I Need Help! Analyzing Frequency Components of Accelerometer Vibration

[Crispo]

Hi,

I'm trying to analyze a voltage output signal from an analog accelerometer using Labview Full Development System.

I do not have Sound and Vibration Toolkit, I am new to Labview and also new to Signal Processing and Analysis. So it would be helpful if anyone could tell me if i'm doing it wrong. (or right.. )

My objective is to identify the frequency components of the vibration.

I'm attaching a .doc file showing the block diagram of the code and sample readings i get. Could you please look at it and tell me if there's something wrong with the way I did it?

Sincerely,

Jason Ray I. Crispo

[jyang72211]

1.      To increase the resolution, you need to look at the formula for resoluation, Delta f = (sampling freq)/(number of sample read).  For you the result is 20k/1k = 20.  You can increase the number of samples read to increase the resolution.

2.      Yes, you need to rename the axis.

3.      I think it is just an artifact from the FFT.  Try subtracting out the average from your data and see do you still get the spike or not.  Maybe someone can common more on this?

[preston_johnson]

May I ask what hardware you are using?  Normally, we use AC coupling for an accelerometer to remove the DC bias which is from its power supply. 

An alternative is to find the median value of the time series signal and subtract this value from all the data. 

[Crispo · Accepted solution]

---

jyang72211 wrote:

1.      To increase the resolution, you need to look at the formula for resoluation, Delta f = (sampling freq)/(number of sample read).  For you the result is 20k/1k = 20.  You can increase the number of samples read to increase the resolution.

2.      Yes, you need to rename the axis.

3.      I think it is just an artifact from the FFT.  Try subtracting out the average from your data and see do you still get the spike or not.  Maybe someone can common more on this?

---

@jyang, thank you for that. I didn't know about the formula for resolution. this is why i like forums. I would never have figured that on my own.

I'll try what you said on #3 and post back the result here. 

[Crispo]

---

preston johnson wrote:

May I ask what hardware you are using?  Normally, we use AC coupling for an accelerometer to remove the DC bias which is from its power supply. 

An alternative is to find the median value of the time series signal and subtract this value from all the data. 

---

I'm using a 2-axis ADXL322 Accelerometer. The x and y output of the accelerometer are each in series with a 0.1uF capacitor.

One Vss pin of the accelerometer is connected to 3V source. the X-axis is connected to AI0 of the SCB. The Ground pin is connected to AI GND of the SCB. The SCB is connected to an NI PCI-6251 DAQ.

I attached a schematic of the accelerometer.

If i do AC Coupling, how will I know what capacitor to use?

For the other alternative, you suggested subtracting the median value from the time series signal. Will there be a difference if I use the mean/average of the signal (device turned on but without vibration) instead of the median? I will be trying this today and post back later how the experiment went.

Thank you preston.

Message Edited by Alterock07 on 02-22-2010 08:09 PM

[preston_johnson]

I like the median function as it sorts its data and picks from the middle of the list.  In other words, it discards outliers.  The Average or Mean adds these into the equation.  A spike from noise of from something else will impact your average and risks not giving you a true DC value to subtract.  My experience suggests the statistical median is much more robust in removing DC offsets from signals. 

[jyang72211]

No problem.  Keep us updated.

Yik

[Hummer1]

Here is a little background on sampling theory.

The faster you sample, the higher the resolution you get in the frequency domain when you do the FFT on the data.

http://en.wikipedia.org/wiki/Sampling_(signal_processing)

Hope this helps

Hummer1

[Crispo]

---

preston johnson wrote:

I like the median function as it sorts its data and picks from the middle of the list.  In other words, it discards outliers.  The Average or Mean adds these into the equation.  A spike from noise of from something else will impact your average and risks not giving you a true DC value to subtract.  My experience suggests the statistical median is much more robust in removing DC offsets from signals. 

---

Hi,

I've tried both the mean and median and although I didn't really find a significant difference, I'll go with the Median since your explanation is very logical.

Anyway, it works! After removing the offset DC by subtracting the Median from the data, the large noise in the low frequency domain disappeared!

Thank you!!!

🙂

I attached a .doc file with the block diagram for you to see.

I also added the "Extract Multiple Tones.VI" to automatically get the frequencies with high amplitudes. I was able to display its output but I can't seem to connect it to WriteToSpreadsheetFile.VI.

The output of the ExtractMultipleTones.VI is:

3 Numeric Indicators ->inside a Cluster -> in an Array -> in a Cluster -> in an Array.

How do I break this down so I can write the data on a spreadsheet file?

Any suggestions on how to do this?

Sincerely,

Jason Ray Crispo

[Crispo]

---

jyang72211 wrote:

No problem.  Keep us updated.

Yik

---

Your suggestion worked! thanks again jyang! Finally the resolution is now per 1Hz. Finally my experiment in moving forward again after being stuck with that obstacle. You can see my block diagram from the attached file from the previous post. I'm also facing another problem now which I stated at the previous post. Do you have any idea on how to go about that?

Again, thank you for your help!

Sincerely,

Jason Ray I. Crispo