Repairing a Agilent Infiniium 54825N Power Supply (NX300-6003)

[pandrew]

Hey All,

I'm trying to repair an NX300-6003 power supply.
The fuse is blown, the diode bridge is shorted, and so is one of the PFC mosfets (Q102) (all three pins, including the gate of this one mosfet were shorted)
The gate of each mosfet is connected to a dedicated network of 2 resistors and one diode. And the two resistors associated with the shorted mosfet also failed open.
The diode appears to be okay though. (See attached dead_resistors.jpg )
Both mosfets are eventually driven by an ML4812CQ chip that's on a daughterboard, and the output pin of this chip measures as ~1 ohm to GND, so I suppose the controller chip is also gone.
This cannot be bought new anymore but I can get one on ebay/aliexpress, probably salvaged from another product.

Replacing the controller chip is gonna be a pain, not because of the PLCC package, but because it's sitting on a vertically mounted super long daughterboard soldered in via 90 degree 2mm pitch pin headers.
I suppose my best bet might be to remove the two mosfets and heatsink on the edge of the board, and to remove the two capacitors next to the chip, and maybe that will give me enough access to solder the new PLCC chip on.
For pins 4..8 I will have to rely on hot air only to solder it, with the rest I'll probably have enough access to drag-solder it. (See attached how_I_plan_to_get_access_to_the_controller.jpg)

The diode bridge looks like it's supposed to be heatsinked by the separator sheet metal that is between the input filtering and the rest of the power supply. But there was no contact, and whatever thermal compound was on it, was completely cooked. I suppose I will have to make sure this is properly heatsinked. (See attached diode_bridge.jpg ). I bent up the sheet metal to remove the bridge, but before that it was sitting at 1-2mm distance from the surface of the diode bridge. I'm thinking of using like 2-3 layers of thermal pad ontop of each other, then bend the sheet metal back over the new bridge. (I could also drill a hole into the sheet metal, and put a screw through the bridge rectifier and sheet metal to keep them together, and in that case only use a single layer of thermal pad) Alternatively I was also thinking about leaving the sheet metal bent up, and instead putting a small "glued" heatsink on it, but i'd be afraid of it falling off in use  later on, especially since if I remember right this normally sits upside down. Not sure If there are any pre-made heatsinks I can buy for that form factor. Any recommendation on what to replace the bridge with? I was thinking "NTE53002", since it has a built in heatsink metal, so It'd have a better interface to the sheet metal, and otherwise it has similar specs.

I would appreaciate any advice,

I also read through this related forum post which I found interesting: https://www.eevblog.com/forum/repair/agilent-54825a-infiniium-oscilloscope-dead-power-supply/

Cheers!

[pandrew]

A couple updates:

I have replaced the dead resistors, and dead IRFP450 (replaced both of the pair, just to be sure).
I have removed the the other mosfets in order to gain better access to the control daughterboard, and
I successfully removed the two capacitors, and the dead ML4812CQ. (See this_is_how_much_space_i_have_now.jpg . Hopefully I'll be able to solder in at least the bottom pins of the ML4812CQ with just hot air, and then drag-solder the side and top pins without damaging anything around.)
I accidentally touched my soldering iron to the corner of the big high voltage capacitor, and melted it a little (See melted_corner_of_cap.jpg) although I don't think it's a problem, there must be metal under that heatshrink-like-looking outer packaging. Although I'm conscious that capacitor better not be damaged since it gets very high voltage it being on the output of the PFC, that's why I'm mentioning it.
The dead ML4812CQ OUT pin still measures shorted to PWR GND even out of circuit (~1 ohm).
I measured around the daughterboard and I didn't see anything out of the ordinary. Resistors measure as <= their value in circuit. Diodes are okay, except for one which I couldn't measure cause it seems to be in parallel with a 47 ohm resistor, and I didn't want to start removing more components just to measure them out of circuit.
I have also replaced the diode bridge, and drilled a hole in the sheet metal, and attached it with a screw to the diode bridge, with a thermal pad inbetween. I don't _love_ what it looks like but at least now there's some proper heatsinking. (see diode_bridge.jpg)

I don't have the ML4812CQ yet, and it looks like the wait is gonna be longer than expected, the Aliexpress seller hasn't sent it yet. Anyone know another place where I could get it, other than ebay/aliexpress?

[pandrew]

I also made this 4-layer image that can be useful for reverse engineering how the PFC subcircuit works (posted here as 4 individual images, but one can put them input something like GIMP, on separate layers, and toggle the layers as they wish for example)

The images are perspective corrected, the bottom images are flipped, and they are aligned over each other.

Obviously the ML4812CQ is missing, and so are the two caps: 1uF/50V , 56uF/35V

[pandrew]

Success!

I didn't do any drag soldering.
I added solder to the PLCC pads, as much as they would reasonably hold, I also wetted the bottom of the PLCC pins with the same solder, to make sure it will solder easily. (not much solder was retained by the pins).
Then I removed all old flux, added fresh tacky flux, and used only hot air to solder the chip. Then I double-checked all pins with a multimeter, so that there are no adjacent pins shorting, and all connections are made. The ML4812CQ I originally had in the power supply was Fairchild branded, but the one that I replaced it with had the older Micro Linear branding.

The first two power-ups of the power supply were successful.
The third power-up blew the 8A fuse (and a 16A B curve MCB in the electrical panel), and one of the IRFP450 PFC switchers went shorted again. The PFC controller survived, it was only the one mosfet that died.
The only significant thing I did differently on the third power up, is that I waited less than 3 seconds between unplugging and replugging the power supply.
I think this power supply might be sensitive to unplugging/replugging immediately. (either that or it died again by random chance)
I replaced the dead IRFP450 with the old working one I removed from the original pair. (I didn't have any new ones at hand, I didn't order any extra).
Then the power supply started working again, and it's been working fine ever since.

I have measured that the voltage on the big PFC bulk capacitor goes from about 380V to about 30V, in 16 seconds. (From 30 volts down the rest of the discharge takes a lot longer)

I have now tested many times unplugging the power supply and waiting 16 seconds, and plugging it back again, and there have been no failures.
I'm sorry but I won't test purposefully killing it with shorter than 16 second unplugs, so we probably won't know if my theory is correct.

The oscilloscope works perfectly (after re-seating the RAM. It wouldn't boot without a re-seat)

The only disappointment I have is I tried a 128GB mSATA with an 44-pin PATA to mSATA adapter and it didn't work (this one: https://www.emag.ro/convertor-ide-44-pini-la-msata-frame-2-5-pana-la-7-mm-delock-62495/pd/DDZJ5ZBBM/ )
I dd'd over the original harddisk image without modification, including the MBR.
Interestingly it recognized the size of the disk in the BIOS, but I got a "Missing Operating System" message.