How to dissipate up to 1kW of heat on a floating ground linear supply?

[Bendba]

Hi,

Last week, I found that really nice 24V DC power supply made out off a big constant voltage transformer.
I've ben planning for a long time on building a good floating ground linear power supply. I got dozens of TO-3 NPN transistors, one nice lot is of 2n3773
So I might look into modifying the supply into a linear supply.

But that isn't really the question. The question is about heat dissipation. Is there a foolproof system to dissipate up to 1kw of heat without any grounding of the heat sink?

The strait forward aluminium heat sink with fans, mounting the heat sink on nylon or Teflon posts and screws but with the summer temperatures we get here, the heat sink would be about as big as the supply, and noisy.

The other idea is water cooling. But if I use main water, the water circuit as to be galvanically isolated from the transistors and I can't really figure out how to thermally couple them.

I could use an isolated water circuit but that complicates greatly the system whereas with tap water, I can just dump the water in the garden irrigation tank.

Is there any other good dissipating system?

Also, are there any operating temperature ranges that are better/worse at limiting thermally generated noise? Or is it the cooler, the better?

[rstofer]

There are mica heatsink insulators and plastic bushings to insulate the mounting screws.
These are for TO-3 but there are other styles:
http://www.mouser.com/ProductDetail/Keystone-Electronics/4636/?qs=sGAEpiMZZMsQtlBhqKq43cDG0TrwBSwA

1 kW is an enormous amount of heat to dissipate in a power supply.  I would consider a SMPS as a pre-regulator followed by some kind of linear regulator.  The idea would be to supply the linear portion with a voltage somewhat higher than the drop-out voltage but not so high that power dissipation becomes an issue.

[CopperCone]

unless you evaporate the water, its just moving heat from one place to another..

[mariush]

Wouldn't it be easier to unwind that big transformer and make taps at various points and switch between taps as needed to reduce power wasted?

[Bendba]

There are mica heatsink insulators and plastic bushings to insulate the mounting screws.

1 kW is an enormous amount of heat to dissipate in a power supply.  I would consider a SMPS as a pre-regulator followed by some kind of linear regulator.  The idea would be to supply the linear portion with a voltage somewhat higher than the drop-out voltage but not so high that power dissipation becomes an issue.

Yes, actually forgot to mention mica insulators.
Might have to make a custom heat exchanger for water cooling. (Better get the aluminum casting gears ready)

I know I could do with an SMPS, actually planning on adding that option to the output of the transformer as another selectable supply output. The all linear is more of to challenge myself, not to be efficient.

unless you evaporate the water, its just moving heat from one place to another..

Any cooling system is just moving the heart from one place to an other. What I like about liquid cooling is the heat capacity of the liquid. If my calculations are right, if the water enter at 20°C and exites at 50°C, I can dissipate 1kW (which is the worst case scenario, I don't think I'd be dissipating much more than 200W normally) with a flow of less than 35l/h. So, even 24h of use wouldn't use more than the amount of water used in the garden in one day.

I know, it's not efficient, for that, I'll work on an SMPS but after finishing my big Z80 project.
It's just one of those old project I want to do, just because I can 

The question is more, is there any good heat sinking option I should consider? Keeping an open mind.

Wouldn't it be easier to unwind that big transformer and make taps at various points and switch between taps as needed to reduce power wasted?

Yes, that could be an option. I got to do a lot more reading about those ferroresonant transformers to see if it could work.


Please don't think I'm ignoring constructive comments, they are all very welcome. I know it is a very inefficient system, just want to do it for the sake of it. You know, like some people put a V8 petrol engine in a Corolla, just because you can.
I don't want to appear like a stubborn person who doesn't listen.

[TheUnnamedNewbie]

As others have said, the main reason for water cooling is just moving the heat away to somewhere where you have more room to deal with it. This can be because you are limited by the thermal resistance of the material you are attached to (at some point, just throwing more copper or aluminium at the problem won't make it any better) or if you don't have the room to deal with it in that location (which is why it is used in certain supercomputers and RF amplifiers etc). By using water, you can move the heat away fast to a location with a lot of available space to put a lot of poorly performing and cheap heatsinks with little airflow.

Regarding thermal noise: As the name suggests, thermal noise gets better (ie, less) with lower temperature. Shot noise does not improve with temperature.

I wanted to post at this point but saw you added a reply:

If my calculations are right, if the water enter at 20°C and exites at 50°C, I can dissipate 1kW (which is the worst case scenario, I don't think I'd be dissipating much more than 200W normally) with a flow of less than 35l/h. So, even 24h of use wouldn't use more than the amount of water used in the garden in one day.

That would seem like a waste of water tho. Usually in water cooling you don't want too much of a differential - if you make the water flow faster you get a more even temperature, and less of a temperature gradient in the system. Perhaps looking at some computer-builder oriented water cooling stuff can be usefull. Those guys have stupendously large heatsinks for water cooling that can dump kilowatts of energy with differentials of just a few tens of degrees.

[nctnico]

I wouldn't use isolation material between the transistors and heatsink. You easely add 0.5 to 1K/W of thermal resistance that way. A piece of PVC or rubber hose between the tap and the heatsinks should be all the isolation you need. Tap water isn't that conductive. Still I would go for a solution with temperature controlled fans. Also make sure to have thermal limit switches on the heatsinks so things don't go bad due to overheating.

[fourtytwo42]

If this was me as a semi-permanent addition to my workshop I would make the heat do something useful, a simple enclosed heat exchanger plumbed indirectly into my hot water cistern tank using a time honored gravity feed system would work very nicely and if things are getting to hot run a bath I have run amps and psu's with open evaporative pot's way back in the 70's but the running temperature is to high and corrosion is a problem BUT with a closed loop system you could add a touch of glycol (vehicle anti-freeze) and keep the temperature (between baths) to no more than 55C or so.

[Bendba]

As others have said, the main reason for water cooling is just moving the heat away to somewhere where you have more room to deal with it.

...


That would seem like a waste of water tho. Usually in water cooling you don't want too much of a differential - if you make the water flow faster you get a more even temperature, and less of a temperature gradient in the system. Perhaps looking at some computer-builder oriented water cooling stuff can be usefull. Those guys have stupendously large heatsinks for water cooling that can dump kilowatts of energy with differentials of just a few tens of degrees.

Thanks for your input.
Indeed, the way I saw it was to take the heat away from the transistors and throw it out, along with the water, that's why I did a quick calculation with a slow flow. We use up a lot of water for the garden, unfortunately, we have to use tap water, we don't have any rain water tank.

As you mentioned, smaller differential would be better, then a closed loop would be the way to go (with glycol for a better transfer). On one hand, it complicates things because you can't just dump the water out. On the other, if the pump is isolated,it would be easy to make a loop that is not grounded.

(
Just a thought. If I were to use a closed loop liquid cooling. Would it make a significant difference, noise wise, to use a non conductive coolant, enclose the transistor heat exchanger in a Faraday cage and connect it to the rest of the loop with rubber hoses?
More of a theoretical question at this point and time.
)

I wouldn't use isolation material between the transistors and heatsink. You easely add 0.5 to 1K/W of thermal resistance that way. A piece of PVC or rubber hose between the tap and the heatsinks should be all the isolation you need. Tap water isn't that conductive. Still I would go for a solution with temperature controlled fans. Also make sure to have thermal limit switches on the heatsinks so things don't go bad due to overheating.

Interesting about the non conductivity of the tap water. I haven't done any measurement, I just assumed the water would be enough of a conductor to be significant.
The idea of not using isolation between the transistors and heat sink is why, if I was to use fans, I'd simply mount the heat sink on insulating posts so it doesn't ground to the case of the power supply.

The fan option isn't excluded of course, I'm more looking around for what I haven't thought of.

And of course, thermal switches and temperature monitoring/control are a must.

If this was me as a semi-permanent addition to my workshop I would make the heat do something useful, a simple enclosed heat exchanger plumbed indirectly into my hot water cistern ...

I would love to, unfortunately, we are in a rental and I don't have a lab, just invading the dining table. In an idea world, I would run two coolant circuits around the house to recycle heat and lack of heat to heat water/living space and keep fridges and freezers cool. But that's not an option at the moment.

[schmitt trigger]

For best electrical isolation, corrosion protection, and preventing water loss, you could use a mixture of distilled water and glycol, and use a radiator on a closed loop system.

[ZeTeX]

I don't understand why you would need to dissipate 1kW.
You said you have 24V DC power supply, and let's say that after rectification and filtering the voltage is at 24V max load.
You would need over 41A at 24VDC to get almost 1kW dissipation. Are you sure that this is what you want and you have over ~1.3kW transformer?

[Kleinstein]

With a simple rectifier and filter capacitor the power factor is not that good, especially with a larger transformer. So a 1000 VA transformer might be good for 500 Watts of DC power. So one might tend to overestimate the power. Alternatively one might consider at least some power factor correction - this usually already comes with some voltage regulation. Besides classical high frequency SMPS there is also the option to use a kind of SCR or 100 Hz switching (with MOSFETs instead of SRC) for a pre-regualtion. The older HP high power supplies used SCR's. It might be slightly easier with MOSFETs.

Water cooling is mainly closed cycle to remove localized heat to a larger heat sink. However it is quite some effort and one still needs the heat sink to cool down the water. Usually just air cooling with a fan is the easier solution In this case the heat sink can be inside the case and does not need to be isolated. For 1 kW it is relatively hard to getaway without a fan. It might work quite as long as the power loss is low.

The mica isolation is still an option, though not that attractive with the relatively high power density 2N3773. It might need a few more output transistors.

[DBecker]

There is much to cover, but I'm sure others will do some of it.  So I'll add only a bit.

Mica is pretty much obsolete for production.  But you are using TO-3 packages.. I guess you could call mica "period correct".

Water flowing in a through-drilled aluminum block will result in far less heat transfer than you are hoping for.  To remove 1KW you'll need multiple channels with 'trip points' to create turbulence.  For a one-off you'll have better results cutting multiple channels with a ball nosed end mill and pressing in copper tubing.

Water is conductive enough to cause a safety problem.  With a total loss system you'll want to ground the heat sink, and carefully monitor corrosion.  Even a little leakage current can cause water quality problems, including releasing lead from brass fixtures.

[jbb]

I wouldn't use isolation material between the transistors and heatsink. You easely add 0.5 to 1K/W of thermal resistance that way. A piece of PVC or rubber hose between the tap and the heatsinks should be all the isolation you need. Tap water isn't that conductive. Still I would go for a solution with temperature controlled fans. Also make sure to have thermal limit switches on the heatsinks so things don't go bad due to overheating.

You'll get away with it for a while.  But you're likely to get corrosion (and possibly leaks!). Also I wouldn't trust that for anything to do with human safety.

For best electrical isolation, corrosion protection, and preventing water loss, you could use a mixture of distilled water and glycol, and use a radiator on a closed loop system.

This is a classic coolant mix.  However, the distilled water can absorb CO2 from the air and become a tiny bit conductive, which might become an issue; I don't know (maybe glycol prevents this?)...  The large motor drives I've worked on have deionised water and include special ion exchange systems to continuously extract conductive ions.


Let's think about an air-cooled heatsink in a hot summer.
P_tot = 1 kW
T_amb = 35 C
Assume T_heatsink = 70 C
Thus \DetlaT = 35 C
Hence R_th = 0.035 C / W

That's a large air-cooled heatsink!  Or a moderate water cooling system.

Now consider transistors (2N3773):
T_junction = 110 C {leave margin below 125 C)
T_heatsink = 70 C
Hence \DeltaT = 40 C
R_th = 1.5 C / W {direct mounted, using On Semi data sheet value + 0.3 C/W for grease}
Hence P_single_device = 26W

That means you need 43 devices! (Adding 10% margin for sharing.) Argh! Danger Will Robinson!  Making those share load will be a nightmare.

Moral of story: dissipating 1kW is hard

[ZeTeX]

I wouldn't use isolation material between the transistors and heatsink. You easely add 0.5 to 1K/W of thermal resistance that way. A piece of PVC or rubber hose between the tap and the heatsinks should be all the isolation you need. Tap water isn't that conductive. Still I would go for a solution with temperature controlled fans. Also make sure to have thermal limit switches on the heatsinks so things don't go bad due to overheating.

You'll get away with it for a while.  But you're likely to get corrosion (and possibly leaks!). Also I wouldn't trust that for anything to do with human safety.

For best electrical isolation, corrosion protection, and preventing water loss, you could use a mixture of distilled water and glycol, and use a radiator on a closed loop system.

This is a classic coolant mix.  However, the distilled water can absorb CO2 from the air and become a tiny bit conductive, which might become an issue; I don't know (maybe glycol prevents this?)...  The large motor drives I've worked on have deionised water and include special ion exchange systems to continuously extract conductive ions.


Let's think about an air-cooled heatsink in a hot summer.
P_tot = 1 kW
T_amb = 35 C
Assume T_heatsink = 70 C
Thus \DetlaT = 35 C
Hence R_th = 0.035 C / W

That's a large air-cooled heatsink!  Or a moderate water cooling system.

Now consider transistors (2N3773):
T_junction = 110 C {leave margin below 125 C)
T_heatsink = 70 C
Hence \DeltaT = 40 C
R_th = 1.5 C / W {direct mounted, using On Semi data sheet value + 0.3 C/W for grease}
Hence P_single_device = 26W

That means you need 43 devices! (Adding 10% margin for sharing.) Argh! Danger Will Robinson!  Making those share load will be a nightmare.

Moral of story: dissipating 1kW is hard

Moral of the story: it's unpractical to dissipate 1kW. OP needs to find a solution not to how to dissipate 1kW, but how to reduce the power dissipation by 10 at least.

[tszaboo]

Here is the deal: You have a button, and you are trying to get a coat tailored for it. The TO3 transistors are oldscool. They dont have very good thermal resistance. I worked with FETs, that had 4-5 times better Rtjc. So you needed 4-5 times less component to build a system from it. Also, they came isolated by design.
Anyways, here is what you can do, if you absolutely must use these. Mount the TO3s on an aluminium slab. place the insulation between the slab and the chosen heatsink. This way, there will be a huge area for the thermal insulator so the relatively bad thermal performance will not matter as much.
It is not impossible to dissipate that much power. The biggest machines I worked on put 40 KW of heat into water, by precisely controlled transistors.

[ZeTeX]

Here is the deal: You have a button, and you are trying to get a coat tailored for it. The TO3 transistors are oldscool. They dont have very good thermal resistance. I worked with FETs, that had 4-5 times better Rtjc. So you needed 4-5 times less component to build a system from it. Also, they came isolated by design.
Anyways, here is what you can do, if you absolutely must use these. Mount the TO3s on an aluminium slab. place the insulation between the slab and the chosen heatsink. This way, there will be a huge area for the thermal insulator so the relatively bad thermal performance will not matter as much.
It is not impossible to dissipate that much power. The biggest machines I worked on put 40 KW of heat into water, by precisely controlled transistors.

Yes, but your machine probably wasn't a simple linear bench power supply, and you didn't have many options to reduce the power dissipation.
Different things.
if OP is not mistaken, and he wants 40A 24V linear supply, then he should of use a pre-regulator like Kleinstein suggested.

[jeroen79]

Wouldn't it be easier to unwind that big transformer and make taps at various points and switch between taps as needed to reduce power wasted?

Or add a switching preregulator/SCR to keep the dissipation low.

[C]

Last week, I found that really nice 24V DC power supply made out off a big constant voltage transformer.

A constant voltage transformer does supply some regulation on it's output.

First question I would think is "Does your loads really need better then this?"
If true, then you have more options then just creating a big heater with a linear regulator.

First, Some of the constant voltage transformers I know of have some electronics that could be improved. This control circuit runs at a lower power level, so an improved circuit would not generate as much heat.

Second, some better designed loads can function with the output of this power supply.

And for a final option,
Just use cold air or cold water.

[julian1]

Just get a few old Pentium P4s off ebay. Otherwise try some sha256 asic miners.

A common design for higher power linear supplies/loads - is to use aluminium box section - with the transistors mounted on the outside, and fans at each end to duct airflow efficiently through the system. 

Obviously some kind of fail-safe temperature monitoring and cut-out is required, so that if the fans die, it does not pose a fire risk.

Otherwise a SMPS (or buck converter immediately after the transformer) + LC low-pass filter just look a lot simpler and more power efficient - assuming the need for variable voltage.

[schmitt trigger]

This is a classic coolant mix.  However, the distilled water can absorb CO2 from the air and become a tiny bit conductive, which might become an issue; I don't know (maybe glycol prevents this?)...  The large motor drives I've worked on have deionised water and include special ion exchange systems to continuously extract conductive ions.

Thanks for the info

[Kleinstein]

The thermal calculation is not that bad. With a fan, one can keep the heat sink temperature to less than 70 C, at least most of the time. The maximum allowed junction temperature is 200 C - so reducing the temperature to only 110° C is a massive margin, especially if this is for full power on a hot day. Allowing up to 150 C on a hot day (and thus considerably less under more normal conditions) one could allow a pwer dissipation of something like 75 W (which is half the specified P_tot) per transistor. So it would take something like 15 maybe 20 transistors.
Some TO247 case FETs might have lower thermal resistance, but due to the small area it gets worse with a mica layer. Also many (especially modern) MOSFETs have SOA limitations and current sharing is more difficult with MOSFETs. So FETs are usually not a good choice - especially if the 2N3773 are already there.

Depending on the circuit/ parts one might not get the initially mentioned 1 kW of heat loss: The transformer might not even deliver that much power (without PFC at least).  So of the power loss word be at the rectifier, fuses and shunt's. The full heat loss would be only at zero output voltage and thus a dead short. Fold back current limiting could reduce the current in this worst case. Even a simple pre-regualtion (e.g. limit charging of the filter caps with MOSFET switching) can reduce the power loss to something like 1/2 or 1/4.

At this power level it might start to become attractive to also include an inductor for raw voltage filtering.  Inductors (like transformers) get more effective when they are larger and also the price is going up slower than linear with power. An inductor will also provide some power factor improvement. Just a filter cap would produce massive current peaks at this power level, as the transformer will be low impedance. So the inductor would also save on the rectifier and transformer rating not only with the filter caps.

[Bendba]

Hi,

Thanks a lot for all the comments.

I get that wanting to dissipate 1kw is a bad idea. I probably should have stated that I'm talking about a variable linear supply, so the 1kw would be the worst case scenario. It would take a load of 40A down to almost 0V output, unlikely to ever happen. I just want to make sure it can take the beating if need be. I'd say the worst that could happen in normal use would be around 20A 5V output, so (24V-5V)*20A = 380W to dissipate.

The transformer I have is actually rated for 24V DC 40A, I know, it's an unusual rating but by the size and weight of the transformer (over 25kg, about 45x25x25 cm) that sounds about right.

I'm still trying to get informations from the manufacturer but apparently they are tied up with the medical equipment factory not to divulge information. I do believe that the rating is for the actual rectified output. That would mean I do not have to account for losses in the transformer and bridge rectifier.

It also appears to me that if I can pull it off as a purely linear regulation, I could build the transistor module in such a way that I could use it as an electronic dummy load.
Or even a high power, low frequency function generator.

There is a lot of very interesting comment and I'm taking note for designing another variable supply more efficient.

I hope you understand that by trying to do it that way, that give me a chance to learn and experiment. It doesn't mean it's a good idea, I get that. That's why my original question what focused on the specific heat question.

From the quick calculations I did, I came up with 27 transistors to take the load but I allowed for higher junction temperature.
I thought maybe I'd try a 100W module first, 3 transistors.

Moral of story: dissipating 1kW is hard

But not impossible, right? I could use it as a room heater in winter.

So, from what I get, I should go with a close loop liquid cooling, no insulation between transistor and heat exchanger, get maximum turbulence in the liquid for a good exchange. I can get some HVAC radiators from the scrap yards, any size, so getting rid of the heat wouldn't be a problem.

I should be able to cast some good exchangers for the transistors, with internal copper pipes (I don't like running water through aluminum when there are other metals in the circuit)

Or I could cast a few big radiators, and maybe use car radiator fans. I think I'll do a test with a water cooled 100W module first.

At this power level it might start to become attractive to also include an inductor for raw voltage filtering.  Inductors (like transformers) get more effective when they are larger and also the price is going up slower than linear with power. An inductor will also provide some power factor improvement. Just a filter cap would produce massive current peaks at this power level, as the transformer will be low impedance. So the inductor would also save on the rectifier and transformer rating not only with the filter caps.

Do you think it would be worthwhile going to that trouble where I already have the big bridge rectifier and capacitors there? (3 x 100,000 microfarads)


PS: once I'm done with my inefficient linear design, I'll reuse all your advices again for a switch mode power supply. I'll probably go for a wider output range.

And again, thanks for your help.

[tszaboo]

The moral of the story is this: You can ditch the transformer, 3-4 AC-DC power supply and make the usual relay switched multiple range power supply, and have a copletely manageable 150-200W dissipation. Or use transistors, which are designed in the second half of the last century. Or make a DC-DC pre-regulator. We just think, that what you described is wasteful and harder than the alternatives.

[rancor]

Hi,

Thanks a lot for all the comments.

I get that wanting to dissipate 1kw is a bad idea. I probably should have stated that I'm talking about a variable linear supply, so the 1kw would be the worst case scenario. It would take a load of 40A down to almost 0V output, unlikely to ever happen. I just want to make sure it can take the beating if need be. I'd say the worst that could happen in normal use would be around 20A 5V output, so (24V-5V)*20A = 380W to dissipate.

The transformer I have is actually rated for 24V DC 40A, I know, it's an unusual rating but by the size and weight of the transformer (over 25kg, about 45x25x25 cm) that sounds about right.

I'm still trying to get informations from the manufacturer but apparently they are tied up with the medical equipment factory not to divulge information. I do believe that the rating is for the actual rectified output. That would mean I do not have to account for losses in the transformer and bridge rectifier.

It also appears to me that if I can pull it off as a purely linear regulation, I could build the transistor module in such a way that I could use it as an electronic dummy load.
Or even a high power, low frequency function generator.

There is a lot of very interesting comment and I'm taking note for designing another variable supply more efficient.

I hope you understand that by trying to do it that way, that give me a chance to learn and experiment. It doesn't mean it's a good idea, I get that. That's why my original question what focused on the specific heat question.

From the quick calculations I did, I came up with 27 transistors to take the load but I allowed for higher junction temperature.
I thought maybe I'd try a 100W module first, 3 transistors.

Moral of story: dissipating 1kW is hard

But not impossible, right? I could use it as a room heater in winter.

So, from what I get, I should go with a close loop liquid cooling, no insulation between transistor and heat exchanger, get maximum turbulence in the liquid for a good exchange. I can get some HVAC radiators from the scrap yards, any size, so getting rid of the heat wouldn't be a problem.

I should be able to cast some good exchangers for the transistors, with internal copper pipes (I don't like running water through aluminum when there are other metals in the circuit)

Or I could cast a few big radiators, and maybe use car radiator fans. I think I'll do a test with a water cooled 100W module first.

At this power level it might start to become attractive to also include an inductor for raw voltage filtering.  Inductors (like transformers) get more effective when they are larger and also the price is going up slower than linear with power. An inductor will also provide some power factor improvement. Just a filter cap would produce massive current peaks at this power level, as the transformer will be low impedance. So the inductor would also save on the rectifier and transformer rating not only with the filter caps.

Do you think it would be worthwhile going to that trouble where I already have the big bridge rectifier and capacitors there? (3 x 100,000 microfarads)


PS: once I'm done with my inefficient linear design, I'll reuse all your advices again for a switch mode power supply. I'll probably go for a wider output range.

And again, thanks for your help.

I don't think dissipating 1KW with a water cooling loop will be all that hard 2 360 or maybe even one 360  computer rads should be all you need with adequate fans. The problem is going to be getting the heat into the water loop.  You might look into using CPU water cooling blocks. They are efficient and you are extremely unlikely to beat the heat transfer capabilities. The downside to using all these computer water cooling parts is that if you chose a computer water cooling pump they won't work will with high water temps. They are not meant for water temps much above 35C.