System hears when machines run hot
Unforeseen machine failures cost a lot of money. However, sound-based emissions from machines could be analyzed with comparatively inexpensive microphones to detect damage at an early stage.
A new AI monitoring system by researchers of the Kaunas University of Technology (KTU) from Lithuania analyzes sound emissions, as machine operators once did, in order to detect incipient damage at an early stage. Experts estimate that unforeseen machine failures cost the global industry around one trillion dollars every year.
Expensive machine downtime
"Since sound data is easy to collect due to the relatively low installation costs of microphones for existing plants, sound data-based methods are of great interest," explains KTU computer scientist Rytis Maskelinas. In very noisy factories, however, sound emissions from machines are superimposed by extraneous noise, which often leads to misinterpretations. In this way, damage is indicated that does not exist. The result is expensive machine downtime.
Maskelinas and his colleagues use a damage detection method based on real sound data from industrial machines in perfect working order. The algorithm developed by the researchers in Lithuania compares this data with the sound emissions of the machine on which faults are to be detected. In a training process, the software learns, so to speak, to concentrate only on the sounds of "its" machine and to block out other noise.
Deployment in poorer countries
Modern machines are equipped with a wide range of sensors that ensure reliable early detection of damage. But there are still countless plants, especially in less developed countries, that are not so well equipped. It is for these that Maskelinas has developed the system. He has drawn on an extensive data set of sounds from four pieces of technical equipment. The plan is to expand this data set to include other systems.
