SMPS transformer core getting hot advice

[Salvador]

I had a bit of rattle about this here before but I hope you folks ain't fed up with me yet.:)
So after some trial and error I've built a working smps for my amplifier , it's pretty high power ,when testing dummy loads i got up to 2KW , since I don't need more for my project I never went further as I had no need.
It's not regulated but it dropped some 15 volts at max dummy load so works for me.

The problem is this , after trying out some leftover ferrite cores simply to understand that there's no luck with them I bought a new EPCOS made core.It's the ETD59 , the material is N87 with permeability of 5300nH.
the size of the core is 59.8 x 31.2 x 22.1mm

Based on other smps using the same materials I made 12 turns in the primary (maybe 11 can't remember)
And the secondary even though doesn't play a role in this problem is some 2x7 turns.the primary is wound with some thick litz wire salvaged from some old smps.Haven't exactly measured but I would say some 3 mm2.
the transformer frequency is +- 50 Khz, driven by a pair of half bridge IGBT'S. The problem is this , under load the core heats up , I kind of predicted this so I made my heatsink such that it also touches the core , in other words I have an aluminum box with large surface areas and everything is thermocoupled to the box.
But still I think the core gets too hot even with lighter loads , something probably isn't right.The windings themselves stay cool , only slightly warm but then it could also be the effect from the heat given off by the core.

When I was younger i made a few mains transformers with hand and I once got my turns calculations for the primary wrong and ended up winding fewer than necessary turns , but back then the windings got warm not the core , since there wasn't enough inductance and the wires went through some resistive/ohmic heating on each mains cycle.

So I wonder what could be the problem , my own guess is too few or too many turns in the primary, rather too few but I'm not sure.

 

[Baluncore]

... , under load the core heats up , I kind of predicted this so I made my heatsink such that it also touches the core, ...

A heatsink on a ceramic core will likely fracture the core due to differential thermal expansion between the hot and cooled regions. Cracks in the core will reduce both thermal and magnetic conductivity. You are better to cool the entire transformer in free or forced air.

So I wonder what could be the problem , my own guess is too few or too many turns in the primary, rather too few but I'm not sure.

Using thinner wire will permit a smaller window and so a shorter magnetic circuit on a different core.

An optimised design will have a core temperature the same as the windings.
If the core is very hot and the windings cool then either;
1. The core material is not optimised for the selected frequency of operation.
2. The flux in the core is too great. Increase the section area of the core.

Have you investigated adjustment of the air gap in the core assembly ?

 

[Salvador]

Well my core has no air gap, the two E sides match together with no gaps and I did not make one since according to what I know it's not needed.I just put them together and the whole core is then kept together by a adjustable metal squeezer so the core is pretty tight held to the side of the heatsink which also works as its support.

the frequency range is noted as 25 to 500khz if I'm correct so 50khz should be good enough.
Not sure what to do , the thing is the very transformation is working good enough, the IGBT's run barely warm , the heatsink for the smps in general is barely warm , everything is barely warm except the ferrite core which after some 20 mins of music through a light load speakers got pretty hot to touch.

I think I have 11 turns of primary now , should I try to take one off? Would not be that complicated since I made everything such that I can access the critical parts.

 

[Baluncore]

EPCOS ETD59 in N87 is available with a gap of 0, 1.0, 1.5 or 2.0 mm. One or both of the E core halves provided may have the gap as a shorter centre pole. You will probably not notice it without a straight edge across the faces of the E core poles.

The inner pole is shortened so as to avoid flux leakage from an external gap. You can experimentally adjust the gap by using a shim of non-magnetic material that will survive the core temperature.

 

[Salvador]

Oh now I remember , the cores did come in air gap sizes and I bought the one that has no airgap.
prior to assembly which is rather straight forward , I looked at the two half sides put together and there wasn't light coming through the middle pole on which the bobbin sits so for all practical considerations there is no air gap in this core.

 

[anorlunda]

http://www.brighthubengineering.com/diy-electronics-devices/96783-designing-your-own-transformer/

That link has good info on transformer design. Procedures, formulas, data tables.

 

[Nidum]

Have you considered the possibility of there being a shorted turn somewhere in your metalwork ?

 

[Salvador]

What do you mean by shorted turn in my metalwork? If you think a shorted turn in either the primary or secondary winding , then no because given such few turns I would immediately notice the difference in output voltage.
Output voltage hasn't changed since the first day I wound this particular transformer.

thanks for the link @anorlunda , I will check it out.

 

[Nidum]

A shorted turn results when any combination of metal parts in the transformer assembly forms a complete loop around the transformer core . Most commonly loop consists of the transformer centre bolt and external connections via mounting brackets , chassis , earthing straps and heat sinks .

 

[Salvador]

Today I plugged it it once more and listened to some music , wasn't long some 15 mins. maybe bit less and the core got to the boundary between very warm and hot.
So before I take the transformer out and start solving the problem would your 'consensus" be that I need to add an extra or two turns to the primary winding?
I now have 11 turns in the primary.

 

[Baluncore]

I would first put an 0.2mm insulated spacer between the outside limbs of the E cores and test it again. That will identify if the gap is important in the design.

 

[jim hardy]

So before I take the transformer out and start solving the problem would your 'consensus" be that I need to add an extra or two turns to the primary winding?
I now have 11 turns in the primary.

Can we put a number on your flux density ?
On back of an envelope
how many volts per turn from your transformer? 200/11 = somewhere around 20 ?

Now to get 20 volts in one turn requires by e=ndΦ/dt
around 20 Webers per second ? Twenty microWebers per microsecond ?
At 50 khz, a half cycle is ten microseconds ?
So your rate of change of flux is 200 microwebers per half cycle? Meaning flux likely swings from -100 to +100 microWebers?
What is area of your core ?
Flux density is ~ 100 microWebers divided by that area...

Repeat with your real numbers...

How does your result compare to this rule-of-thumb 500 to 2500 Gauss from Magnetics ?
http://wcmagnetics.com/wp-content/uploads/2015/02/appnote11.28.10.pdf

Choice of a Core Material
–––––––––––––––––––––––––––––––––––––––––––
––––––––––––––––––––––
SMPS transformers are typically not saturation limited, but they are loss limited. Most all SMPS transformers use man-
ganese zinc ferrite cores because this family of magnetic materials has the lowest losses at SMPS
frequencies, and is relatively low cost. Manganese zinc cores typically have a saturation flux density of
4000 to 5000 gauss, and SMPS transformers are typically designed to operate between 500 and 2500 gauss. Even
if most SMPS transformer designs use a manganese zinc ferrite, within this family of materials there is considerable
choice available to the designer. Manganese zinc ferrites are simply not all the same and it is necessary to understand
their properties in order to make an informed choice for any SMPS desig

That's 0.05 to 0.25 Teslas if i remember my SI ...

Better calculation procedure here but it'd take me a week to learn.
http://ecee.colorado.edu/copec/book/slides/Ch15slides.pdf

old jim

 

[Salvador]

Since it's a half bridge and my mains is 230 AC , the rectified DC is 325 volts of which the primary sees only half so about 160/11 would be closer.
I will try to do the maths after I come home after two days.

Well I like your suggestion Baluncore , mostly because it's the first easiest things to do.I wonder what kind of material I could use for the spacers to separate the side legs to get the air gap in the middle one ?
Ordinary cardboard? thick paper?

Also can you briefly explain what does the air gap in a transformer actually does? All I know in low frequency transformers or audio output transformers it only decreases performance.I assume since air is much much worse at conducting flux or permeability the air gap would kind of "brake" the magnetic field a little?
Isn't the magnetic field also stretched in the place of air gap due to the field lines being concentrated in the core but as they leave they spread?

 

[jim hardy]

160 / 11 = 14.5 volts/turn
14.5 microwebers per microsecond
145 microwebers per half cycle is Δflux
half that is 72.5 microwebers

22.1 X 31.2 mm is 0.000689 m²
72.5 microwebers divided by 0.00068952 m² = 0.105 Tesla ? 1050 Gauss ? Check my arithmetic.?...

Does your core datasheet suggest an operating flux ? I couldn't find N87 curves.

old jim

EDIT found one
http://en.tdk.eu/blob/528882/download/4/pdf-n87.pdf

 

[Salvador]

I will continue this after I get back as I have to go on a small trip today.
My parameters in short are 50khz , 160v primary with 11 turns , the duty cycle is rather large , almost to it's fullest which would be 50% if I'm correct , well that's about it , yes it's an N87 core made by EPCOS.

 

[Tom.G]

Unless I messed up somewhere, I get 40W to 50W in core losses, which would explain the temperature rise. This is based on Jims calculation of core area, the core diameter in the OP, and the graph of losses Jim posted. Please double check this, it's late & I'm tired.

 

[Salvador]

thanks Tom for contributing, About 40/50watts of core loss is overly high for such a small ferrite core which cannot dissipate heat not nearly as good as larger iron cores and even then the losses would be bit high.

I wonder Tom can you write down the math that you used to come up with these losses ?I want to understand which variable has to be changed to get smaller losses as I am sore these cores are capable of some 1KW in power through without getting so hot.

 

[jim hardy]

I had better luck with searches this morning

Well now

72.5 microwebers divided by 368 mm2 X 10⁶ mm² / m³ = 0.197Tesla ? 1970Gauss ?

Which puts us on the ~200 milliTesla curve instead of where i drew the oval in earlier post

in fact let's use the TDK N87 curve i stumbled across just a couple minutes ago
it's from their 625 page catalog at http://en.tdk.eu/blob/519704/download/2/ferrites-and-accessories-data-book-130501.pdf , page 84
and there's tons of info there, download a copy for your e-library...

maybe 275kw/m³ ?

275 kw/m³ X 51200 X10^-9 m³ = 14 watts ?
Still seems a lot for such a small core.

You'd better check my arithmetic.
They say those curves are for sinewave and i don't know how to adjust them for square wave.
But this seems a reasonable approach for making an estimate. It's not like we're building a Steinway here .

I learned a lot today, thank you Tom.G and Salvador !

old jim

 

[Salvador]

Well you don't have to say thank me, Jim I'd rather have to say thank you as I am the one puzzled by this.

 

[jim hardy]

Well you don't have to say thank me, Jim I'd rather have to say thank you as I am the one puzzled by this.

@Tom.G Does this look right ?

@Salvador I don't know how difficult it is to add primary turns.
You've got me curious so i hope you try it.. Is this heat just from flux amplitude or something else ?

old jim

 

[jim hardy]

But still I think the core gets too hot even with lighter loads , something probably isn't right.

Great observation.

If it runs hot unloaded that's a clue because flux is proportional to volts per turn .
To first approximation, Current heats the windings and voltage heats the core.

 

[Salvador]

Well I'm always intrigued about physics stuff , also I will most definitely try out adding primary turns , I think I will do it tomorrow , it's not that hard of a fix in my case as I can access the transformer simply since when I built the smps I made it an easy accessible part as I knew there might be readjustments needed along the way ,of all the things I don't like in smps there is one i simply love , it's the fact that changing the transformer windings is so easy simply because all you have is few turns of wire around a small sized piece of ferrite and that's all.so I will probably wrap those turns as I drink my morning coffee.
I'll come then here to tell the results of that .

On the other hand I remember winding my audio output transformer not long ago which took me about 3 days with nothing but hands as I had no tools but only a wish to resurrect the old tube amp I had that other thread about.That was painful both mentally and physically.

PS , should I put 16 turns in the primary or bit less?
Also there is this thing I'm not getting , maybe due to the heat I'm having here right now, why does it happen so that putting up more turns would solve the problem , because you just said and I know that too , that more turns equals more flux per given core area so putting more turns will increase the flux density atleast I think so ,then what is the thing that heats the core in my case if not flux density ?

 

[jim hardy]

because you just said and I know that too , that more turns equals more flux per given core area so putting more turns will increase the flux density atleast I think so ,

Sorry if i wasn't clear.

If it runs hot unloaded that's a clue because flux is proportional to volts per turn .

Volts PER turn.. as in $\frac {volts}{turns}$$\frac {160}{11} = 14.5$ volts per turn

$\frac {160}{16} = 10$ volts per turn

so adding 5 turns will lower flux to $\frac{10}{14.5} = 69$% of whatever it is now
and that semilog chart suggests losses will drop to more like $\frac{100}{275} = 36$% of whatever it is now
i'd think you'll be able to feel that much difference.

Of course you'll have to add secondary turns to get output volts back up...
....................

Volt-second product gives flux
it's one of the shorthand notations magnetics guys use so often they forget to explain it , so magnetics gets an unearned reputation for being complicated
from your core datasheet

applied volts(per turn) X duration of a half cycle gives Δ flux during that half cycle, that's what i did in posts 12 and 14
but that gives Webers(flux Φ) instead of Teslas(flux density B)
ts easier for me to keep my alleged thinking straight that way.

any help ? Are you getting a feel for flux ?
What's important is to not forget it's $\frac{Δflux}{Δtime}$ that gives volts, so Δflux is volts X Δtime .
Φ = ∫volts dt if you prefer.

I hope this works !

old jim

 

[Tom.G]

@Salvador
Post #3:

core is then kept together by a adjustable metal squeezer

Post #9

A shorted turn results when any combination of metal parts in the transformer assembly forms a complete loop around the transformer core .

Could the "squeezer" be acting as a shorted turn? Something is using a lot of energy if the Duty Cycle is at maximum.

@jim hardy Post #20
A quick look, seems reasonable. I'll go thru in detail later, a bit pressed for time today. Where would I find the volume of the actual core used?
RMS of Square Wave: Good explanation at http://masteringelectronicsdesign.com/how-to-derive-the-rms-value-of-pulse-and-square-waveforms/

A fine thread; we're all learning lots of good stuff!

 

[jim hardy]

Where would I find the volume of the actual core used?

Post 18 - i found the actual datasheet , i think, from TDK
It's Ve, 'effective magnetic volume of core' , just below point of red arrow, 51,200mm³

http://en.tdk.eu/inf/80/db/fer_13/etd_59_31_22.pdf page 2 of 9

.............................

Sine vs square: thank you -
RMS of a square wave is same as its peak so i think our estimate is probably closer than i'd hoped ! No fancy integrals required.

Any day i learn something is a good day. SMPS's are starting to look less fearsome ..

 

[Tom.G]

RMS of a square wave

Uhh, I think Duty Cycle is involved. See Eqs. (4) and (5) in the referenced article.
http://masteringelectronicsdesign.com/how-to-derive-the-rms-value-of-pulse-and-square-waveforms/

 

[Tom.G]

Post #17

I wonder Tom can you write down the math that you used to come up with these losses ?

Not anything that would be useful. I'm pretty sure I messed up the calcs in my post #16. Please disregard. I'm currently trying to match the @jim hardy calcs. To follow as soon as I can get them typed in.

 

[jim hardy]

Uhh, I think Duty Cycle is involved. See Eqs. (4) and (5) in the referenced article.

indeed

My parameters in short are 50khz , 160v primary with 11 turns , the duty cycle is rather large , almost to it's fullest which would be 50% if I'm correct , ...

uh oh i took that to mean each half of the bridge is on for 50% of the time.
i took his 160 volts X a half cycle at 50khz for volt-second product
so flux will be lower by however short of ten microseconds is his on-time ..

i'm stuck in old fashioned terminology from 1960's high school electronics days
a square wave has 50% duty cycle
if it's not 50% it's not square wave it's a pulse train

maybe we'll get to see a 'scope trace

anyhow thanks for the heads-up
looking forward to seeing your results

old jim

 

[Tom.G]

@jim hardy

I backtracked your steps and mostly agree. However our intermediate results differ in flux density by a factor of 100! I haven't carried my results further as it's a matter of 'just read the graph' from here on.

I feel like we are negotiating to find an acceptable compromise. Anyhow, here is what I've done this time. Please point out my errors.

I'm using the equation from http://wcmagnetics.com/wp-content/uploads/2015/02/appnote11.28.10.pdf, page 9, Equation 2.

B = [V x t_on x 10⁸] / [2 x A_e x N]

Where

B = peak AC flux density (gauss)
V = primary voltage (volts) . . . . . . . . . . . . . . . . . . . . . . 160V
t_on = primary switch on time (seconds) . . . . 1^-5 (10uS)
A_e = core area (cm²) . . . . . . . . . 368 3.68 cm²
N = number of primary turns . . . . . . . . . . . . . . . . . . . . 11

B= 160 x 1e-5 x 1e8 / (2 x 368 3.68 x 11)
= 160 x 1e3 / 8096 80.96
B= 19.76 1976 gauss

 

[Salvador]

No Jim you got it right the first time not the second time about the duty cycle.
Mosfet's as well as IGBT's make their switch on time directly from the signal wave that they are driven with at their gates respectively and since my scope is not for higher voltage i measured the duty cycle on the gates rather than the actual transformer primary as there isn't much difference , the waveforms are the same only amplitude differs.
So my waveform is close to 50% duty cycle for a square wave or in other words every half cycle is full , when the upper one ends the lower one starts immediately , there is almost no dead time only as much as to not cause switch overlapping.
If I take smaller on time and larger dead time or lower duty cycle I get more significant voltage drop at heavier loads at output so i just stayed with that.
Although I think the duty cycle is not to blame for this as with the previous ferrite cores i used here before i bought this EPCOS n87, one of them was actually very mildly warm.SO I think the key here will be the turns ratio,

stay tuned as I am now rewinding the transformer let's see what's the result.

 

[jim hardy]

I'm using the equation from http://wcmagnetics.com/wp-content/uploads/2015/02/appnote11.28.10.pdf, page 9, Equation 2.

A lot of magnetics manufacturers still publish cgs . SI Tesla gets rid of that pesky 10⁸

i love it when intuitive and book formulas agree .

@Tom.G Thanks !

ps

Last edited: Today at 2:54 AM

You too ?
'Tis a fine madness, is it not ?

old jim

 

[Salvador]

Ok, I'm back , here are some observations , maybe i will go into more detail later as this is the last day with high heat here and when temperatures are lower my head works better.

I rewound the bobbin.I put some 15 turns of primary instead of 16 as 15 turns matches an even number of secondary turns for the voltage I need , since it's rather messy to have say 17 turns of secondary which produces overvoltage or 16 which in my case would produce under voltage in case of a 16 turn primary.
So I put 15 on primary and 16 on secondary.
My observations are as follows , I'll start with the less dramatic ones.As awkward as it may seem but for the same loads it feels like my smps aluminum heatsink block is warmer than it was with the 11 turn primary.So it must waste more heat now. Atleast the volt turns are following some logic I put 15 turns on primary so it goes like 162/15=10.8 , so I should get about 10 volts per turn in the secondary winding.Which I get.

The ferrite core is still quite warm , there has been maybe a small improvement and the core isn't hot by touch anymore but it's temp is still i'd say some 50/60 degrees celsius.The savior for this is the fact that the core sits tightly to the common aluminum heatsink and so the heat slowly drifts away.
my whole box is one big aluminum heat sink so that kind of helps set matters evenly.

I will have to do some research more into this as the turns ration and core material might not be the only ones to blame
I did a test earlier with the 15 turn primary today I took out the fuses that connect my dc output rails from psu to my amp boards and simply connected a dummy load for test purposes which consisted of a water heater and a halogen lamp in parallel , for the output voltage of 160v DC the power i calculated from ohms law was somewhere around 300/400w I can't know for sure since the filament of the 200w bulb changes it's resistance upon heating up.
The core and the smps heat sink were warm with the core being a bit hot but still ok.
Now I connected back the fuses and listened to some music for some 30 mins maybe bit less and to my surprise the heat sink was almost as warm as with the previous dummy load , and the speakers were quite small and i was running only two out of 4 channels , the other two disconnected channels had warm heatsinks too which tells me that the idle currents are maybe higher than they need to be.
So maybe the core is also getting hot because the amplifier has 4 x500w boards with idle currents running and the smps output has 2x 15kOhm resistors as maximum idle voltage limiters etc , in other words it wastes quite some power even with light music loads.

Anyway the situation is better but not as good as your calculations showed , definitely not a 30% or so gain.
And the heatsink seems warmer now which I can't understand , maybe the ferrite core is better pushed to the heatsink now and causes the elevated temp, I'm not sure.I have added some pics just in case you Jim want to see my far from perfect and far from ideal home made magic box :D
Even though the wiring seems messy when I connected all the grounds in all the right places I miraculously have no hum or background noise in the output.
the amplifier boards are at the left side , in the middle there is the protection board and at the right side there is the smps unit.the pics are rather bad but i hope you are able to see the stuff and how its put.

 

[jim hardy]

The core and the smps heat sink were warm with the core being a bit hot but still ok.
Now I connected back the fuses and listened to some music for some 30 mins maybe bit less and to my surprise the heat sink was almost as warm as with the previous dummy load , and the speakers were quite small and i was running only two out of 4 channels , the other two disconnected channels had warm heatsinks too which tells me that the idle currents are maybe higher than they need to be.

Have you understood yet that flux depends on voltage across the windings not load current through them ?
So if it's flux the core should run same temperature loaded or unloiaded.

SMPS cores lose less heat as they warm up note 25C and 100C curves on the graph

and is this a metal loop as Tom.G asked?

Is the core pressed hard against aluminum ? I can't tell.
If so you might well be making eddy currents in the aluminum heating it right where core touches it. To test for flux fringing superglue a penny or piece of timfoil on top of the core and see if it gets hot. Beware , don't let it fall across something dangerous.

The core and the smps heat sink were warm with the core being a bit hot but still ok.

It is easy in the heat of the chase to not realize when we've achieved success. Got a contact thermometer ? Or a thermistor?

old jim

 

[jim hardy]

By the way - awesome project . Your homebrew SMPS is really impressive, as is that you don't get hum anymore.

An old engineer from Tennessee once gave me his litmus test for parts running too hot: "Does it fry spit?"

water heater

?? How big? They can be kilowatts.

Looking forward to more observations

old jim

 

[Averagesupernova]

"Does it fry spit?"old jim

^^^^Like^^^^