FFT and sample window

[jeff_scharpf]

I'm using Signal Express but this question applies to any software. I'm hoping someone can shed some light on this. I'm using an NI-9233 card to measure acceleration. I set my sample rate to 25K, and put a filter of 10K. I use the software to trigger.

My setup is the accelerometer mounted to the back and rear of the handle of a hammer (lower where you would hold it). I was asked to measure the result when a small steel ball is dropped (lightly) to impact the head of the hammer. The accelerometer is a PCB brand 1000g pezioresistive accelerometer, (mechanical BW around 10KHZ).

My question is this: Is it feasable to perform an FFT of any sort when you are just going to "hit" the hammer and measure the result? I am not measuring the actual impact, only the vibrations after it's hit. When I sample for, let's say .5 seconds, I get a nice time signal, and then I apply an FFT. When I shorten the acquisition time to 40 milliseconds, I'm obviously only capturing the very beginning, and my FFT looks different (higher magnitude), so I began to wonder what is "real"? Does and FFT need a repeating signal? Is there a limit to how "small" or "large" of a time signal I have?
Just looking for some general guidance as my time domain - frequency domain skills are so so right now.
I have the option of using an impacting sensor to just measure the impact energy but this is a large sensor. My accelerometer is small. I am not exceeding the rating of the accelerometer as far as I know. Am I wasting my time with an FFT?

Regards,

Jeff Scharpf

[johnsold]

Jeff,

Does the FFT need a  periodic signal?  No.  You can Fourier transform any kind of time signal (within some mathematical limits that won't be a factor for signals derived from real physical systems).

Limit in how large or small?  Mathematically, no.  However the duration of the sample set affects the frequency resolution. It can also appear to change the shape of the spectrum (a very large topic referred to as windowing).

The question you should be asking is "What do I want to measure?" Simply saying the vibrations of the hammer is not very specific.  Are you looking for the amplitudes or energy in certain frequency ranges or in certain time intervals?  What are you gong to do with the information?

Your data acquisition device measures acceleration at sequential times.  The FFT lets you look at that same information as a function of frequency.  If the frequency varies with  time and you want to see that variation, you need to use joint time-frequency methods.

Let us know what you need to do and perhaps we can help. 

Lynn 

[jeff_scharpf]

Wow thanks Lynn for the response. I feel that I may be in over my head 🙂

"What do I want to measure?" I have two hammers. They are almost identical with the exception of the handle material. I am told that one hammer does a better job of absorbing the shock or vibrations. I am not sure if this is just a marketing ploy or not. I need to somehow measure this if that's even possible. I've tried actually using the hammer and I don't notice a difference.

I was given a setup consisting of a wooden frame that hangs the hammer with some bungy cord. There is a steel ball that is on a string, such that you can lift it up and release it. It swings down and hits the head of the hammer.

I have several concerns with this including the fact that it's not simulating a real nail-in-to-wood impact.

In any event this is what they gave me for the moment. I am told to put an accelerometer on the handle and measure what happens when the head is hit (the hammer is hanging free with the bungy cord, I'm not holding it).

So, beyond that I'm told "figure it out".. I believe some marketing guy wants to "show" this difference in hammer handles but that's my speculation.

Amplitudes would be great if they were relevant. I just don't know.

I've noticed that if I change the total time that I sample. (same rate but longer time) then I do the FFT, it changes the amplitudes of the various frequencies.

Other than this I'm sort of running blind lol.. :

Jeff

[johnsold · Accepted solution]

Jeff,

OK.  You have a definition of the problem: Tell the marketing guy that you can measure some difference between the two handles. 

I assume that both use the same head, that the head is solid metal and that the handles are of different materials.

You have  a means of generating a consistent stimulus.  If you always release the ball from the same point you should get fairly consistent excitation of the hammers.  It is possible that the way the hammers are hung in the frame could affect the results, but my guess is that this will be a minor effect.

I would guess that the response of the head is a complicated resonance excited by the impulse produced by the ball.  It will be complicated because the shape of the hammer head is asymmetric (unless it is a sledge hammer) and because of the loading produced by the handles.

I would try measuring the response of  the accelerometer at a few places on the head of the hammer (places where it will not get hit by the ball!).  If you can get something here which is fairly consistent between both hammers then you have a common input reference for comparison.  If the responses of the heads are quite different, then it will be more difficult.

Then I would look at the response of the handles in comparison to the heads.  Two accelerometers, one on the head and one on the handle, measured simultaneously is ideal.  I would look for differences in damping.  That is, does the ringing die down faster on one handle than on the other?  This is a time domain measurement, but it may require filtering to select a single resonance at a time.  I think you want to get as much data as possible, the half second or longer that you mentioned earlier.  The short segments are likely to be harder to interpret and may not have al the information you want.

If you can post some data, someone may be able to give you some other ideas.

Lynn 

[jeff_scharpf]

Thanks again Lynn! I had not thought about putting an accelerometer on the head. I was a little nervous about that but I'll see what I can do. Yes the head is the same as far as I can tell.

 .

I didn't measure the weight of the ball but I chose an abitrary drop height and tried to maintain this. It is hanging with a string. The hanging mechanism for the hammer was a piece of bungy cord on either side held in place by a standard hose clamp. Not sure if tightening the hose clamp would affect things. 

I will try to get some data attached. Right now I have some large excel spreadsheets which are probably too big to post. (several meg each).

[johnsold]

One way to post data is to run your LabVIEW program which acquires the data.  Then mark the array indicator which contains the data with Edit >> Make Current Values Default.  Then save the VI again.  Post the VI with the saved data.

Lynn 

[jeff_scharpf]

Ok I believe that I am at the point where I can continue. I won't post the data just yet as I want to start fresh. This is being done in Signal Express so I don't have any LV code yet.

Thanks for the assistance, Lynn. I will see what kind of numbers I get and post them either in a jpg format or something not so big.

It will take me a few days as I was just pulled off of this again 😞 

Regards,

Jeff

[johnsold]

Jeff,

I understand about getting pulled off.

A text file with one value per channel per line will work fine.   That way we can get at the actual data.  Excel files require Excel to read (or some other program which can interpret them) and .jpg files are compressed so the raw data is not there.  Anything i do with the data would be done in LV, so I would prefer to avoid using intermediary programs just to get to the data.

Please come back when you have more questions.

Lynn