Maximizing Heat Dissipation in High-Power SMPS Design

[Salvador]

Ok folks, I'm back, now here are some fun observations.I now installed the 7815 voltage regulator , it's working nicely , the mosfets driving steady.even when one of the older IR2110 stopped working (it went through alot) the regulator dropped to 2.3 output voltage , then i unplugged the IR2110 and measured the regulator voltage again and saw it coming back to 15 volts , the IR developed a fault somewhere inside but the nice thing about the voltage regulator is that it sensed that fault , probably due to overcurrent and dropped the output saving itself.
Now I changed the IR2110 to a fresh one and the whole show went on.This time I attached a lamp load and a water heater , the old ones which you put in a cup or a pot and just boil some water.the water was heating for some 15 minutes atleast togethr with the lamp.the exact power used is hard to calculate due to the uneven resistance due to heating.

now here's the interesting thing , after all that time i started getting some cracking noises as the water approcahed the boiling point.i disconnected the power and finger measured the heating of various devices , here are the results, the mosfets and their heatsink -cold.everything else was also cold.
two things were hot and one was extremely hot. the output high frequency rectifier diodes (mur1560) with their heatsink were kind of hot but not to the point of device SOA.
Now the 3c85 rated ferrite core itself was so hot i couldn't hold my finger on the ferrite material.and it also made the cracking noises.
what would this indicate, that the ferrite itself is so hot? as much as I can tell either the frequency is too high , which is not the case as it's about 50khz (measured) or the power transmitted through the core was too intense for the core size?

now I am thinking maybe the ceramic cap in series with the resistor across the primary is bad , but I am not sure whether that would have some profound consequences on the core heating.
also i switched the device on with the same load for a few mins and the output rectifiers (mur1560) are running quite hot , is i normal for high power high frequency diodes to run hot under high load? I'm asking this because low frequency mains rectifiers almoust never even heat up under any kinf of load.
im using a standard aluminum plate heatsink for the output rectifier , like the ones that are used in ordinary PC psu where the switching semiconductors are attached to.the particular one is actually from an old pc psu.I think you know about what kinf od aluminum heatsink I am talking about.

 

[Baluncore]

now I am thinking maybe the ceramic cap in series with the resistor across the primary is bad , but I am not sure whether that would have some profound consequences on the core heating.

You are going on a witch hunt. So long as you blame the ceramic capacitor, which is unable to heat the core, you will ignore the transformer.

Is the transformer core hot because it is being over-driven to saturation, or is it being heated by the windings? It is best to design the transformer so the core and windings both run at the same temperature. I think you need less turns, but of thicker wire. Make sure the wire fills the available space in the core.

Maybe you should do something actually meaningful, like measure and report the actual output current and voltage.

 

[Salvador]

the witch hunt is mainly because I don't have a decent scope.lets be real without one building something like an smps is quite a task.but so far I am actually doing pretty good, except for the gate overvoltage mistake and as one good old song goes " with a little help from my friends" ...

Somehow I want to agree that a few turns less and a thicker wire would help.when I made the transformer I used it's own original wire (litz), the one which was on the core from factory, I just reconfigured the turns ration to suit my needs in terms of output voltage.

measure the output current and voltage under what load? I always chekc atleast the voltage , as that's the bare minimum one can do, and when I was boiling my water the voltage stood steady at about 134 volts. +-1 volts as the AC mains ' walks" a little.even at the very last moment of the transformer core becoming so hot as to crack the voltage was still Ok, as soon as I'll get back to the scope I will atleast measure the secondary waveform if not the pirmary also, and then i'll be sure what happens there.
without the scope measurment it's always going to be more or less a guessing thing.

Oh by the way , if the transformer isn't saturating and the wire for example is the correct size and turns ratio is it possible for a ferrite core to heat up simply due to the transformer driving a static heavy load for a considerable amount of time?
atleast soft iron mains frequency transformers do heat up under load , sometimes even till the point were it becomes hard t hold my hand on the core iron , I for example have a japanesse made transformer which i now use in one amplifier and sometimes under heavy load for long hours the transformer heats up quite a lot , hence I added an aluminum heatsink which ventilates the heat into the chasis.

 

[Salvador]

oh and one more , I was searching google without much luck, now everycore size and material has the maximum power limit it can transfer , how could i determine the limit beyond which either excessive heating or losses occur?
I think it has to do with core size and its maximum magnetic field density it can take up until saturation point.and frequency or rate of change ofcourse.

 

[Baluncore]

measure the output current and voltage under what load?

Exactly. The voltage may be regulated, but what is the output current? How many watts is it really delivering?
The output voltage and current gives the load. W_Load = V * I. R_Load = V / I.

The question comes down to... What is the ratio of W_core to W_windings ? You will need a thermometer.
Once your core has cracked, do you expect it to have the same magnetic characteristics ?

 

[Salvador]

Well , except for the times when i haev measured the output amps directly I've always calculated using the resistance i have and the voltage which drops across that load resistance. and the average comes out as 500w. on my typical loads.

the core hasn't cracked , it was just expanding and so the sounds originated , although I am not 100% sure and will have to check.I would like to put a thicker wire on the traffo, although I would really like to understand what is the power limit this core can deliver as there is no use in putting a wire that;s too big given that still the system will only be as powerful as it's weakest point is.

also I'm curious , I don't have a thick litz wire but could I use ordinary soft mains copper wire with the insulation , the question is about the skin effect since in litz wire each of the individual strands is also lack coated and insulated from the oher strands but in a typical mains wire all the strands are bare to each other so in the case of high frequency wouldn;'t the stranded wire be just as bad as a single monolith wire if the individual many strands don't have their separate isolation ?

 

[Baluncore]

Is the wire the source of the heat ? Or is it the magnetic material ?
Twisting bare copper does not make litz. They must be separated.
You can make your own litz wire by twisting more strands of magnet wire together.

 

[Salvador]

it will be hard to say whether the wire or the core is the source of the heat, since the bobbin surrounding the core on which the wire is wound, is kinda of really encapsulated , well the core does get hot but it happens with time and it could be the result of the wire getting hot first or maybe not.

well ok I've taken the transformer open mayeb some of you could advice me how many turns to put on the pirmary knowing the schematic and that the ferrite material is 3c85

 

[Baluncore]

Is the transformer core you used really a Siemens ETD59-31-22 ?
Where the schematic shows "ETD59 (1.0) Siemens" does that specify a core with a total 1.0mm air gap ?
What air gap does your core have ?
You might need to insert an extra spacer in your core to set the correct total air gap.

 

[Salvador]

now I am not sure waht core it actually is but I am definitely sure it isn't an etd59 core , nor do i think it has any airgaps ,if i remeber correctly it came from an old philips tv. but the core is quite large in area definitely larger than other CRT's smps that i have seen or repaired.
all i know there are the letters 3c85 writte on the side of it which describes it's ferrite properties.

luckily i have pictures that i just made so judge for yourself.
also i counted and the turns ratio for pirmary is probably to blame , as it's about 36 +- turns maybe even 38 as i could have missed some or added some anyhow , i think it stoo high of a count , the blame is that previously i used this thing for a low power mosfet smps and it required more tunrs but here the power driven through the primary is high so the core probably goes into saturation and the wire into heating.
what do you think , but the question is how many turns should I put on the core?
something inside me tells about 20

 

[Baluncore]

if i remeber correctly it came from an old philips tv.

That looks like an iron powder core for lower frequencies.
Get the right part. Set the right gap. Wind the specified number of turns of the specified size wire.
Stop wasting our time.

 

[Averagesupernova]

Stop wasting our time.

-
I will admit that this thread has been a bit irritating but seriously, who is holding a gun to your head for you to stay here? I consider the possibility that I am wasting my time every time I reply to a thread.

 

[Salvador]

let's be honest I don't blame anyone, we all get our bad days, but i think we all have wasted so much of our time each one of us in our lifetime's that replying to a guy who tried to make a simple smps and uses parts of the hand is merely a coffe pause think process than a time wasting thing.

anyhow , the problem for me is too many primary turns, I've did some formulas put in some numbers measured the core dimensions and if everything matches i should put somehing like 12 to 15 turns for the primary , since i have only the original litz wire that came with the transformer I will wind it double like two wires in parallel.

one thing that Baluncore mentioned about the core material, well I'm not perfectly sure wheher it's iron powder or not and yes i am aware that Iron powder ir meant for lower frequencies, but on the side it says 3c85 and when looking in google 3c85 and 3c90 have ferrite parameters.

 

[Salvador]

you know the internet is not much of ahelp anyways so i don't expect anyone here to share even the slightest comment maybe not even one which says about time waste , but here's what i got out of my project.
I rewound the transfomer , I used 12 turns for the primary , I took the litz wire i had and doubled it so each turn has two wires in parallel, so i then made my secondary also with two wires in parallel.
I added a light load at first and no voltage drop , the total voltage of the secondary being 160c DC.
then I added a heavier load , like a water heater and a lamp and i was impressed , still no secondary voltage sag not even one volt.the output was hanging at 160volts steady. i let it sit there ofr a while , but i knew something is going to happen with those god damn no good semiconductors. and guess what , magic smoke after a while.
the IR2110 did what it does best and I have observed this many times, it blew itself nicely.the mosfet heatsinkwhich is pretty big was also quite hot , the transformer which was hot before when the mosfets were dead cold was now cold and now the mosfets were hot, they exchanged roles this time. :d

as for the semiconductors are all dead this time , everything is shorted starting from the voltage regulator both IC's and the mosfets .

If anyone still cares to " waste all his life..." you can share a comment on wheher you think what caused this.
My guess is not enough primary turns for the transformer.previously it had too much so it overheatd but this time it has a bit less than what's necessary. So i guess I will just increase the primary turns and try once more.

 

[Averagesupernova]

Do you have a way to measure current in the primary? Had you done this before you would know how your changes affected it.

 

[Salvador]

yeah I agree working with no instruments is a blind game , just right now i don't have the money to buy myslef fancy measuring equippment , although I will see maybe I can get a simple ammeter from someone.
Anyway, surely the primary current might have been twice as what it was before beacuse the transformer secondary was giving out wuite a load with no voltage decrease, surely that power must have come from somewhere.the transformer performed nicely , I mean there was no sag of voltage whatsoever this time.
I used both two parallel litz wires for primary and the same for secondary.
basically there's only two chances , either the mosfets were overloaded , although the IRFP 460is rated at 18 amps which is not that little , yet I might have been close my secondary load was a mains230 volt 1000w water heater and a 230w light bulb in parallel, the output was a steady 160v DC.

now the other posibillity is that i need a few more turns of primary , I mean the mosfets worked fine for some time until they overheated.if the amperage would have been too high to begin with wouldn't they fail in seconds ?atleast when the voltage is too high for them they fail in a split second.

oh just a quick question , since the transformer is already wound nicely , if I come to the conclusion that the primary needs some additional turns can I just wind those turns over the secondary ? using proper insulation between ofcourse.I mean just take one of the exiting primary wires and continue the winding over the secondary and then exiting that wire again , I think it should't make any difference.

 

[Averagesupernova]

I won't answer your question about winding transformers. That is a subject I wish I was more informed about along with about a million other subjects. Concerning ammeters. An ammeter can be thought of as a sensitive voltmeter in parallel with a very low resistor in which the current to be measured is passing through. So get a couple of high wattage low ohm resistors, put them in series with the primary and use your scope to probe each side. Again, be careful where you ground your scope probe. I remember a time when I was in your position building and troubleshooting things that did not work. I did not have the luxury of having some of the test equipment that was necessary at the time so I improvised when I could. This is one place that you could improvise.

 

[Baluncore]

What happens if the cross section of the core is too small?
The core saturates earlier and at a lower current.

What happens if you have too many turns?
The magnetic product of (amp * turns) saturates the core.

What happens if the core saturates?
The primary current rises steeply, limited only by resistance and may destroy the drivers.

What happens if you try to increase the load without over-current detection on the drivers?
When the increased load saturates the core, it destroys the drivers.

Other than over-current sensing, how can you help prevent or delay saturation?
Set a driver on-time limit, use a bigger core, or use a core with an air gap in the magnetic path.

Why did the original design specify Siemens ETD59 (1.0) 12t + 12t CT ?
Because the material and dimensions optimised power output of that half bridge design.

What is the maximum output power that an old TV core can handle?
Probably the same as the TV it was originally fitted to.

How do you estimate relative transformer power capability?
Measure the cross section of the magnetic path of the core, it is usually equal to the total of the side sections.

Why do you illogically want more turns on your transformer?
Because there is room for more turns and you know that core utilisation is important.

Is it easier to wind 12 turns of thick wire or 24 turns of thin wire?
Less turns of a thick wire or litz bundle is much easier to wind.

Why do you not use thicker wire or the correct core section and material?
Because that would make it too easy and you need more failures if you are to learn.

1. The most common problem area in switching designs is the inductor and the most common difficulty is saturation.
An inductor is saturated when it cannot hold any more magnetic flux. As an inductor arrives at saturation it begins to look more resistive and less inductive. Under these conditions the current flow through it is limited only by its DC copper resistance and the source capacity. This is why saturation often results in destructive failures.

 

[Salvador]

ok guys , thanks for still stoppin by this thread.

So an overall conclusion here is that too many primary turns cause core saturation beacuse of more amp turns create more flux for given area, too few turns would probably result in not enough flux and drop in output power??
Too short of a deadtime can cause shootthrough and or saturation of the core ,. too long deadtime , again drop in output power and the core not being fully utilized.

now here's some questions, could the increase in wire thickness cause core saturation even though the turns count was decrease by 2/3? I assume due to a twice as large wire diameter causing more current to flow which again causes a stronger field more flux = core saturation?
also is it correct that the heavier the load on the secondary the sooner the core saturates or no?

well i think the thicker wire and decreased primary turns count was the right way to go beacuse now I am getting no voltage decrease under load , probably because the single litz wire with too many turns caused excessie resistive losses and not enough flux under load.
i think i should increase my deadtime even more and closely monitor what happens , I might instal one of those cheap thermoswitches on the mosfet heatsink and connect it in series with the driver IC's supply voltage rail. say a 90 degrees celsius switch which cuts the mosfets off when the heatsink reaches 90 degrees.you think i should go for a lower safety temperature?

surely probing the primary would be nice , well as for the ammeter i think that I can get and will.Also I am thinking , right now I'm having as I said 12 turns of two litz in parallel primary, would putting afew extra urns really help in anything , seems like the answer would be no , maybe i first really have to put my deadtime to maximum to prevent core saturation if it's happening.
But then here is the weird part , before I had like 36 turns of single litz primary , and whatever was happening in the core , the only symptoms were slowly heating core until after some time it got hot and also sagging output voltage under load but the mosfets were dead cold , now the transformer seems to be fine yet the mosfets are hot as hell.
I mean if the core was still saturating and if that was also the cause for mosfet failure due to overcurrent on each cycle after saturation then why the secondary performed good under load?

 

[Averagesupernova]

More of: "If THIS were happening, then would THAT happen? What if I increased THIS? Would THAT change?" You don't know any of the things it takes to make a determination of what you should be changing. You don't know if the core saturates or not. As far as I know a core will saturate with no load easier than loaded. Please someone who knows better than I correct me if I am wrong. Have you run this supply for an extended period of time with no load and observed results? Unless I am mistaken I don't think you have run it long enough to be sure that the load is what is causing your problem. Don't you think that if you found that you had the same problems after running the supply a while with no load it would be a little easier to start troubleshooting at this point? How about running it for a long time with no transformer hooked at all?

 

[mheslep]

If you want to wind up an old large iron core, 60 Hz transformer based, non-switching power supply you can get by with a volt/amp meter. You can not build and diagnose a high power 50 KHz switching power supply without a scope, any more than you can hammer nails without a hammer. The goal was to learn about SMPS operation. Even if you happen on a properly designed core, with this try-something-and-fire-it-up,measure-nothing approach you'll learn no more about SMPS design than you would by rolling dice.

 

[Baluncore]

An optimised switching supply MUST be engineered, it cannot be built using Monte Carlo techniques. You do not have the necessary equipment to work safely and successfully on the development of a high voltage half-bridge supply. Without an isolated development platform you can't measure enough to know the actual situation. You also require a fast and accurate over-current trip circuit to protect the driver transistors.

Whenever you get anything working you increase the load until saturation of the transformer destroys the driver transistors and ancillary chips. Your blind and undisciplined approach guarantees repeated failure.

 

[Salvador]

Speaking in terms of one of the most popular music genres I would like to say this little update is "4 all U haterz and non-BELIEVERS" out there.

I did a bit of reinforcing on my PSU in the meantime I installed the psu board into the aluminum mainframe which is essentially the amplifier box also housing the amp channels etc.
Now I added quite a bit of heatsinking on various crucial elements, like the mains rectifier (actually got pretty damn hot when I was loading the psu with 1.2kW), connected the output rectifier heatsink to the aluminum chassis which I made out of aluminum for exactly the resons of heatsinking.made a heatsink contact from the mosfets to the chassis using double isolation (first isolator between mosfet and one heatsink then another layer of insulation between that aluminum pad and it's connection to the chassis to which everything is heatsinked.
I even took an old CPU aluminum heatsink and attached it to the transformer that got hot under heavy load and now it's only moderately warm even when I was driving a 1kW+ load on the output.I used some aluminum cooking foil as the "thermalpaste" to even out the surfaces between the transformer ferrite and the smooth cpu aluminum block.

My monster load is two old clothing irons connected in parallel , each rated for 1600watts mains 230 volts , they pose a resistance of about 20 ohms when hot to my psu output and doing some calculations I concluded that the psu supplied at its peak about 1.2KW into that load since the voltage was 155 volts approximately, when the load was connected.only a 10 volt drop from my usual 165 volts of DC output under lighter loads such as a water heating coil and a halogen 230w bulb in parallel.I'd say not bad.given that it's not regulated , since the feedback was causing me trouble with voltage swing and unstabillity I just disconnected the feedback IC and to my surpise the output is very steady it only fluctuates when the mains itslef sometimes fluctuates.but since i have pretty decent mains filter caps they kinda compensate for that.
those irons by the way were submerged into a metal can filled wth cold water and the water started to boil about a bit more than a minute after power on.also my mains cable from the wall socker got pretty warm I'd say during the heavy loading.even the lamps on the ceiling went a bit dim.
I'm no that goot with AC power comsumption calculations mabe one can estimate based n my output numbers the approximate power drawn from the wall socket ?
the output was 155 volts DC over about 20 ohms of resistance the maths on DC side give about 7.7 amps.Now I'm finishing my other transformer and will soon be putting that in and also heatsinking it to the chasis.I laready have the other half bridge mosfets installed and have the place for the second IC ir2110, a small question if I may , is it okay for one SG3525AN to drive two IR2110, they all share a common small psu.but anything other than that the IR2110 are separate only that they share one psu and the outputs from the SG3525 chip which are going to both the inputs of one and the other IR2110.
the half bridges themselves are separate units and will be driving separate amplifiers, one half bridge driving two channels that are also able to bridge mode.

The thing I have learned so far is that when designing an smps with considerable output power one needs to have heatsinking with quite some capacity, I mean everything heats up , even the mains rectifier , actually I'm surprised but of all the things giving me heat the switching mosfets are the coldest ones of them all.
I increased the frequency a bit over the specified 50khz and increased the deatime with my reostat so that might give my mosfets a better waveform to keep them cold.
Also even though I have done both of my transformers with double litz wire in parallel both for primary as for secondary i feel the litz wire couldbe even thicker and it would only help.
Also a larger transformer core than necessary or specified would't harm probably either, giving some extra headroom for temperature and saturation if it happens.