I have 2 spare units which were purchased privately from US sellers.
Both turned out to have intermittent faults. One of them, I later found, had been back to Sandel several times for an “overhaul” (hmmm, many of us know that that is: a bench test and if it works then you tag it “serviceable” and sell it to some sucker) but the paperwork didn’t show any actual work done.
So I set out to repair these, because they might be handy one day for something.
The internals were documented here and this refers also.
This unit uses three yaw rate gyros and three accelerometers. That is the classical “3D navigation” setup, established since the V2 rockets (they were simpler) and the Polaris missile had it done properly. However, these components drift a lot and are not suitable for kind of navigation, despite being mounted on a large aluminium block heated to +65C. As noted previously, these are ADXRS613 and ADXL321J respectively. The 1st is discontinued and has apparently led Sandel to redesign the product, but actually the ADXRS623BBGZ is exactly the same and is a current chip.
The 1st SG102 had a weird fault: after a period which could be under an hour, or could be several days (obviously only on a bench test since no flight will be that long), and it would start a wild rotation
This video was done using the MAX Technologies PCMCIA card (costs $xxxx but was found on Ebay Germany for peanuts) with their $xxxx software which is free for 1 TX and 1 RX channel, running on an old winXP laptop. They also do a PCI-400 and IPM-429 (PCI card with a daughter board) for installing inside a PC and this is a much more robust approach.
The next part describes the repair.
It looks like a pitch failure but actually the data from all six sensors goes into some software (a Kalman filter, for those who like to look like experts 
) and this processes it into pitch, roll, and heading. The heading needs a fluxgate connected too. So actually it could be any of the six. However, there is an RS232 module (which costs $xxxx too but I picked it up included with one of the SG102 units) and you can get debug data
In the above, the Y (pitch) gyro indeed shows -19 which looks wrong. However it later turned out that the Z gyro was duff too and the -0.033 output just looked right…
Changing the gyros is a right bugger. They are BGA packages
and come with solder balls underneath which are just the right size. One problem is that these are unleaded solder which needs a lot of heat to melt, and you can easily damage the board. I found the best way is to remove these balls (with a soldering iron) and replace them with 60/40 solder balls, which are much smaller than original but are easily formed by running an iron over the pads, with plenty of flux applied.
Here are the accelerometers, of which there seem to be two (one does x,y and one does z)
In the end, to make it work, I replaced all three gyros… There is a residual error of 2.5 deg on both pitch and roll; this is from the accelerometers only. There is obviously some factory calibration process for the SG102 but Sandel won’t tell me what it is The RS232 debug port takes keyboard input and some key should get you into a factory cal screen but I haven’t found it yet. Anyway, such a small pitch and roll error is easily fixed by mounting it on a spacer.
The gyros can be obtained from Mouser.com for about £50 each, or about $7 from China as I found out later This is 25 year old stuff; you can get much better devices nowadays for the motor vehicle market, from e.g. Bosch.
The test setup showing the fluxgate and the SG102 RS232 adapter
It turns out that the ADXL321J is also discontinued!
I posted this on a US site in the hope that there is somebody out there who knows the “secret” calibration commands which Sandel use to zero the gyro offsets. No luck at all; just total silence even from those who should know.
I now have three SG102 units here. For experimenting, I replaced the three yaw gyros in 2 of them and they work, but since the gyro offsets are not nulled-out, they are useless, because any significant offset confuses the code (Kalman filter etc?) which processes the gyro and accelerometer signals.
The 3rd is the new SG102 version (which incidentally mounts upside down, and does not use the rubber mounts anymore) and that one runs fine, for days or weeks, but just sometimes goes off “on a trip” where it does what the other two do (pitch, roll, yaw/heading just go round and round). Power cycle it and it is again fine for days. I have never opened this unit and anyway would struggle to know where to look since it happens so rarely. So I keep bench running it, sometimes putting it in a -25C freezer and then back on the bench. According to Sandel this unit has never been back to them (they are able to confirm this kind of thing) so there is no reason to suspect that it has an intermittent fault (usually this is evidenced by having been back to the factory for an “overhaul” multiple times, always “no fault found” ).
All 3 were bought from private owners in the US who removed them, reportedly from working installations. The 3rd one was “never installed” and looks it.
These units sometimes come up at SEA in Florida – the usual source for US avionics
https://www.seaerospace.com/sales/search/sg-102
for 1k-2k USD. The one currently in the aircraft came that way, about 2 years ago, and that is the new model also.
There is very obviously a big reliability issue with these boxes. I reckon they have a 100% failure rate within 5-10 years. On the two whose gyros were replaced, not only at least one gyro had failed but also a multi-channel delta-sigma ADC (which converts the gyro and accelerometer signals, plus I believe returns the heated block temperature displayed in the RS232 debug) failed on both units, undoubtedly due to thermal cycling to +65C at every startup (that chip should not be heated like that).
There are two heated block temp sensors. One is in the ADC and the other is an LM50 chip. I believe one of these is used to control the temp and the other drives the readout in the RS232 data. I can’t easily tell which one is which.
Yet another lesson in buying used avionics: always buy 2 or 3 
Maybe 30 years from now somebody will come across this thread and will tell me… I PROB90 won’t be around by then
If I recall correctly: On the Harrier we had two pretty critical control modules – the FECM and the FUCM – which scheduled the flaps in accordance with nozzle, undercarriage and lots of other parameters. Important stuff because if they played up crashes like this were the result https://aviation-safety.net/wikibase/55518
We had one rogue unit that kept getting passed around aircraft, bench testing and manufacturers repair but would still randomly throw a wobbly at unknown times. In the end a Warrant Officer took said unit and it had a unfortunate accident from height in the car park and was finally scrapped..
None of this may be helpful to your Sandel situation of course
Intermittent failures in avionics units resulting in removals, followed by NFF at shop and return to service are relatively common occurrences in transport aircraft maintenance too. However, there is sufficient redundancy on flight-critical ones that it is extremely rarely a safety concern. That Harrier design is a bad one.
NFF SNs are traced both at shop and airline level, perhaps for the occasional car park accident?
That is OK if in warranty, or in the MOD 😁
