
The WT3000 appears to have about a 0.1% tolerance for each input channel when measuring DC (0.05% for reading, and 0.05% for range), so 0.4% for an efficiency measurement, but as you are doing a comparison measurement using the same channel ranges and levels, and mitigating as many other accuracy issues as possible, then I guess the comparison results could have an accuracy tolerance acceptably under 0.1%.
The test winding with tape appears to be pushing the winding close to the max fill height of the bobbin, so that makes sense as a good test case.
Is it practical to do a bifilar winding test case with/without tape, as I could imagine that would be the main other practical configuration that could be manufactured.
A nominal reduction from circa 1.4W to 1.1W dissipation in the transformer makes good sense to me, but I guess that depends a bit on what the 'in situ' core temperature benefit turns out to be.
I know what you mean ..From my experience, testing the same unit multiple days or weeks apart, the efficiency readings have typically been within 0.05% of each other. So in that sense, the overall stability/precision of the WT3000 seems very good, although the absolute accuracy is not incredible. (Well actually it is still quite good). For example, we made a change to add a ferrite bead to the flyback MOSFET. We tested efficiency 4 times to confirm. No ferrite, add ferrite, no ferrite (sanity check), add ferrite (sanity check), and the difference of about 0.4% was detectable each time.
And bi-filar for which winding? Or what do you mean?
-
Core shape: We found more open, round center leg modified EE types like EER ETD etc have better heat dissipation and more rugged than the modified pot cores.
Thanks for the good tips Jon. Interesting note about the EER vs POT style cores. I've also found that EER gave better (overall) power density when compared to something like PQ or RM. The tricky part is more open design also means its more open for EMI. So far, pretty good luck with EE/EER for LLC resonant transformers which have generally softer waveforms. But, maybe think twice before using it for flyback, et. al.
Any clues on why pot type cores tend to perform worse? I haven't physically built and validated many transformer designs yet but from simulations so far it seems that pot and modified pot type such as RM perform better than open types such as EER or ETD so I'm wondering what's being missed in simulations.
The only case I've noticed open E type cores out performing more enclosed pot type cores is when copper loss is quite high so the more open cores tend better allow heat dissipation from the windings but when magnetic and copper losses are well balanced and optimised it seems pot type cores perform best.
It seems that POT and modified POT perform better in what sense? Although Jonpaul says the ferrite material often isn't very effective at EMI shielding, it does help to some degree, and I personally don't think I'd run anything hard-switching with a EER/ETD core. I figure there has to be a good reason the majority of flyback power supplies use POT, PQ, or RM style cores. ;)
I don't have much feedback on the reasons different shapes make work better than others. I guess we can over-simplify in terms of different effective Ae, Le, or Ve values for a given physical size.
-
Rebonjour a tous..
I try to clarify and respond to your questions:
1/ Progression history of core shapes:
original EE, EI square cross sec
1930s..1950s pot core
1980s modified pot for larger window and less enclosure: TDK PQ
1980s modified EE with round center leg ETD, EER, etc.
2/ POWER Cu><Fe: The old rule of thumb, equal core/copper loss was determined for mains frequency transformers! NOT for design of modern SMPS magnetics.
Most often the Pcu is much higher than Pfe as cores losses have improved in modern materials.
The curve of core loss is temperature variant and certain power materials have peak efficiency at a certain T hotspot! (check some old and new power ferrite spec sheets, eg TDK, EPCOS, NiCera)
3/Core geom: Every core shape has windows, the more or less open geometry will not assure EMI shielding.
We used copper or foil Faraday shields interlayer, and then if needed an external copper shorting strap to intercept the EMI.
We had no issues in EMI compliance even for medical and avionics magnetics, with PQ, EER, ETD, and other open shapes.
For HV transformers, a UU with round leg cross section is needed to reduce corona and allow great P-S spacing.
4/ Materials properties: The point of ferrite is to have a non-conductive magnetic material!
Ferrite is a ceramic, the conductance is NOT like a conductor! The various mixes differ widely in the conductance.
IF a core was highly conductive then eddy currents would cause great losses!
Je espère que ces informations préciser les situations!
I hope this information clarifies the situation.
Bon soiree,
Jon
-
Forgot: Measurement of efficiency is very difficult taking P1-P2, in the calcs mentioned as you had two numbers very close, and they that have tolerances.
We used a temperature rise with a thermocouple at the core/winding hot spot, embedded during the winding.
Jon
Indeed...I'd say (generally) at < 0.5% change, the test requirements get significantly harder, and results which seem to be because of 'parameter X', may in fact be due to 'parameter Y', or something tolerance related.
On that note, the sample transformer with margin tape measured approximately 2% lower leakage inductance compared to the sample transformer without tape. The magnetizing inductance of the margin-tape transformer was about 0.5% higher than the transformer without margin tape.
For a quasi-resonant flyback, higher magnetizing inductance pushes frequency lower. In this case, I estimate about 500Hz lower, maximum. Lower leakage inductance = lower loss...So, now the question is: Would these slight differences account for the overall 0.35% improvement in efficiency seen on the margin-tape transformer. A larger test sample size would be one way to confirm.
On the other hand, the theory still holds true that fringing flux around the gap will cause additional loss, so given how easy it is to add this margin tape, it may not be worth launching some crazy investigation.
-
Tim agree on the tape/investigation.
Any feedback on my other points re cores shapes, Pcu, Pfe, losssses vs temp?
Jon
-
It seems that POT and modified POT perform better in what sense? Although Jonpaul says the ferrite material often isn't very effective at EMI shielding, it does help to some degree, and I personally don't think I'd run anything hard-switching with a EER/ETD core. I figure there has to be a good reason the majority of flyback power supplies use POT, PQ, or RM style cores. ;)
I don't have much feedback on the reasons different shapes make work better than others. I guess we can over-simplify in terms of different effective Ae, Le, or Ve values for a given physical size.
My experience so far (again based on simulations) is that the best optimised design I can come up with for any particular magnetic, be it a HF inductor or HF transformer tends to be some flavour of modified POT e.g. PQ, PM, RM. For the given requirements of a particular magnetic component (frequency, inductance, magnetising current, turns ratio etc.) in the domain of frequency and power (100kHz to 500kHz, 1kW-10kW order of magnitude) I work in, the modified pot cores tend to have the best performance across the board (when optimised as much as I can in simulations) i.e. lowest loss, lowest temperature, lowest size/volume/weight and lowest cost.
If you'd be willing to post some specs (for this or any other random example design we could try) I could try post my attempt at a simulation optimised design.
Rebonjour a tous..
I try to clarify and respond to your questions:
1/ Progression history of core shapes:
original EE, EI square cross sec
1930s..1950s pot core
1980s modified pot for larger window and less enclosure: TDK PQ
1980s modified EE with round center leg ETD, EER, etc.
2/ POWER Cu><Fe: The old rule of thumb, equal core/copper loss was determined for mains frequency transformers! NOT for design of modern SMPS magnetics.
Most often the Pcu is much higher than Pfe as cores losses have improved in modern materials.
The curve of core loss is temperature variant and certain power materials have peak efficiency at a certain T hotspot! (check some old and new power ferrite spec sheets, eg TDK, EPCOS, NiCera)
3/Core geom: Every core shape has windows, the more or less open geometry will not assure EMI shielding.
We used copper or foil Faraday shields interlayer, and then if needed an external copper shorting strap to intercept the EMI.
We had no issues in EMI compliance even for medical and avionics magnetics, with PQ, EER, ETD, and other open shapes.
For HV transformers, a UU with round leg cross section is needed to reduce corona and allow great P-S spacing.
4/ Materials properties: The point of ferrite is to have a non-conductive magnetic material!
Ferrite is a ceramic, the conductance is NOT like a conductor! The various mixes differ widely in the conductance.
IF a core was highly conductive then eddy currents would cause great losses!
Any feedback on my other points re cores shapes, Pcu, Pfe, losssses vs temp?
Core Shape History
I can't comment much on history of core development, there doesn't seem to be much information available I can find online and I do not recall detailed discussion of history in any textbook I have read. Just no one has bothered to write on the detailed history of magnetics design developments over the past centaury it seems. I found this webpage which suggests manufacture of RM cores in Russia starting 1976 https://web.archive.org/web/20191006071021/http://ferrite.ru/products/epcos/rm/rm-ussr/ (https://web.archive.org/web/20191006071021/http://ferrite.ru/products/epcos/rm/rm-ussr/) Certainly your quote is the most detail I have seen yet.
Copper vs Core Losses
Again based on experience from simulations, for a given high frequency core, when optimised to the lowest possible overall loss, the copper and core loss tend to be quite close. Copper loss does tend to be higher but still within order of magnitude of core loss in most cases when the optimised overall loss with the given core is also good. I can usually reach 99% or greater efficiency in simulations.
High temperature materials
It certainly seems most modern core materials are designed for peak efficiency at high temperatures, particularly <500kHz optimised materials (see attached, feroxcube is 100kHz 200mT loss). I'm not sure there are obsolete materials no longer advertised which are different. I lack context to determine what is new and "old".
Core geometry and EMI has not been something I've simulated or tested yet.
Core material conductivity
I understand ferrites (and other nanocrystalline magnetic alloys for that matter) intended for inductor and transformer use are designed to minimise conductivity in order to reduce the core losses. I'm not sure what point that was in relation to discussion points. The purpose of a fringing flux copper screening layer is to selectively suppress the fringing flux (through eddy current effects) overall eddy currents and their associated loss is still being minimised.
Again thank you for your experienced input Jon!
-
Cool fact, as far as fields are concerned, any loss effects skin effect -- it needn't be direct electrical conduction, hysteresis will do it as well. So there's a maximum core thickness to frequency relationship, beyond which you need to use smaller cores stacked up.
Coincidentally, no one actually makes cores so large, the thickest I think being ca. 2.5 x 5 x 10cm bricks. Which, unsurprisingly, are exactly what are used to construct very large power transformers -- I've seen >1MVA units by Jackson Transformer, using edge-wound rectangular copper tubing, brazed together, all water cooled obviously. Impressive units, both in ratings, stature, and unit cost... ;D
Laminated structures are also relevant to physics applications, like shock lines and magnetic compressors. These are nonlinear applications that are complicated by saturation: as mu drops (transiently), the loss also drops, and skin depth increases (indeed, transient saturation occurs as a propagating shock wave of sorts). So there's a whole lot going on at once, and they may need that little boost to performance.
As for higher frequency materials, I'm not sure offhand if resistivity tracks between types; MnZn ferrites are all roughly similar, maybe give or take a factor of 10 though, I haven't checked in detail. The next big step up is NiZn which has substantially higher resistivity (by like 1000x), as well as lower hysteresis loss -- at expense to mu (topping out at ~1000 e.g. mix #43) and Bsat (~0.2T) of course, but if you're working at some MHz, neither of those are a big priority.
Tim
-
And bi-filar for which winding? Or what do you mean?
I was thinking more about the primary layers (inner and outer). But just noticed your secondary is bifilar but relatively large wire gauge, but that may be a practical optimum compared to say trifilar. At some point the checking of options becomes too costly/onerous for any likely benefit.
Do all winding layers have their wires butting up to each other or is there some splay available in some winding layers? Eg. if there was some wiggle room, then perhaps spread out the winding as much as possible in the middle of the layer.
Do you have a view on why the padding has caused a noticeable reduction in leakage inductance, and increase in magnetising inductance? Perhaps it relates to a better coupling of the primary winding to the core as it is now closer to the outer legs of the core.
-
As for higher frequency materials, I'm not sure offhand if resistivity tracks between types; MnZn ferrites are all roughly similar, maybe give or take a factor of 10 though, I haven't checked in detail. The next big step up is NiZn which has substantially higher resistivity (by like 1000x), as well as lower hysteresis loss -- at expense to mu (topping out at ~1000 e.g. mix #43) and Bsat (~0.2T) of course, but if you're working at some MHz, neither of those are a big priority.
Fair-rite has some nice tables which include material type and resistivity. https://www.fair-rite.com/materials/ (https://www.fair-rite.com/materials/)
MnZn goes from 50 ohm-cm for 76 (10000 mu) which is a suppression ferrite to 300 ohm-cm for 75 (5000 mu) which is a material for inductors.
NiZn goes from 10^5 ohm-cm for 43 (800 mu) to 10^9 for materials 51 and 52 (350, 250 mu)
Some lower permeability mixes seem to also have lower residual flux.
-
And bi-filar for which winding? Or what do you mean?
I was thinking more about the primary layers (inner and outer). But just noticed your secondary is bifilar but relatively large wire gauge, but that may be a practical optimum compared to say trifilar. At some point the checking of options becomes too costly/onerous for any likely benefit.
Do all winding layers have their wires butting up to each other or is there some splay available in some winding layers? Eg. if there was some wiggle room, then perhaps spread out the winding as much as possible in the middle of the layer.
Do you have a view on why the padding has caused a noticeable reduction in leakage inductance, and increase in magnetising inductance? Perhaps it relates to a better coupling of the primary winding to the core as it is now closer to the outer legs of the core.
In this case, the secondary is triple-insulated wire. The insulation is relatively thick and eats away at the bobbin utilization. So, if I switched to tri-filar, basically the cross sectional area of copper available for a single layer goes down.
Regarding the winding layers "butting up to each other", I'm not 100% sure what you mean by that...but I think you are asking "Do all windings have full winding 'breadth' utilization?". The answer is yes...all layers are completely filled out.
I cannot say that the padding is what caused the leakage inductance reduction and magnetizing inductance increase. It's more likely that variation in magnetizing inductance is just due to manufacturing process control. Small variation in gap dimension is easily possible due to how tightly the "core fixing" tape is wound around the two core halves before the varnishing process. Regarding the leakage inductance, with slightly larger mean-length-of-turn (MLT), we might expect the leakage inductance to go ~up~ ever so slightly for the padded/margin tape version, but in this case it went down. Again, I think we have to attribute this to process variation. All of the wire layers are wound by machine, which should do a good job winding the wires as tight as possible...but something like the cuffed copper foil shield may introduce some variation as it has to be applied by hand, taped down by hand. Could be something like that.
Anyway, ran one more test today with a second sample of the no-padding transformer. Compared to the previously tested padded transformer (which measured Llk = 9.05uH, this non-padded sample measured 5% lower leakage inductance at 8.57uH. The 8.57uH non-padded transformer still could not beat the efficiency of the padded 9.05uH transformer, and was still about 0.15% short of the padded efficiency.
So, to finish the thought, from what I can tell, the padding is still effective...just the actual difference may be smaller than the 0.35% originally reported. Probably in the range of 0.25-30% improvement due to the padding. YMMV. And as Teslacoil Tim said, it matters most when you have a large gap dimension, relative to the distance between the gap and the first copper winding.
-
Thanks by the way, for going and actually testing this. :)
Tim
-
Thanks by the way, for going and actually testing this. :)
Tim
Pleasure...after blabbering on about invisible electrons and invisible EM waves all day :scared:, sometimes you have to get down to the fundamentals and make sure reality is actually behaving as you claim it is..It's more for mental health than anything else. :-DD
-
TimNJ, ScandleCandle and TeslaCo, many thanks for the notes....great discussions...not many experienced SMPS/magnetics folks still around...
RE ferrite conductance , my note was about effect of the core as an EMI shield for E fields. Since ferrite is a poor RF conductor, regardless of closed (potcore) or open (EE, UI, UU) shapes,
the radiated EMI will not be very different, if all other factors are the same.
RE temp dependence of core losses, modern materials are targeted for a max hotspot temp rise and have worse losses above or below the target temperature.
The trend in materials has been to higher frequencies, lower flux and more sensitivity of Pfe to T.
For core material curves, besides the ones shown, see TDK and old Philips = EPCOS.
RE history, Phillips in Netherlands, Siemens in Germany and Bell Labs USA developed ferrites back to 1930s for use in radio and telephone equipment.
After the round pot cores, the square variants RM arose to save space on PCBs. Much better bobbins and PCB pinout, still in use. We used RM6, RM 8 in HV and Avionics in 1980s.
RE sizes, largest we used in prod was a PQ 40, 50 from TDK, at about 500-750 W.
https://www.tdk-electronics.tdk.com/en/529402/products/product-catalog/ferrites-and-accessories/epcos-ferrites-and-accessories/pq-cores-and-accessories (https://www.tdk-electronics.tdk.com/en/529402/products/product-catalog/ferrites-and-accessories/epcos-ferrites-and-accessories/pq-cores-and-accessories)
the largest cross sections available are EI and UI square leg, ~ 25-35 mm thickness dimension. 1970s..1980s, Stackpole, no longer butavailable.
Power was up to 2-5 KW, but much easier for several transformes in parallel.
RE ratio of power losses core vs copper, this should be researched a bit further, we think the balance of equal power does not yield the optimum designs at high frequency.
This was researched in 1980s by an old friend at Teledyne, Herr HENDRICKS, ex Philips Eindhoven. See the IEEE transaction on magnetics and power electronics may have some papers on this interesting topic.
J'espere que cette informations sera intéressante a tous.
Bon soiree,
Jon
-
RE ferrite conductance , my note was about effect of the core as an EMI shield for E fields. Since ferrite is a poor RF conductor, regardless of closed (potcore) or open (EE, UI, UU) shapes,
Mind that "poor" of course depends. It's not so bad of an AC conductor, thanks to the having resistivity, and the relatively high dielectric constant as well. It can be worthwhile to ground the core, shunting the small capacitance from winding to core. This is especially useful around very high impedances -- high voltages say, indeed you can get arcing to a core just fine, HV doesn't mind!
So, high impedances, electrostatic fields.
For E style builds, this can give some priority as far as which winding (and which end of the winding) goes down first, nearest the core (highest capacitance through the bobbin).
Conversely, E fields are trivially shorted out by any shielding, and B fields somewhat less so. Core grounding, and flux shorting band, may be irrelevant inside a metal enclosure. Not to say they don't still have beneficial effect, just that it's probably not worth it. So they're most worthwhile to consider, when doing an open-frame or plastic enclosure design. :-+
Tim
-
Hello again...T3sl4coil, fine notes...
As my work was in epoch 1970s...1990, we did not have any simulations. Sims are fine but do not replace actual experience and winding and testing the hardware.
Instead, we made designs on paper, wound prototypes, ran tests. One or two prototypes were sufficient to trim the design.
We did, however use Magnetics Designer, fine software goes back to 1980s (DOS) days. You could check fit, bobbin fill, number turns, etc.
http://www.intusoft.com/mag.htm (http://www.intusoft.com/mag.htm)
RE HV windings we placed the primary first near center leg, secondary after, and sometimes separated P-S on different ends of the bobbin, with a barrier or bobbin flange.
Did not use grounded cores, we had no issue to pass EMI CISPR, medical UL. (PQ shapes).
ref on magnetics design:
Magnetic Components Steve SMITH
Electromagnetic Compossibility, Heinz M. Schlicke
Enjoy !
Jon