Use of the Fast Fourier Transform (FFT) for experiments

[wadi]

G'day all,
With all this talk of resonance/impedance matching I'd like to share this little bit of info with the forum.

We sometimes use a strain gauge which is like a piezometer; there will be a change in voltage (minute) if whatever object the strain gauge is attached to deforms. So we attach strain gauges to reinforced beams of concrete and hit them with a small hammer; which causes the beam to vibrate. The strain gauges are recorded using a datalogger sampling at three microseconds.

Using the data recorded and plotting the voltage vs time we get this sort of graph:-



Which would be expected. We then use a fast fourier transform (fft) which is an algorithm to determine any frequency peaks that should be avoided. So we then manipulate the data (further steps - not related to hydroxy experiments) and we can then determine whether the beam is cracked or still intact.

However peaks are found using the fft, which are frequencies that should be avoided. Do you see any peaks that would look out of the ordinary in the first graph?

This is the result of the fft being applied to the data from the strain gauges:-



There is a huge spike around 22Hz, and a smaller one around 10Hz; whilst the rest appears to be noise. If we were able to cause that reinforced beam to deform at 22Hz it would fail; this is known as it's resonant frequency. And by the term resonance, I mean minimum input for maximum output.

This is done using matlab; a very good program we use. I can upload the script for this program so all that would be required is:- input the data and hit enter and it will generate the graphs and disp peaks found etc.

Here is what is displayed in text besides the graph:-

[wadi]

Just as an add on to the previous post:-

I would think that this could be applied to cell designs when looking for ways to increase efficieny. A datalogger that could measure the voltage vs time, when current is applied to the cell could help in identifying peaks in the fft.

Problems I forsee is:- A pulse or frequency must be applied to the cell to determine if there are any peaks. In the fft there should be a huge peak which should be the input, but I would hope there are other peaks found.

Then applying those specific frequencies (peaks found in the first fft) back to the cell, then repeating with steps again so that the peaks are being applied to the cell at all times (perhaps similar to the hexcontroller? Except it also applies them out of phase).
This would be very tricky to accomplish.

And finally: if you had a whopping number of 60 cells, do you measure the voltage across the whole lot? Or the middle? Or one? Can a datalogger handle the current and voltage?

Thats my input.
Jaro

[waterbard]

[jai_mann]

This is excellent to see this type of a post. I would like to add to this. The lab that I used to work in studies brain waveforms using extremely similar techniques. You've listed two peak waveforms in what we would call the 'alpha' and 'beta' ranges. I would like to make a suggestion for finding peak waveforms in higher frequency bands. We would use Matlab to produce a Spectral graph. The Abscissa would represent (T)ime as a function and the Ordinate would represent the (F)requency. The actual plot was color coded to represent the strength of the signal. As such we could say at time X, there were signal peaks at band Y. The FFT works well but to the human eye you will miss the high and low "gamma" band activity. The spectral plots allow you to catch otherwise missed high frequency signal peaks. I don't know how the higher bands would impact gas production but we certainly do not want to overlook their contribution. Here is a link with images to give you all a better idea of what these spectral plots would look like. http://sccn.ucsd.edu/eeglab/maintut/time_frequency.html

If you scroll back and forth in that tutorial it gives more graphs with similar layouts.