Looking for Automation System to Differentiate Distinct Sounds

[lollerskate]

I'm currently looking for a system which can automate a part of my process. Currently, there is an operator listening for sound, which is distinct enough for the human ear. Upon hearing this sound, he shuts down the system. I would imagine that if an ear can easily differentiate the sound, a sensor should be capable of doing so as well. However, I've only had the luck of ONE company coming out so far, and he was unsuccessful at getting his sensor to do so (it was a system originally meant for sounds in a grinding process). Can someone please direct me to a company or a system which might serve my purpose? I have been using google, but my search terms must not be up to par to find what I need.

 

[minger]

How predictable is the sound. Is it like a tone or a more complex sound? Would you say that the frequency of the sound remains fairly constant?

If its simple enough, it seems like you should be able to simply mic it up and run the data through some sort of real-time FFT. Observe for the frequency + magnitude within some tolerance.

 

[FredGarvin]

Along what Minger said about the microphone set up, you may want to look at machine predictive maintenance systems. While they are meant for measuring machine vibrations, they include FFT sampling and monitoring based on your limits and will alert you if certian limits are met. They shouldn't care what the real input is. If the sound is a pure tone it is even easier.

 

[berkeman]

One other thought would be to look into computer-based voice recognition systems. They can often be "taught" to recognize words or phrases, so maybe you can teach it to recognize the sound.

 

[lollerskate]

The sound is metal on metal contact. It is repetitive since it is made by metal bouncing on a metal shaker table. The sound is not very predictable since the size of the metal coming through can vary. It is only liberated metal we wish to detect, which makes the task harder. The table normally has chunks of rubber going through it, with beads of metal implanted in the rubber. While this metal can hit the table, it doesn't make the same sound as items such as bolts, or chunks of hardened steel.

Problems being observed:
The material originally drops onto the table from a height of ~1.5', causing interference
Since the material sometimes has metal that is only mostly embedded within it (which is fine), there is also interference when from it hitting.

There was a system that was applied to our process, but it didn't look promising. However, with the sound being detectable by ear, there should be a system available.

Thanks for the comments so far. I will look into real-time FFT systems.