Elusive Instrumentation Amp Burnouts - AD8293G

[seancsnm]

Hello,

I've been working with a few guys to do some data acquisition from a load cell. The setup is very simple - a power supply to provide power to the load cell and a signal conditioning circuit, the load cell, the conditioning circuit, and an NI data acquisition device configured with a NI 9215 ADC.

The benchtop power supply is regulated by an LM7805 regulator to provide 5 V to both the load cell and conditioning circuit. The load cell is in full-bridge configuration, with excitation voltage 5V/0V so that the no-load output voltage will be 0V differential, 2.5 V common.

The signal conditioning circuit consists of an AD8293G160 instrumentation amplifier (http://www.analog.com/media/en/technical-documentation/data-sheets/AD8293G80_160.pdf) with a reference of 2.5 V provided by a 100k/100k voltage divider. Some RC filters are also in place to remove high-frequency noise, and power supply input has a bypass cap (albeit not ideally).

The last stage is an NI cDAQ with a 9215 4-channel, +/- 10V differential 16-bit ADC. The input impedance is pretty high. I've attached a DaveCAD-style schematic for circuit specifics.

Seems simple enough, right? We soldered the SOT-23 package to a breakout board, built the circuit on a protoboard, and started collecting data. Things were fine for a couple hours but at some point things just stopped working. No accidental short circuits, no out-of-the-ordinary events occurred. Turns out the op amp died. Thought maybe it was a defective amp (since I've used this amp pretty extensively in a similar manner in the past). So desoldered this one and soldered another on. Same thing - worked for a couple hours, then died. At this point I start looking through the datasheet to see if maybe there's something we're doing wrong. It's pretty tolerant of the soldering process - rated to handle 300C for 10 seconds. Pretty sure I was within that constraint. Has ESD protection diodes. We weren't touching the circuit while testing was occurring or near the time of failure, so I doubt this would have been a problem anyways, but good to know an ESD death is very very likely not the issue here. Can handle input voltage range between GND and +VS. There's no way for the load cell voltage to go out of this range for more than a couple microseconds due to emf causing transients through the 3-4 ft length of the load cell cable. Can also handle shorting of the output for an indefinite amount of time. This should never be a problem anyways, even if the DAQ were to malfunction and short the output, since the output is connected to the DAQ through a 10 kohm resistor. All circuit components appear to be working as well with no measurable resistances across capacitors. Voltage regulator voltage is stable (in fact, between the first and second amp, we switched power supplies entirely to make sure this wasn't the issue).

What is interesting is that as one of the op amps was failing, it got very hot to the touch. Seeing as there should be minimal current flowing through it, this didn't really make sense. After failure, it was found that the resistance between +VS and GND on the 2nd amp was pretty low. Measured about 1.5 kohm. The third amp measured about 40k ohm after failure. A good amp is in the megaohm+ range between these two terminals. If you look at the simplified schematic, you'll see there are several fets between +VS and GND, with one of the gates being connected to an input. So there may be indication of one of one or more gates failing, causing the amp to short out through the fets.

But even if this is what is happening, I have no explanation as to why this is happening. Any thoughts or ideas for further investigation? I'm very stumped.

[DC1MC]

Last time I've used a load cell was on a huge US made crane for measuring the arm load. Some undiscovered genius electrician removed the shielded dual coax cable from the cell to the amplifier to use it for TV antennas  (a big deal in the '80s) and he replaced it with some similar looking 3 wires electricity cable, he even connected the neutral to the ground. The crane was having engines with 600-1000A startup current and our FET based amplifiers expensive AND hard to obtain by an Eastern block country were suddenly dieing in an hour or two of work.
The electronics guys were not able to see what cable was used, being hidden in the crane skeleton, and this was completely incomprehensible for us (not many scopes available at the time).
After the terrifying  secret police was involved (the PH crane was a strategic asset) and each and every maintenance operation was reviewed, the culprit was discovered, the was cable replaced (the poor soul didn't have time to cut it in pieces and sell it) and the guy who did it was never to be seen, nobody dared to ask what happened to him, part and parcel in living in communist dictatorship.

 The moral of this anecdote:

 Look if your signal cable is not picking any inductive load that will kill the inputs in common mode.


 Good luck,
 DC1MC

[wraper]

Dunno how load cell is attached, wire length, shielding but IMO it's not a good idea leaving it without any input filtering/ESD protection at all. Yes AD8293 has some input ESD protection but it's relatively weak, diodes are rated for 5mA continious only. Cutoff frequencies of internal and output filters don't match, strange component value selection. Capacitor values at 7805 are strangely low, may cause oscillation. I use following circuit in product which has thousands units in a wild with inputs completely exposed to outside and available to the user. Often with ~1kW servo motor running nearby. I'm not aware of any inamp failures.