Accelerometer data post processing

[henrikh2]

Hi,

I would like to extract out of acceleration measuring data a chart, which consists acceleration value, its existance length in seconds and the occuring frequency. I think it is called interval average acceleration versus time chart.

The acceleration data is basicly noise filtered values of acceleration in g-s, measured with 512Hz frequency.

Signal processing is new field for me. Is FFT time domain is the right track where to continue?

PS. I am a mechanical engineer, thats-why its unknown field for me.

Is there a simple solution in MATLAB or Excel for that?

Best Regards,

Henrik

 

[Skogsgurra]

Yes, FFT seems to be the way to go. It will produce a spectrum where you can see peaks and valleys.
Matlab or Scilab (free) have many frequency analysis and plotting functions. Excel is also a possible way to go, but you need a little more than the standard package.

Do not understand "FFT time domain". Looks like you are mixing two things there.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[henrikh2]

To be honest, i dont remember much about Fourier analysis from school time. Only thing I can remember is that it allows to replace an orginal function with bunch of sine or cosine functions. So I looked into some digital signal procesing handbooks and the first thing i noticed was the time domain topic, which could be related to that kind of data transformation.

 

[BobM3]

By "existance length in seconds" are you thinking of an impulse response function? That would show you (in the time domain) how the system responds to an impulse. You'd FFT the data and then calculate the impulse response function from that.

When you say 512 hertz, I assume that's your sample frequency?

 

[GregLocock]

You would be better off asking in the Mechanical Acoustics/Vibration engineering forum.

What you are describing sounds a bit like wavelet analysis, or STFT. it would be easy to do it in Matlab Scilab or Octave, but it would also be easy to get it wrong.

It'd probably help if you could give more detail of what the data looks like, and what you actually want in the way of results, as a general solution would be quite laborious to describe, and i suspect you don't need that.

Cheers

Greg Locock

I rarely exceed 1.79 x 10^12 furlongs per fortnight

 

[henrikh2]

The data is in following format:

-0.0367906
-0.00433198
-0.00831627
....

There is 1025243 lines in this file. Its the resuts of accelerometer from automotive vibrations. Measured with sampling rate 512Hz.

I would like to get a chart which describes which acceleration levels have the longest duration and occur most often.

 

[Skogsgurra]

Sounds like a simple 'plot and look' task to me. Plot the data and have look at the resulting curve. You can do it in Excel, but Excel doesn't like more than 32 000 lines of data in each plot. Never tried to use a million data points in Matlab. Could work.

I recently had 28 gigabytes of data to analyse. Too much for any available(to us) analysis package. So, my colleague wrote a simple routine that sorted out regions that were above a certain, settable limit. We then got a set of smaller files that we could look at and plot from our ARCUS.

That was a very effective way of finding extreme vibration levels in data collected during about one month. We found six occasions that had lasted between two and five seconds each and I think you could do a similar thing. FFT and variations thereof is good for many things, but not necessarily for all things.


Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[Skogsgurra]

Re previous post: Attach a plot of vibration data extracted from 28 GB. Vibration level is the blue trace. Other traces are motor voltage, pressure, DC link and such.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[henrikh2]

I am not trying to find average or max values of acceleration. I am trying to create a 3D plot with following axes:

x - intervals of acceleration duration in seconds
y - average value of acceleration in these intervals
z - probability density of these intervals or the count how often did the average acceleration did occur

I am developing a device, which is supposed to extract with pendulum-like mechanism maximum amount of energy from random road accelerations. So in order to get max efficency for this this device I have to optimize the pendulum inertia and resistance to acceleration + acceleration duration values. It is not a vibration energy harvester, its swinging or bouncing energy harvester.

 

[BobM3]

I don't see any frequency dependance for those three axes. Is the plot for each frequency line?

 

[henrikh2]

No there is no frequency dependance. I am trying to extract basicly the area under acceleration curve (the work, which the acceleration does). And then line it up to see which average acceleration values are most useable (which occur most often and last the longest).

 

[BobM3]

Are you familiar with power spectral densities (PSDs)? They give you the energy contained in the signal as a function of frequency.

The area under the acceleration curve (accel vs time) is the velocity.

 

[Skogsgurra]

henrikh2

Are your data instantaneous acceleration values or are they rms values? I have the impression they are rms. Even if the negative values contradict that. Sampling with 512 Sa/s does not seem meaningful otherwise, since vibration data usually contain components in the kHz range.

I think it would be good to have all information possible instead of 'free thinking' in different directions.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[GregLocock]

(Gunnar 512 hz may be acceptable for what he's doing, most road vibration is at less than 20 Hz)

OK, you don't need FFTs. With a million points I wouldn't use Excel. I haven't seen a plot like the one you describe, although it is very similar to what we'd look at when taking roadloads for durability. There we don't bin by the length of each event, we just use a cumulative frequency plot.

Your proposed analysis needs a bit more definition.

Is the thing you are looking for describable as "duration of event in which RMS level (measured over what time?) exceeds a given level". You are going to have to define an averaging time for the RMS measurement, and I suspect you will end up binning the amplitude data.

I suggest you play around with a short sample to iron out the definition.

And I still think this is the wrong forum.







Cheers

Greg Locock

I rarely exceed 1.79 x 10^12 furlongs per fortnight

 

[GregLocock]

OK, here's my take on what you think you want. The process is easily recreated in a proper programming language. The most time consuming step will probably be the moving average.

I haven't charted the final result. Also note that the sorting and histogramming are not live functions in excel, so this spreadsheet would need a lot of work before it was useable.



Cheers

Greg Locock

I rarely exceed 1.79 x 10^12 furlongs per fortnight

 

[Skogsgurra]

hmm.. 20 Hz? OK. Rubber tyres, car body. Low pass, can see that. Not ordinary machine vibrations.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[henrikh2]

I have a PSD chart from the same data. The g-vlaues are not RMS-ed.

This data post-processing field is too unknown for me . I should do my homework properly. I´ll just start seeking handbooks and articles. Then I´ll get back to you if I am still in trouble.

Greglock, thanks for the spreadsheet. Question about it:

range F12 - there should be instead of A4 a B4 range?

 

[henrikh2]

I wrote the question in so strange sentence, that I don´t even understand it myself . What I wanted to ask from Greglocock is:

Range F12 - there should be inside equation instead of $A$4 a $B$4 range?

 

[GregLocock]

No, it is right, a4 is the time step increment.

There may be smarter ways to time each 'event' but since it is just proof of concept I didn't bother working out a neat method.



Cheers

Greg Locock

I rarely exceed 1.79 x 10^12 furlongs per fortnight

 

[zappedagain]

Excel maxes out at 4096 samples going into an FFT analysis. Watch out for that. When I've had to perform a FFT on a larger, single data set I use Origin to process the data instead of coding something.

John D