[toncho11]

Hi,

I am not an electrical engineer, but I have managed to recap a few power supplies in the past.
I am thinking of recapping two SGI Indy power supplies - one that works and one that does not. Both are Nidec.

1) How do I open it? I do not see where the screws are?
2) Are the capacitors the main source of problem for the Nidec? What are the ICs that a common point of failure for the Nidec PSU?

[weblacky]

There is no common failure on Indys PSUs, thats why they lasted nearly 20 years, for many owners. Very robust.

They have a large amount of caps and the board is 1980's style PCB, delicate tracks on a thin, single-sided, PCB.

Indy Nidec doesn't have much "smarts" so you'll need to bench test for all rail voltages, it will run with some voltage rails not working (I've seen this). Most ICs are screwed to a common heatsink wall, you'll destroy the original fasteners getting them off, but you can use modern hardware to reattach the ICs to the wall, easily.

If you don't own a thermal imager, don't bother troubleshooting an Indy PSU with semiconductor damage, the layout is confusing with several regulators and high-speed diodes technically sitting on the HV side of the board...but connected to LV side instead.

Also, if you have a nonworking power supply, it's not the capacitors, anymore. It's the capacitors and (at least one) semiconductor(s). You're better off recapping a working power supply, then you're golden. Once the caps get so bad that the power supply has trouble the semiconductors will quickly be damaged. Their job was to preserve the semiconductors through filtering. Once that filtering is gone, the semiconductors will start exploding. Once that happens, you have semiconductor damage and no amount of capacitor replacement is going to fix that. So preventative maintenance is much easier than investigating a finally broken power supply.

If the Indy Nidec power supplies had more of a brain, they would shut down earlier, when the regulation was having trouble. They don't have enough of a brain to do that until they're critically wounded. And once they're critically wounded, there's more trouble than just filtering.

They rely primarily on their large, high-voltage side, filter capacitors to do the vast majority of filtering. So these capacitors are actually very important to be replaced. These power supplies pre-date power factor correction, and so the only way that they can function well as having all their filtering elements changed. So it gets expensive and laborious.

Most people have trouble removing them and many don't wanna spend the extra money so they leave the original filters in place because bad advice online says that these never go bad. That's a total crock, they absolutely do go, and when they do they just don't filter much anymore. Do they explode? No, do they leak, I've never seen it. But that doesn't mean they're doing their job anymore and their job is the largest, and first, line of defense on filtering, and also preventing excessive back EMF into your electrical system. So expect to replace every single electrolytic cap on the board, including the large ones. Also, there's a couple semiconductors that I've seen damaged on the large daughter board that functions as the brain. But there's only a few power transistors on that board. The vast majority are where you would expect them.

Don't attempt this job if you don't have a desoldering gun. You'll rip the tracks and vias right off the board if you attempt to use desoldering wick on the majority of these.

The design also has about 2 to 3 power absorbing resistors that are used to drain the power supply after it shuts down. These are unfortunately, in line with the power system the entire time and have a tendency to really cook and often get out of spec (you'll see the burning on the stabilizer adhesive). It's a standard part that I change in all my rebuilds now and I've change the resistor formulation and upgraded the wattage from the original design. We have new materials that didn't exist when these were being designed and produced. New breeds of resistor material are not only fire resistant, but also have higher wattage resistance in the same physical size as the old ones, so you can easily get one watt higher in a lot of the smaller bodies that were available in the older wire wound or thin metal designs.

[robespierre]

There's a difference between a bleeder resistor (used to provide a path for stored energy to bleed away after power is shut off) and a load resistor. The former is for safety when the unit is serviced.
The latter is for stability when the power supply may be used without a load on every power rail: it provides a minimum load even if none is connected externally.

Switching power supplies for low voltages don't normally need bleed resistors, because the primary stored energy is drained by the main chopper transistor. Some regulator feedback designs may require a bleed resistor on the primary side reservoir. It goes without saying that the energy stored on the secondary low-voltage side is not dangerous and does not require bleeders.

Better regulator designs maintain stability over a wider range of loads, reducing the requirement for load resistors to be used. When the powered device is known to sink power from all of the rails, the load resistors can simply be removed. That is not always the case; for instance, an Indy or Challenge S with no hard drives would not draw much power from +12V, and yet this rail must be stable to avoid damaging the fan. So there is (at least one) +12V load resistor.

The power dissipated in a load is the same no matter what kind of material is used to construct it. P = V^2/R. The reason that components dissipating higher power must be physically larger is that power is lost as heat, and the heat flux into the environment is limited by thermal resistance. With air cooling, the thermal conductivity of air in W/mK is terrible (about 0.025, similar to polyethylene) and limits the heat flux for the whole system, and so higher power requires more surface area.

[weblacky]

There's a difference between a bleeder resistor (used to provide a path for stored energy to bleed away after power is shut off) and a load resistor. The former is for safety when the unit is serviced.
The latter is for stability when the power supply may be used without a load on every power rail: it provides a minimum load even if none is connected externally.

Switching power supplies for low voltages don't normally need bleed resistors, because the primary stored energy is drained by the main chopper transistor. Some regulator feedback designs may require a bleed resistor on the primary side reservoir. It goes without saying that the energy stored on the secondary low-voltage side is not dangerous and does not require bleeders.

Better regulator designs maintain stability over a wider range of loads, reducing the requirement for load resistors to be used. When the powered device is known to sink power from all of the rails, the load resistors can simply be removed. That is not always the case; for instance, an Indy or Challenge S with no hard drives would not draw much power from +12V, and yet this rail must be stable to avoid damaging the fan. So there is (at least one) +12V load resistor.

The power dissipated in a load is the same no matter what kind of material is used to construct it. P = V^2/R. The reason that components dissipating higher power must be physically larger is that power is lost as heat, and the heat flux into the environment is limited by thermal resistance. With air cooling, the thermal conductivity of air in W/mK is terrible (about 0.025, similar to polyethylene) and limits the heat flux for the whole system, and so higher power requires more surface area.

That's interesting...the only thing I know for certain about the Nidec design (because I used thermal imager on it) is that the resistor in question is hot all the time the PSU is running and it remains hot for nearly 15-20 seconds AFTER shutdown (lingers)...then fades quickly.  So I made assumptions based on observed behavior.

[toncho11]

Thank you all for the information!

And what about using a (more) modern power supply? 

These are the requirements for the Indy:

+3.3V 7A
+5V   25A
+5V   standby 0.02A
+12V 4.5A
-12V  0.75A

No need for -5 volts. Most old power supplies have -12 volts. We need PSU that provides higher or equal current for these voltages. Although higher probably means bigger risks if something goes wrong.
Like this we also loose the speaker sound. 
At least pins 3 and 4 of the smaller 20 pin connector will need to be adapted for the new power supply to start.
Also the new PSU should somehow fit in the case of the old one. Except if it is not made as an external power supply.

Sounds like a lot of work, but ultimately doable (in theory)?

[Raion]

You can't use a more modern power supply. There is logic in the current power supply that cannot be replicated in an ATX power supply. Don't do it. Someone was working on a PSU replacement plan but ultimately gave up.

I'm working first on an O2 replacement and probably Indigo 1, 2 and maybe Octane first but then I'll consider coming back around, a sort of gentleman's agreement with Weblacky who wants to do that.

The -5V in the Indy is for the amp, if you skip it, it may have unintended consequences.

[weblacky]

Regardless, in the end, it'll cost you a heck of a lot more than the current rebuild offerings to build any kind of modified unit. Even if you manage to find an ultra small form factor power supply that would fit in the area, even if you created a circuit to emulate the logic, and even if you managed to get the thermals/cooling replaced that the power supply provides to the entire system (alone). You would be well over the price of a rebuilt, including shipping.

If you already had professional knowledge of CAD design as well is a good 3D printer then you could probably do it for just about the same cost if not a little cheaper if you just happen to manage to find everything and knew that much about logic electronics. I certainly don't have those things so it's much easier to just rebuild the originals as they lasted for 20 some years so there's nothing wrong with them, fundamentally.

But if you already own a professional soldering iron, and a desoldering gun, and a steady enough hand, you could always rebuild the power supply yourself as you suggested. After researching all the caps and finding their upgraded equivalence you can place an order and try to refurbish your working one and see how that goes.

Most people don't own the tools needed to do the rebuild. So again, the pricing works out. Similar to not having a huge car lift to do the minor maintenance on your car. You pay someone who has the lift to do it. Otherwise, you buy the lift yourself, right?

[toncho11]

For the working one I am thinking of replacing the caps. I have access to a professional disordering machine.
The dead one will have to wait. Probably I will ask for a professional to do it.

Please tell me at least how do you open it? I do not want to damage it.

[weblacky]

For the working one I am thinking of replacing the caps. I have access to a professional disordering machine.
The dead one will have to wait. Probably I will ask for a professional to do it.

Please tell me at least how do you open it? I do not want to damage it.

I'll give you enough of a hint that you should be able to figure it out.

There are no screws involved in the final assembling of the outer case of an SGI Indy power supply. Look at the sides along the top, longest, edge. You see clips there. Examine around the entire top and I'm sure it will become apparent how it must come a part.  It cannot be initially disassembled with your bare hands, you will need prying tools like a hard putty knife or chisel to initially separate the top casing.  Only one out of the four sides (on the top) is where you separate. You can fully reassemble the PSU casing with your bare hands. The clips do not bend at all, do not attempt to bend them or open the clips on the sides.

There is a small circuit board on the face with buttons and a speaker. These buttons do absolutely wear out, I tried two power supplies where the buttons intermittently worked or were dead. If your power button is dead swap it with your reset button for the time being to at least get back in fighting shape. But make sure to test the buttons on that board before you reassemble the unit. Also that small board cannot be easily put in once you put the rest of the guts of the power supply together. So the small daughterboard that carries the speaker interface and the power, reset, and volume buttons must be the first one in the empty case if you want to make it easy on yourself after removal. Be aware to that for best effort you'll need to separate the long daughter card from the rest of the board. It should be tacked together with stabilizer adhesive. You'll need to melt through the adhesive on a low temperature soldering iron, carefully to remove enough stabilizer to separate the card. You also have three points of stabilizer inside clusters of parts that you'll need to do the same thing with. You don't have to put back stabilizer if you don't intend to ship the power supply anywhere. I do however reapply a 3M silicone stabilizer for this purpose to reaffix the board and some of the part clusters that are originally tack down. You can technically desolder some of the components while some of them are clumped together but most of the clumps involve large inductors (coils) on the board. So you'll have to separate clusters from neighboring inductor in order to remove them from the board. That will be the second task after you actually remove the guts of the power supply. 

Lastly I'll give you one other piece of advice, because even if you do it right you have about a 20% chance of damaging this. You'll need to remove the huge filter capacitors that use snap terminals on the high-voltage side to replace them. There's so big that once you melt the solder in them you can sort of walk them out while the solder is molten. But even so you'll hear a creaking and screeching noise as you try to get them out and even then about 20% of the time you'll pull out one of the villas with them because the boards aren't particularly strong. If you pull a via out don't immediately panic. Carefully retrieve the via by desoldering it from the cap snap terminal using your iron and a set of tweezers, clean it off with soldering wick, and put it back into the empty hole on the board, and afix it using something like soldering mask or conformal coating. And then check that the Villa is in contact with the track that it's meant to touch. Since the track is on the bottom of the board but the via inserted through the top normally a simple soldering bridge will fix this for you.

Several of these vias are very tight compared to what you'd expect. So desoldering with wick will not clear them for you. That's why I said even for some of the smaller caps you want to clear them with a desoldering gun because some of the holes are just barely the diameter of the caps that they support.

There is no stabilizer on the bottom of any of these caps. And no adhesive on them either that held them to the board during soldering.