I have spent several hours looking at your data and various ways of analyzing it. I focused almost completely on the "separate samples..." files.
I tried spectral analysis without much success, although I think it could be made to work. One issue is that the fundamental frequency varies over a 3:1 range in your data set. This makes setting arbitrary frequency thresholds problematic. Similarly the highest magnitude harmonic is often not the fundamental. This makes frequency "tracking" difficult. Pump "mech7" may be slowing down when the piston strikes. It seems to hit on every other revolution and two frequencies appear overlapping in the data: 63 and 71 Hz.
Several of your "good" pumps are actually bad or marginal. I suspect they are just touching, and not on every revolution. Good7 is the worst but good6 and good8 are also questionable.
What I found that worked combined two measurements. One is a simple amplitude check. If the Peak to Peak amplitude in the file is greater than 1.5, the pump is bad. The other resulted from my observation of the time domain signals. The bad pumps tend to have much more variation from cycle to cycle or over 100-300 ms than the good ones. The way I measured this was to divide the array into multiple segments. Calculate the peak to peak and rms amplitudes for each segment and their ratio. When the standard deviation is > 0.20, the pump is bad.
I have attached a VI which demonstrates this with your datafiles. It is not highly polished but should give you something to try. You will undoubtedly make some adjustments to the thresholds. Note that you need to press the Next File button each time you want to read a data file. The event structure makes the program just wait until you push the button.
When you get to the production line watch to see what happens when a pump goes from noisy to quiet. My guess is that you will see a change from mech5 to mech6 to good8 to good1 over a rather small change in your stroke length. You may find that watching the change may be a better guide than absolute thresholds.
Lynn
