[Gabriele, Federico]

Last night we left the acoustic emission test bed running, with an excitation that gave us about 210 um peak to peak motion of the blade tip. As before, we collected one-hour-long alternating stretches of data, with drive on and off. The difference with respect to before is that microphone 2 has been moved further up the blade.

Both microphones signals still show wider distribution when the blade is driven with respect to when it's not driven. However, microphone 2 (the one further up the blade) shows a smaller increase in the distribution width. We don't have a good explanation for this just yet. The following plots show the histograms of the microphone signal RMS at 4kHz.

To improve the analysis, we considered only the driven data, and tried to correlate the distribution width with the force we were applying to the blade. The force is estimated using the DAC output, in coils. We haven't calibrated it in netwon. 

The following plot is a bidimensional histogram: the color gives the number of times the microphone was at a given value while the DAC was at given value. For each bin of the DAC value, we normalized the microphone histogram to the total number of points. In this way we can get rid of the different number of points that have given DAC values (in other words, the DAC value, being a sinusoid, stays more at the extreme than at the center, and we have to compensate for this effect in out histogram, otherwise we can't compare the microphone signal distributions at different DAC values).

The colorscale is hihly saturated to show the tails of the distribution, and the colorscale is logarithmic. It's apparent that the wider distribution that we saw in the first plot are there only when the drive is close to the maximum values. So we have an increase of the microphone noise when the absolute value of the oscillating force we are applying is large. We can't conclude that this is crackling noise, but it's a step in the right direction, even though we were expecting to see an increase of noise when the force derivative was larger. 

Next steps:

• run with different amplitudes and frequencies, to see how the noise changes
• derive a numerical estimation of the "ditribution width" to have quantitative results
• repeat the above analysis with the full rate signal which should have a gaussian distribution if it's only sensor noise