[jan-jaap]

I'm busy rebuilding power supplies for my old 4D series. The one in my Crimson would kill the system before it had even fully booted. When opening up the internals I found lots of leaky capacitors, plus additional damage so I decided it was time for a comprehensive rebuild.

This PSU is rated for a total output power of 1050W. The PowerOne SPM5 series is a modular power supply. Not like modular PC power supplies where it simply means you can remove unused cables, but it could be configured with a range of output modules for various voltages and currents. The SPM5 chassis can handle at most 1500W. The SGI configuration was rated for:

• 5V @ 170A
  
• -5V @ 5A
  
• 12V @ 16A
  
• -12V @ 7A
  

The reason I'm saying this is that old USENET posts are sometimes confused between the 1050W and 1500W numbers and draw incorrect conclusions about which models are supposed to be capable of running a Reality Engine.

Here are two SPM5 chassis side by side:



The left is from an SGI, and has all output modules removed. The right is an SPM5 configured for 5V @ 150A, 3.3V @ 150A and triple 12V @ 10A. I bought it for spares. You can see why I like modular PSUs: with the output modules removed you can poke around measure. Just beware that almost everything left of the SGI PSU in the photo is not galvanically separated from the mains! I use an isolated oscilloscope probe.

This diagram is from the PowerOne datasheets:



We'll start at the mains input terminals. This is the rectifier board, removed from the PSU:



From left to right: input terminals, EMI filter, rectifier (under heat sink), filter capacitors.



You can see that the film capacitors have cracked with age (they absorb moisture). I replaced them. I also replaced the thermal paste between the rectifier and heatsink. I measured the main filter capacitors. There are two banks of 3x  1500uF and they checked out fine so I left them alone:



Replacements would have costed me around 100 bucks and availability is problematic.

Did I say modular PSUs are convenient to work on? Here's a test of a rebuilt rectifier:



Next up, the backplane board. this is an old photo but it shows where various parts of the functional diagram are located:



In my experience, functional problems with these PSUs have always been with the bias supply. So I decided to figure out how it ticks. This is the relevant part in more detail:



The implementation follows the UC3844 datasheet:



Basically, the BUK456 is Q1, and the PWM frequency is set with Rt/Ct where half of Rt is the blue potentiometer. So that can be used to adjust the PWM frequency.

This is the drain-source voltage drop over the BUK456:



This is with an x200 (isolated) probe, so you're looking at a square wave of some 250V. It is sent to the transformer which has many secondaries. One of them powers the fan, one the control logic, and one each for every module slot.

This circuit is always 'live' when the PSU is connected, even with the system off. At this point I screwed up. When you remove power from the PSU, the voltage drops and at some point the PWM cuts out. You hear the fan spin down. But some 70V remains on the main rectifier filter caps and it takes > 10 minutes to drain. I heard the fan spin own and thought it was safe to move the scope leads. I shorted drain and source of the BUK456, it sparked and it was dead. I was pretty sure the high voltage had killed the UC3844 and probably the BUK456 switcher. There's a handful of other semiconductors, all small stuff. I decided to simply replace the lot. Of course nothing is simple when you're dealing with obsolete parts and global shortages of components, so it took a bit of time to rebuild the circuit.

It didn't work.

That's when I found out the fuse had blown. I should have measured it and I didn't. It looked intact...

But I'm happy to report that after replacing the fuse, all is good again.

I think there's an attachment limit to a single post so I'll stop here. There's an interesting story to how the inhibit circuit works (standby vs operational results in different fan speed, and how the control circuit works). I also have some photos of the carnage of the filter capacitors from the output modules which were beyond disgusting.

[jan-jaap]

Part 2, where we focus on how the SGI PowerOne PSU is controlled using the 12-pin connector on the input module.

The SGI version of the PowerOne SPM5 series is different from a standard model in two ways:

1. The meaning of the inhibit signal on the 12-pin connector. The standard SPM5 series enable outputs by default when input power is applied (they are 'ON'). The SGI model doesn't. If you connect the inhibit signal to ground, the SGI PSU enables outputs. You may have noticed a blue jumper wire on the 12-pin connector in some of my photos: that's how I switch on the PSU outside a system. This behavior is configured with positions 4 and 6 of SW2 in the control circuit and documented in PowerOne "High Power Modular Series Applications Note #H7 For Product Series: HPF3, HPF5, & SPF3".
   
2. There's an extra circuit board installed, between the 12-pin connector and the PSU control circuit. It's a simple circuit, I traced it and it seems to release the inhibit signal if the overtemp signal goes active. In other words: it switches off the system if the PSU overheats, rather than just signaling this condition to an external system controller. 
   

This is the PSU control circuit:



I didn't fully reverse engineer it, but you can recognize major components:

• It's powered from a secondary winding of the auxiliary bias supply I discussed before. So it's galvanically isolated from the mains. It has to: it connects to the 12-pin connector on the input module.
  
• You can spot an 7805 regulator.
  
• You can recognize some voltage comparators and an Airpax overtemp switch.
  
• The SW2 controls behavior including the inhibit function, but is largely undocumented.
  
• The (white) MOC8104 optocoupler allows the control circuit to control the auxiliary bias inverter.
  

I was interested in the function of that optocoupler, so I did some measurements. Remember: the aux bias supply is always on, even if the PSU outputs are inhibited. In this state, the voltage drop over the LED side of the optocoupler is 1.17V (DC). The voltage drop over the C-E is 0.2V. In other words: LED is on, phototransistor in conducting. If I switch on the power supply, the voltage drop over the LED rises to 2.67V, and the drop over the phototransistor to 13.7V. This voltage drop over the LED is only possible if no current is flowing through it. The phototransistor is no longer conducting.

What this does to the aux bias inverter (the circuit built around the UC3844N and the BUK456) is kind of interesting.

This is what the PWM output of the UC3844 looks like when the PSU outputs are off (inhibited):



And with the PSU switched on, it looks like:



Somehow, when the PSU is in standby, every 3rd pulse from the UC3844 is missing, and the amplitude is much less as well. The voltage generated (and used) by the aux bias supply goes up when the PSU is switched 'ON'!. The UC3844 is not like a digital circuit that requires a fixed supply voltage. When the PSU is switched 'ON', the voltage goes up which also makes the PSU fan spin faster. I knew this, I just never realized how this works...

The application note hints at this as well:

2. Inhibit Signal
What is the purpose of this signal?
By activating this signal, the end user can inhibit all output modules simultaneously. This signal interfaces with the pulsewidth modulation circuitry of all modules and turns the pulsewidth modulation IC either ON or OFF.

The pulsewidth modulation IC discussed here is probably on the output modules, because obviously the aux bias supply inverter is always 'ON'.

[Geoman]

I love these restauration and repair posts so much! 

An inspiration to work on my spare PSU as well

I also have some photos of the carnage of the filter capacitors from the output modules which were beyond disgusting.

Rifa! The magic smoke has escaped! (SCNR, 07:00)

[jan-jaap]

I also have some photos of the carnage of the filter capacitors from the output modules which were beyond disgusting.

Rifa! The magic smoke has escaped! (SCNR, 07:00)

I was more talking about the electrolytic capacitors... the thing is, I never expected it to be this bad. After all, the power supply worked. The functional problem was with the aux bias supply on the mainboard, not with the output modules.

But then I removed the capacitors from the -5/-12/+12V module and found this:





The PCB looked like this (C12 site has been cleaned already). Fortunately it cleaned up fine and in the end the PCB wasn't damaged (yet?). I thought of dumping the whole PCB in pure alcohol in the ultrasonic cleaner, but not all components like ultrasonic cleaning so I just cleaned it by hand with lots of alcohol.





This is what the little PCB on the bus bars of the 5V module looked like, before and after cleaning. Minor damage to the silkscreen, but copper layer and PCB intact. This is what came out of the 5V module:



These capacitors were completely shot. Still, the PCB cleaned up nicely. It has to be 100% clean, or new solder won't flow properly:



To the left you can spot four mosfets on individual heat sinks. They likely form a H-bridge inverter circuit. The hole is where the transformer goes. Below you can see what it looks like, with it's four diodes making the rectifier.



Transformer and diodes are a pretty compact unit, if you know that 170A can flow through it.

Right now, the Crimson's PSU is probably fine again (I ran it for a couple of hours with a 40A load and it didn't blow up). But the internals of the PSUs of my 4D/380VGX rack and my 4D/440 are likely a similar toxic waste pile so I don't dare running those anymore until I've rebuilt the PSUs. I will also have to check up on the 4D series circuit boards, see if anything is leaking there...

NB: I used mostly Panasonic FR series capacitors for the rebuild. These are long life 105c capacitors with very low ESR and able to handle high ripple currents. Ideal for SMPS applications.

[robespierre]

I thought of dumping the whole PCB in pure alcohol in the ultrasonic cleaner

You shouldn't do this because it produces an explosive vapor. Cleaners must be nonflammable to be sonicated.

[vishnu]

Yowzers! 🤯

What do you use for a solder-sucker, if I may be so bold as to inquire? 😎

[jan-jaap]

Yowzers! 🤯

What do you use for a solder-sucker, if I may be so bold as to inquire? 😎

I did the first one by hand with my old soldering station and litze. It was torture so I bought this thing: https://www.eleshop.nl/desolderingstation-zd-915.html  There's a newer version (https://www.eleshop.nl/zd-8915-desoldeerstation.html) but apparently that one is more likely to clog up so I got the older model. So far it works fine. For 100 bucks I can't complain, really.

I thought of dumping the whole PCB in pure alcohol in the ultrasonic cleaner

You shouldn't do this because it produces an explosive vapor. Cleaners must be nonflammable to be sonicated.

Thanks. Guess I'm stuck with  8L of pure alcohol. I'll use it in the burner when we make a cheese fondue

[jan-jaap]

Finally, a question for the electronics experts: in these power supplies there are places where a sort of 'kit' (glue? sealant?) is used where wires meet a PCB. I guess it's to give some mechanical support but considering the voltages it better be non-conductive. To the left you see what's left of it in a ~ 1990 vintage PowerOne SPM5 from an SGI. To the right what you find in a generic SPM5 unit from around 2001. The material from the older unit looks quite tired. It crumbles when put in the ultrasonic cleaner (with soapy water ;-) ).



Any idea what I should use to replace it? It should remain somewhat flexible/rubbery. Ideally something that I can order in smaller quantities because I imagine it doesn't last once you open it.

[robespierre]

That material is commonly called Silastic, a trademark of Dow Corning. I believe you can purchase a version of it currently under the "DOWSIL" brand. 738, 739, 748, and 3145 RTV look like they would work, from lowest to highest viscosity respectively.
An alternative would be Henkel Loctite 5510 (clear), a modified polysiloxane elastomer with low conductivity and low dielectric. It's firmer than DOWSIL, but thinner uncured for easier application.

[weblacky]

I think I use 738, as it's one of the few that SPECIFICALLY mentions being used for stabilizing capacitors and coils and has good dielectric attributes. Also cures rubbery, not solid, melts easily too for removal. I really like it, though hard to get cheap..but you can on eBay.

However I don't care what you use...I still NEVER cover IC legs or cross component legs or metal areas with it. Aging causes odd things and I still feel it's UNSAFE to link component areas with it that (if it were metal instead of silicone) would short. So I try to use a special dispenser that let's be get close and in between components to better position the stabilizer. So for example, if you're going glue like 2-3 cap together, you place staliber along their trademarked, label, areas but stop before you touch the a top metal can or a capacitor leg. That way even if it turns conductive, you don't care as you've attached it to non-conductive component areas.