H7821 power supply in MicroVAX 3100, SCSI disk enclosures and others

[Peter Coghlan]

Speaking of problems with DEC power supplies...

Some time ago I posted about problems with five 1800uF capacitors in multiple
H7821 power supplies leaking electrolyte, particularly when the machines are
stored in the wrong orientation.

One of my MicroVAX 3100 machines has a slightly different looking H7821 power
supply with a shiney case instead of the matt cases on all the other ones.
The five capacitors in this one seemed to be in good shape although on closer
inspection there may be slight traces of leakage on some of them.

However, after running happily for five to ten minutes, it went pop with the
4A glass mains input fuse blackened.  Access is difficult enough but not
impossible.  This PSU consists of a double sided main PCB where the high power
components are and four daughter boards permanently attached at right angles
to the main board and blocking easy access to many of the heatsink mounted
components.  There are no component references and it is difficult to trace
the circuit because the tracks on the component side of the main PCB are
hidden by the larger components, heatsinks etc.  There was dirty black dust
accumulated by the fans everywhere and it was difficult to clean off because
of all the obstructions.  Starting at the mains input, I tried looking for
shorted capacitors, bridge rectifier, chopper transistor, diodes etc that
might cause the fuse to blow but failed to find anything obviously wrong.
I also looked at the output rectifiers and smoothing capacitors and found
nothing obviously wrong there either.

I decided to replace the fuse and try the old series light bulb trick.  My
150W bulb glowed at about a fifth of it's full brilliance and the power
supply seemed to function reasonably well, the green LED came on, the fans
turned and the unloaded +5V and +12V outputs looked good.  However the
brightness of the bulb suggested too much current was being drawn.  An
el-cheapo digital multimeter suggested there was only 28V across the power
supply mains input.  A slightly better model suggested it was closer to 90V
(out of 220V), however the likely wacky waveform could be confusing the meter.

I left it running in this state for a little while but pulled out the mains
lead when I thought I got a slight whiff of cooking electronics.  It was quite
difficult to quickly check that the input smoothing capacitors had discharged
enough and probe various likely components with fingers to see which might be
getting hot before they had time to cool down again.  After a couple of
attempts, I identified a MOV/varistor/surge arrestor thing in some thick
heatshrink rubber stuff as a likely suspect.  After unsoldering it, I found
it was cracked and blackened on the underside where I couldn't see because
of the heatshrink covering it.  I replaced it with the equivalent one out
of another H7821 which had been damaged by capacitor leakage.

When I turned the power on, the bulb glowed much less, however, there were
unexplained slight variations in it's brightness.  While I was studying this,
there was a slight cracking sound accompanied by a wisp of smoke and the
bulb suddenly got much brighter.  The replacement MOV thing had failed.  It
was connected across one of the main input smoothing capacitors, the other
capacitor also having one which was unaffected.

This fault was starting to look a bit familiar to me.  In a different DEC
power supply, I had a triac used to switch the power supply automatically
between 220V and 110V mains input go short circuit causing a similarly placed
MOV and fuse to fail when the unit was powered from 220V.  I tried looking for
a similar functioning triac in the H7821 which was not easy to do as explained.
The best way to identify this component seems to be to look for a three legged
high power component (which may or may not be on a heatsink - the previous one
wasn't) which is more or less directly connected between the (neutral?) AC
input to the bridge rectifier and the point where one end of each of the input
main smoothing capacitors are connected together.  Trying to follow the PCB
traces was nearly impossible but buzzing out lightly suspects with a
continuity tester identified a MAC220-8 mounted on the same heatsink as the
main chopper device plus a temperature sensing thermistor right at the input
end of the power supply as very likely to be it.  It wasn't shorted which was
disappointing but I decided to take it out of circuit anyway.  It was necessary
to unsolder the MAC220-8 and the thermistor using a solder-sucker to free the
leads, unscrew the heatsink from the PCB and lean it over to one side in order
to be able to unscrew the MAC220-8 from it.

After soldering the thermistor back in, replacing the MOV again and powering
up with the triac out, the light bulb glowed slightly for a brief period at
switch on but after that there was no visible illumination while the power
supply worked nicely.  However, the power supply should only work on 220V
with the triac missing.

I extracted the same triac from the leakage damaged H7821 and compared the
two devices.  It turned out that the one that had caused the difficulties
measured about 2.2 megaohms between MT1 and MT2 in one direction only when
out of circuit while the other triac read infinity in both directions.

I decided to fit the good triac into the power supply so that it would then
work on either 110V or 220V as originally designed.  Happily, it still
worked fine after this was done.

If I hadn't come across a similar fault before in a more accessible PSU,
I think I would have had a very hard time tracking down this fault.


Flushed with success, I moved on to look at a H7816 from a DEC 3000/600
Alphaserver.  This one shows little or no signs of life except for a barely
audible click at switch on.  The fuse is not blown and there are no obvious
signs of distress anywhere.

This one is a real pig to work on.  Firstly, the incoming earth connection is
made to the lid of the power supply so when the lid is removed to work on it,
there is no longer any earth connection to the rest of the unit.  A sticker is
thoughtfully placed on the underside of the lid to warn about this situation.
There are also two sub-boards attached to the underside of the lid.  One is a
fan controller circuit (partially obscuring the sticker) which is easily
removed and put out of the way.  The other sub-board contains the mains input
fuse, bridge rectifier, a 110V/220V autosense circuit and what seems to be
either a soft start circuit or less likely, a circuit to cut the power under
computer control.  It is connected to the main board via a plug and socket and
five leads which are only barely long enough to reach.  The plug has to be
removed in order to take the lid off and then 300V DC is no longer fed to
the main board meaning that measurements cannot be taken on the main board
to investigate what might be wrong.  The real doozie is that the sub-board
has a heatsink on it which is directly connected to the live side of the
incoming mains for no good reason that I can see other than to make it
difficult to work on.  Trying to plug the sub-board back in without the lid
in place would leave this live heatsink flopping onto the components mounted
on the main board below.  As well as all this grief, a whole bunch of low
power components are mounted on another board which clipped under the main
board about 10mm away from it and connected to it using three multipin
plug/socket combinations denying access to the print side of the main board
and the component side of the low power board whenever the unit is able to
operate.  There seem to be 20 Ohm load resistors across all of the outputs,
making it difficult to check the output rectifiers and smoothers without
desoldering stuff, in the event that the power supply turns out to be
tripping due to overcurrent.

On a positive note, there are component references on the boards, the
minor boards are single sided, the unit is clean inside and one side
is completely open allowing some access with the lid on (while keeping
away from the live heatsink...)

I took on this machine knowing it had a power supply fault.  How hard
could it be to fix it, I thought :-)  I'm hoping I have an identical
power supply in another machine which I can use to make comparisons to.

Regards,
Peter Coghlan.