Freesound Forums

Hi,

I am mechanical engineer. Me and my team are trying to create a system to detect irregularities by sound analysis (more flexible than vibration analysis). We will use DFT to transform the time-series data to frequency domain. After that we will use neural network and unsupervised learning to detect any miss-operation.
The software stage is quite straightforward and almost finished. The hardware side is rather complicated.

1) What is the most reliable method to collect the sound waves as time-series from the environment? Do we need a high investment?
2) Any theoritical resource on sound waves?

Any help is welcome

anilbaseski
What is the most reliable method to collect the sound waves as time-series from the environment? Do we need a high investment?

Possible lead ... http://www.innovatemotorsports.com/forums/showthread.php?t=5843

I am assuming that as mechanical engineers and as you mention looking for irregularities, you are making measurements on solid objects (possibly metal pieces).

The best way to make sound measurements directly on solid objects is by attaching a contact microphone.
A contact microphone works, as the name implies, via direct contact with an object and measures the sound within the object or the vibrations transmitted through the object itself.
Normal microphones measure vibrations transmitted through air, and will be subject to all sorts of other complexities and interfeerences which you do not desire (ambient noise, reflections from the walls of the room where measurements are performed, etc).

Now, as far as equipment and investment goes... ceramic transducers make excellent contact microphones and are very very cheap.
I suggest you coat them with enamel paint for protection (they are sensitive to scratches and so on on the ceramic portion. The wires also tend to come off if subject to any pressure or shock and are very difficult to weld back on). So, do apply a varnish or enamel paint caoting for protection.

The ceramic transducers are also used to produce sound, so you can attach two to the object being analysed. One to work as the sound emmiter and the other to work as the sound receiver. They also work up into ultra-sound frequencies, which could be useful for your work.
Producing sound out of ceramic transducers will require a sort of special circuit (I believe it has to do with them being high impedance, but not sure). Projects for these circuits can be found on the internet, so you will be able to feed "sound" from a normal source such as a PC soundcard, via this circuit, and into the ceramic transducer.
Also note some experimenting may be required if you are planning to go up to ultrasound frequencies. I suspect most PC soundcards will not be able to generate such frequencies adequately as there may be filters on the output designed to remove frequencies outside the human hearing range. Also, other equipment on the signal chain (such as amplifiers, or the sound input used for the measurements collection) may have similar limitations in dealing with ultrasounds.

Finally, in were it relates to the attachment of the piezoelectric ceramic transducers to the objects, I have in my experiments obtained good results with tape and glue tack. So again, experiment with different adhesive materials to find what works best for you.
Depending on the shape of the object, holding the transducer against it with a simple clip may also work.

Good luck! And let us know how you get along, I will be interested to know!
Also, if you get any results which could be of interest as soundfiles, feel free to post at Freesound. We love unnusual sounds obtained in wierd situations.

Thanks for your interesting proposals. I guess contact microphone is the most suitable solution as you proposed. We need to have some trials, though.

I dont know about your background; therefore I will try to explain the situation as clear as possible. Our project will concentrate on dynamic machines which are making near-periodic motions (at least they are supposed to do so). Every machine, when working properly, will have a charateristic time response which will be used to create a characteritic spectrum.

If there will be any parameter change (of the machine and/or environment) it will create a correspending change in the spectrum which may indicate a failure. (We will try to listen to the machine :? ) It can be done by accelerometer or vibrationmeter; but for continous monitoring of lots of machines it will cost a fortune. If you want I can send you more details. In theory it must be possible to create such a system.

anilbaseski

By education I am a chemical engineer. Although I work in a very different area these days.
So, I do know a bit of maths and physics and such. Possibly not enough to follow something that is very deep or very detailed in terms of maths or mechanics, but I will certainly be interested to have a look, if you want to share. Send me a private message and I will send you my email.

You mention several machines... This is certainly something different from what I imagined.
I thought you were looking at using sound propagation throught static pieces to detect mechanical flaws such as craks or other defects.

If the various machines are placed close to each other some interesting effects can occour... which may alter the behaviour of the machines and/or influence your tests.
As a classic example, if two mechanical clocks are placed in a wall next to each other, the movements of their pendulums often become synchronized. It is thought that vibrations transmitted through the wall mean that the clocks slightly affect each other and become subtly coupled.
Similar effects can occur between your machines, if they are not vibrationally insulated from each other. Of course, physical proximity and the transmission of vibrations through solid surfaces also can mean that one sensor (i.e. piezoelectric transducer) connected to a machine can pickup sound/vibrations from nearby machines. The swithing on or off of another machine could then give the illusion of a failtue on the machine being measured due to the appearance of additional frequencies.

Withouth knowing any details about the machines themselves it is difficult to offer further advice/suggestions...
It is important to know if the machine frequencies will fall in the range that the piezoelectric sensor can detect effectively. Piezo transducers are not necessarily very good with low frequencies.
Sensitivity is another issue, not only during measurement but also during data analysis. A change in frequency of say 1000Hz to 1001Hz could indicate a failure. In this frequency range most commercial sound analysis tools based on FFT analysis will not give you enough definition to detect the change. But then again, I do not know if you are writting a dedicated analysis program...

Happy to discuss further, although I may not be of great help if it gets too much into theoretical physics...

In any case, good luck with your project and feel free to share any sound files that come out of it in Freesound. I am sure I would would be fascinated to listen to those.

Hi,anilbaseski

I am mechanical engineer too. My team are trying to search a system to detect irregularities by sound analysis via piezo. We will use DFT to transform the time-series data to frequency domain also. After that we will use our experience on collected sound to detect any miss-operation.
The software stage is quite straightforward and almost finished. The hardware side is rather complicated.

1) Piezo is the most reliable method to collect the sound waves as time-series from the environmen and no need a high investment.
2) Any theoritical resource on sound waves?

Any further discussion will be welcometo mail to fomosang@yahoo.com.