can i convert a Cray 3.3v/400A psu into a variable voltage psu?

You will need several FETs in parallel, any old chip will give you the pwm even the humble 555 timer.
Your problem will be the large currents involved (physical termination), making sure you turn the FETs on and off quickly so that the FETs don't dissipate too much, watching out for reverse voltages and suchlike when switching such big currents.

I would suggest experimenting with say a 5amp load for a start and then work up slowly.

If the steel tube is getting hot so will all your terminals etc...Interesting project though...let us know how it goes.. it could be useful for when we all build our electric cars!
Del

Thanks for the reply. It's possible to get a 150 amp mosfet for about $30 so two or three of these in parallell is not a problem. Is it acceptable to just common the trigger and would the mosfets self current limit so that one does not overload whilst one under performs? The mosfet resistance concerms me most. Heavy conductor (solid copper sheet / plate) is easy to reduce heating effects but a typical resistance of the heater tube is for instance 22mohm and this is perhaps not too different to the resitance of the mosfet - so i@m worried that halfthe energy will be used in heating the mosfet - I'll need a good heatsink and fan!

Is it acceptable to just common the trigger and would the mosfets self current limit so that one does not overload whilst one under performs?

Yup that's fine.

Personally I'd have an individual resistor to the gate of each FET from the drive circuit, say 10K and a pull down on each gate say 100k just to esure they switch off if the drive circuit goes open or loses power for some reason.

There are some real good FETs with nice low R_DS ON

Have fun

Del

Thanks for the advice - the post following yours has pointed out some useful looking mosfets which might do me the job nicely.

There is some interesting stuff about parallel FETs in this thread:-

http://cr4.globalspec.com/thread/13539/Mosfet-Parallel

Here are some 160A FETs for well under $30 a piece (more like $5 -$10), some with Rds down to 3 or 4mΩ.

Back in the good old days of main frame computers the big IBM, Amdahl, and Hitachi machines used 415hz current and I believe that it may have even been three phase but I can't be sure of that. I don't know what the input requirements were for the Cray's. How were the power supplies cooled, the logic boards were cooled with liquid freon on some models. Some of the other vendors used combinations of forced cold air and refrigirated water. I don't want to be a party pooper but you may have acquired a monster.

An interesting note about the Cray machines. The chips produced both a true and an inverted output at the same time. I was told that this was to provide a more stable load for the power supplies. If in some operation a large number of gates were turned on or of at the same time there was a possibility that the sudden change in demand on the power supply could result in some kind of stress and the regulators wouldn't be able to respond quickly enought to meet the change in power demand. Designing it this way meant that there would a constant and equal number of outputs in the on and off states and therefore the power draw wouldn't change.

It would certainly simplify designing the logic circuits because the designer would always have both positive and negative logic signals available negating the need for inverters and the delay they could introduce.

Oh how I miss computers that weighed tons, cost millions, and needed their own special huge rooms. You could keep your lunch cool by lifting a floor tile and placing it directly on the floor with all of the connecting cables as the raised floors were presurised with cold air that was vented through holes cut in the tiles under the actual equipment.

I am getting caught in a nostalgia loop, I shouldn't have the 60's rock and roll playing the background.

Presumably each unused output had to be connected to a dummy load (equal to the load that the used output was driving)?

If the major part of the load was taken by the chip during switching, I don't see how the complementary output would help. Local supply line decoupling capacitors certainly would (by averaging out PSU demand).

As first sight, it looks to me like having these complementary outputs would increase the overall load on the PSU quite a bit, increase the board heating (and hence the demand on the cooling system), and increase the complexity of the board layout.

I'd have thought designing a PSU with decent load regulation (or even just bumping up the PSU capacity) would have been a better bet.

But maybe I'm just waffling ?

Cheers my friend. Well worth getting a couple to try out the idea.

Dig out the high current transformer in your cray supply. Hopefully it should have a 120V primary and a secondary of a few volts at 300 A or more. Connect your tube to the secondary and connect the primary to 120v through a Solid State Relay. Control the relay with a Temperature Controller that has an output that will drive the Solid State Relay. Connect a thermocouple to the controller and position it to measure the temperature of your tube. There, you're done. If you use a Zero Crossing Solid State Relay you will even have extremely low EMI.

This should take a few hundred dollars of parts and you can get everything designed and ordered in an afternoon. Even if you need a custom built transformer it should only be a few hundred dollars more.

Of course, if you're milking the job for hours and need to spend more to justify your time, go play with MOSFETS. But if you just need to get the job done do it my way.

Don Shepherd

Interesting suggestion. My original plan was to have a low voltage transformer wound for the purpose and do just as you propose. Just connect transformer to mains supply and not a switch mode psu in sight. I'm not sure if you are suggesting going inside the psu to locate the transformer. If you are then when the ssr turns off the output voltage from the psu will fall to zero and when the ssr is on it will obviously go to the original 5v. Do you think this cycling, which would need to be quite fast in order to prevent overheating of my item during the on period (the tube is very small and does not have much thermal mass), would cause the psu voltage regulation to go a bit loopy because when the ssr goes off the psu will think there's an error 'cause the output has unexplainedly drop to zero?

I meant to remove the power transformer from the psu. Just connect its secondary to your tube. You don't need regulated DC to heat it, AC will do as well.

The voltage from the secondary will likely be about 8 VAC so the controller will probably operate at a low duty cycle. If your tube has a low thermal mass the temperature variation may be too high. You may want to put a step down transformer between the solid state relay and the power transformer so that the power transformer's secondary is just a little above three volts when the solid state relay is fully on.

Don Shepherd

Ok, I understand what you mean. I'm no expert with transformers but I thought that the transformer in a switch mode supply tends to be physically small because they operate at high switching frequencies (10s / 100s of khz). If I do what you suggest I'd be powering it from 50 hz and so might the transformer be very incorrectly sized (even though it might have a correct turns ratio)? Sorry if this is nonsence, it's just my basic understanding (I'm a mech eng with a small working knowledge of electronics - enough to get me into trouble, sometimes not enough to get me out of it!)

Your concern is justified. As a switching supply the power transformer is probably designed for a high frequency and 60 Hz will kill it. I was assuming that the unit had a conventional unregulated low voltage DC supply and used switching transformers to PWM the low voltage DC.

You could turn on the supply and check the waveform going into it. If it's 60 Hz it would work in the circuit I described. If not, then forget about using it the way I said. You would need a new transformer for that. A 4V 300A transformer would be 1200 VA in size and probably available for a few hundred dollars.

Don Shepherd

Well it's definitely a switch mode supply and a rather flat psu at that (more the shape of an A4 sized scanner so there's definitely no "conventional" 50 Hz 1200va transformer inside it. Thanks for the advice anyway, valuable as confirmation of another approach to use.

If you could get a diagram of your power supply, there has to be some control within the supply that checks the output voltage and adjusts itself to correct any errors, you need to break in at that point and take over control of that point....said quickly it sounds simple!!

Almost everything that you need is already in the box so to say!!!

You will need to have very heavy cables to the tube so that they themselves are not heated up in any way by the current.

I also feel that your tube should not be of a uniform thickness either, yes where you want to control the heating effect, but say 3 times thicker (or more) at the ends where you connect your cables to so that the heating effects there are dramatically reduced, otherwise you will be heating up the cables etc by conduction.....it may be a good idea to supply cooling air (or similar) at the ends where heating is not needed.....

Sort of tried that idea already. I took a 5v 150A psu and fed a false voltage into the remotesense input (the sense input saw a voltage made up of the output from the psu + a variable voltage in the range 0-5v). If I wanted o.5v out I added 4.5v, if I wanted 1 volt I added 4 volts. It sort of worked but stranegely the fan slowed down as my output voltage dropped - strange because the fan was a 110v fan and so I thought would be independent of the output. Going inside the psu does not appeal because they are so densley constructed it becomes quite a mechanical challenge. To do it your way might work but it would require a big stripdown of the hardware. Cheers.

I think you are making a lot of work for your self with PWM control etc.

Why dont you try a variable shunt controlled welding transformer a 400 amp unit will give you plenty of scope. These are around very cheap nowadays as MIG welding has replaced a lot of them, you should be able to arrive at 300amp and whatever voltage that is required to make the short circuit work

Back in the 80's I worked for CEAG, who built some of the power supplies Cray used. If I remember correctly, they were all constant current devices, instead of constant voltage devices. The voltage just folds back when the amp limit is hit, 150 amps in your case, it looks like. Don't fully understand what you are trying to acomplish, but beleive starting from scratch may be move benificial if you need to vary the voltage. Hope this little tidbit helps, almost gave myself a headache thinking that far in the past.

Then how would the voltage be maintained exactly? the chips used are usually devices needing a specific voltage give or take almost nothing!!

All the PSs I every worked on needed a minimum current flow to regulate properly or they had internally a resistor to allow such a current to flow even when unconnected to anything!! But this would probably only be say 1 amp on a 100 amp supply.....or so!!

I used to work on IBM and other mainframes years ago.......voltages were very very specific and needed to be very accurately maintained or the machine powered down!!! Self checking was built in to everything!!