Vibration Best Practices

[markmakeitso]

Hello all,

Working on a vibration analysis program to replace a hand-held single axis wand. My company is seeing the light on a small herd of triaxial sensors (currently mostly ADXL326 and ADXL335 chips) which are rather in the hobbyist sphere rather than Real Serious Industrial Sensors, but so far seem adequate for our needs. The main goal is a number of sensors mounted to an engine under test, which we can sample at will, and perform better analysis than the wand offers. Right now they just want to be able to fill in the application review forms faster (just max accel and displacement at different locations on the engine), but sometime down the line there might be a need for more complex analysis.

Had a chance to make some sample recordings the other day, and seems like mostly good news. Have a few questions about methodology though:

-We used 6 times edge facing to determine +/-1g sensitivity (and by extension a midpoint voltage) for each axis. Once we have our 3 sensitivities what are the drawbacks of measuring each sensor axis while at rest when first mounted in position, and using that as a 0g reading? Any benefit to using that method versus a dynamic midpoint voltage calculated from each captured waveform average? Or compare the initial calibration reading to each dynamically calculated one and alert if disparity is above a threshold?

-How aggressive should I be in trimming my sample size? Leery of using zero crossings, as in the example posted here I have a fairly repeating waveform, but both axes (ignore Y please) have some 3 cycle patterning going on. Should Peak-Peak values always be for adjacent peaks? Theory is nice, when they show a very simplistic curve, but already my curves are more irregular. For example on Z axis on the shown curve could I pick a local maximum value, then find the minimum immediately before and after, then use the higher P-P difference as my calculated value?
-Our usual calibration house can handle analog output triaxial sensors, with some caveats. We'd need to come up with methods for mounting to induce in all 3 axes individually. They can also only induce +/-1G at different frequencies, and they give a sensitivity curve at each frequency. Would feel a lot better if they could induce past the 16g limit of our larger sensor. Worth looking around for someone with more capability?
-Do you recommend removing the onboard filtering caps on those 2 PCBs? Either for developing our setup or for regular use in field testing? If yes, we've considered selectable caps at the DAQ box, or just handling all of that in software. 

Alright, thanks for reading and for any feedback/suggestions you have for us. Before anyone ask, yes, we have slightly considered one of the vibration toolkits. Some industry friends have good things to say about them, but not really in the project budget right now.

Sorry for the wall of text,

Mark

[BertMcMahan]

I can't answer all of your questions, but I think I can discuss a few of them.

-Regarding zero point vs dynamic zeroing: If you're looking for peak to peak values, zero doesn't really matter anyway. I think I'd just take the max and min values and subtract.

-Regarding your sampling: Do you need to spec X, Y, and Z data, or just max and min? I think I'd just combine them all using vector math (current value = sqrt(x^2 + y^2 + z^2))

-Can't help much with calibration, sorry

-Hardware filtering is nearly always better than trying to do the same thing in software. Usually you combine hardware and software filtering to do really complicated analysis. Look up "antialiasing filters" for further reading. Fully reconstructing a signal close to the Nyquist rate is non-trivial. It looks here like you're way above that, so it won't matter much either way, but I'd prefer hardware filters to software ones any day of the week. I'd also put them on the sensor, not on the DAQ board, but that's just me.

[billko]

For one thing, your expensive test equipment is protected from noisy spikes in your system with a hardware filter, since the filter is physically filtering the signal instead of applying an algorithm to the recorded signal.  It is able to absorb that transient spike before it gets to the sensitive equipment, so if you can put the filter in before the sensor, that provides better protection.

[markmakeitso]

Thanks for your input Bert and Bill,

-That's a good point, that P-P doesn't require an accurate midpoint. Would that be more useful when we're calculating speed and displacement however? There's some discussion about correcting for zero drift with integration (I know just enough about some topics to be a pain in the butt apparently).

-For the application review portions we keep all X, Y, and Z axes completely separate, as completing that with the wand is already a moderate challenging task. I have considered implementing a static check after install that determines (using all 3 sensitivities and 3 zero voltages) whether all axis acceleration components together is within a couple % of 1g.
-We did do a little experimentation with sampling rates to see if we had emerging artifacts from higher rates. Looks like some benefit to that strategy. Our current hardware options both offer 500 kS/s aggregate, either for up to 8 AI channels or 32 for a different box. The latter is nice for a bunch of triax sensors, but setup is much more involved. Ideally the program will support both eventually.
-Those are good points about filtering. For input protection I was considering a couple zener diodes, at least a couple volts away from the 0-5V signal output range. I don't know enough offhand to weigh in on smoothing caps at the sensor versus right at the DAQ box inputs.

Eventually! I'd like to change over to newer motion chips, which have scalable G ranges and digital communication rather than analog outputs back to the DAQ box. Not there yet though.

[billko]

I may be a nerd, but all this sounds like so much fun!  I hope Bert gave you some good food for thought.  🙂

Good luck!

[markmakeitso]

Oh, yes, it is very nerdy fun. I thought that was the best part of the NI forums?