Strain measurement error analysis due to natural frequency

[maxwellb]

I'm having trouble figuring out what the accepted way to deal with a strain measurement system's natural frequency is for the purposes of error analysis.

We have a strain measurement system which is basically a vertical plate mounted on a stick. Gas is forcefully sprayed at the plate, causing the stick to deflect. A strain gauge measures the stick's deflection.

The geometry and materials combine to result in a natural frequency of about 40 Hz. The oscillation amplitude can be as large as 20% of the DC signal of interest. Obviously, using the standard deviation as a measurement of error would result in something like 20% uncertainty. I feel like this is an unfair characterization of my system, because it's a periodic error.

Anyone have any insights on how to express the uncertainty in a strain measurement with a large, low frequency oscillation?

[Seth_B.]

Are you interested in just the DC amplitude of the signal or do you care about frequency components as well?  If you are only interested in the DC amplitude, you might investigate a software low pass filter that will remove frequency components above a certain value.  If you need frequency information as well, you could use a notch filter to remove frequencies near 40Hz.

[maxwellb]

Hi Seth,

I am only interested in the DC amplitude.

I have tried some software Butterworth filters, but have not been happy with the results. The problem is that the time region of interest is short relative to the oscillation period (only about 5 periods or so), and there is a sharp discontinuity in DC amplitude immediately after the region of interest. If I make the order high enough or cutoff frequency low enough to accomplish much smoothing, an unacceptable level of phase distortion is introduced, on the order of half of the time region of interest. If I filter the signal in forward and reverse to remove the phase distortion, the discontinuity causes an overshoot which again drastically reduces my usable time.

I haven't tried doing this filtering after cropping out the region of interest from the signal. That may solve these filter problems, but I might as well just average to extract the DC value, because that's what I will be doing anyway. I'm also worried that a filter strong enough to smooth this signal will also remove other frequency components that might be more meaningful in terms of error. I'm just looking for a good way to express the uncertainty in DC value.

Again, I have no idea how people who look at strain gauge data usually treat this. It may well be that if my natural frequency oscillation is 20% of my signal, my uncertainty is 20%. If that's the case, I'd like to know.

[Henrik_Volkers]

Use the tone detection vi to find that frequency. Cut exact n periods and take the mean value.

Or use the

Basic Averaged DC-RMS VI

Owning Palette: Waveform Measurements VIs

Requires: Full Development System

Calculates the DC and RMS values of an input waveform or array of waveforms. This VI is similar to the Averaged DC-RMS VI, but this VI returns only one DC value and one RMS value per input waveform.

[maxwellb]

Hi Henrik,

I don't have a problem cutting out n periods. I've worked out a good solution for that, so my mean value isn't distorted by incomplete cycles.

My only issue is expressing the uncertainty in this mean. With a normal DC signal, I would just use the standard deviation. How do you calculate the uncertainty in the mean value of an AC signal?

[Henrik_Volkers]

If you do a sliding average, you can do the same thing (well, you miss values for at least one period at the beginning)