Large factories have a lot of machines, instrumenting each one of them to find when a failure would occur is an expensive process. In the age of big data, we can reduce instrumentation and labor cost massively, and substitute it with computation to identify component failures.
In a large factory, with a lot of machines, Numerical Works Instruments the factory floor as shows in the image above. Just simple mics are used around many places in the factory which listen to machine sounds. The mics gather data both in audible and inaudible frequencies. These data are pumped to a centralized database as show in the image below.
Factories have an inventories and repair databases, which hold the data about which part of which machine was replaced. The factory noise and these repair and inventories data are compared by our neural network algorithm, and it slowly learns to predict which machines will fail when and what failure will happen.
The beauty of this setup is that this can be replicated in similar factory floors, the more floor and machines and mic’s we have, the better the prediction. It improves with scale and investment per machinery drops dramatically.