# Comparing vibration frequencies

I have two iron pipes, one bent and cracked, and one whole. On one end Ive attached a vibration motor (200hz) and on the other end Ive attached a vibration sensor.

I do a fft and then get out the amplitudes of the different frequencies I pick up.

Question: I want to compare the signal picked up from the broken pipe to the signal from the whole pipe, how?

At the moment I calculate the "total" energy in the signal, by summing the amplitudes. However, this sometimes fail, especially if im testing a very broken pipe. I can reliably detect smaller damages to the pipes.

Correlation does not work, due to the shape of the signal would be similar, but there would be small variations in the harmonics of the main signal. Also the vibration motor is 200 hz +/- 10hz approximately. Suggestions? Visual check doesnt work as I want to build a completely automated system, where a LED light turns on if the difference in the signal is above some treshhold.

• 200 Hz seems very low for metal structural integrity testing. Also, a single frequency can easily miss relevant properties, even if you assume that there are nonlinearities shaping your output. Have you considered using a broad band stimulus, like hitting the pipe with a hammer or something similar? Also, summing amplitudes doesn't give you energy. – Jazzmaniac Aug 22 '14 at 12:09
• Im not aiming for testing metal integrity, it is only for prototype work. Ive moved on to test "integrity" of Lego constructions. It works well on minor variations in total structure stiffness, but completely breaking the structure also breaks my algorithms somehow. They work but the output tells me there is no difference. But thanks; Summing amplitudes is a dead end. Suggestions to measure energy in the signal? Less structural integrity = less total energy (or clipping/harmonics), right? – Sevenius Aug 22 '14 at 12:32
• Some time has passed and I wonder, whether you found a solution? I am working on a similiar problem (feature engineering for distinction of vibration signals). Which features did you use, to seperate both cases (bad condition, good condition) ? – Nikolas Rieble Oct 10 '16 at 10:02
• Hi, sorry for late reply as I haven't used stackexchange in a while. I never solved it as the company I worked for went bankrupt and I'm now dealing with different problems. However, my latest iteration revolved around searching for peaks around target frequency, and also searching for harmonics of that signal, and comparing their amplitudes basically. It did work somewhat for my original problem, but never to my satisfaction. Sorry I can't be of more help. It was an interesting problem. – Sevenius Feb 22 '18 at 15:49