FFT analysis of striking a automotive brake pad with a hammer

Status
Not open for further replies.

cbuxton

Member
In need of a little guidance!

I work for a manufacturer of automotive brake pads.
At the end of the production line we have people who's only job is to "Tap test" a pad. This involves hitting the brake pad with a small hammer and listening for the tone.
This is done to check for voids or cracks inside the brake pad. A "bad" pad will have a different tone.

So as a tinkering project I decided why not use a microphone, perform an FFT on the signal, extract the the 1st, 2nd and 3rd fundamental frequency peaks and then compare to a known good pad.
There is a reasonable frequency shift between a good and bad pad.

I got much further along than I expected by using the Teensy 3.6 and the audio library. Great job Paul!!

Tap Tester.jpg

The pad is normally struck 4 or 5 times. So I plot a max fft output line (magenta) and look for the peaks. The yellow lines are the last few fft's recorded.

From the scale you can see the frequency range I'm working with.

What I need to do is increase the number of bins (currently 1024 at 43Hz increments) to increase my accuracy of frequency measurements.

I have read many posts on this site and not really come up with a definitive answer on how this can be achieved.

I am not a good enough mathematician to understand the in-depth details of the FFT, so any technical help would really be appreciated.
 
What I need to do is increase the number of bins (currently 1024 at 43Hz increments) to increase my accuracy of frequency measurements.

increasing the bin numbers can only be done with increasing the FFT size.
But this is mostly important for resolving two adjacent tonals.

You have an impulse sound (hit of hammer) followed by a possible long tail of reverberation.
I guess, it is this reverberation that is diagnostic for the brake quality.
My second guess is that you are looking for some resonances in this reverberation that are coming later.

So, what I would first do is not to plot a spectrum but a spectrogram, possibly with a large overlap.
I would then compare then the spectrogram between good and bad brakes.

How to do that? would depend on the type of sound. Do you have a short snippet of a good and bad brake you can share?

Back to your original problem:
If all information is within the 25 ms of a 1024 point FFT and you wanted only a better estimate of the location of the spectral peaks
then the method suggested by DerekR https://forum.pjrc.com/threads/3635...olator-for-Frequency-Estimation?highlight=FFT is what you are looking for.
There may be some multitone derivation of this method discussed elsewhere in the forum.
 
Status
Not open for further replies.
Back
Top