Something that is mentioned again and again as a central advantage, especially among long-time employees, is their experience in recognizing whether a machine is still working or not – just by the noise it makes. But this ability is dying out more and more with ever shorter work cycles. Researchers at ETH Zurich are now trying to compensate for this with artificial intelligence and have developed an AI that is supposed to recognize whether a machine is “healthy” or needs maintenance.
The fields of application for this AI are extensive, because whether it is the railroad, generators or pumps and valves, they all have a unique sound that can have a special order depending on the intensity and dynamics – especially in the case of malfunctions.
AI can detect the sound of healthy machines. Can you? (Source: ETH Zürich)
In order to be able to read these sounds optimally and to perceive even deviations in the smallest decibel range, an AI is needed that supports the employees in their daily work. To this end, the researchers at ETH have developed a machine learning method that interprets the sounds recorded by microphones and is trained on the basis of the empirical values of their human colleagues. The researchers’ goal is to provide experts in the field with a tool that automatically monitors operations, even when all other colleagues are at work. However, the machine learning method not only works for different types of machines, but also for different types of signals, noises and vibrations.
