Sparse Feature Learning is the name of a concept developed within the SKF-LTU University Technology Centre, where researchers together with SKF is working with new technology for monitoring machines with smart instrumented bearings. The method is applied to sensors that records high-frequency sound waves, called acoustic emission. Acoustic emission is generated by for instance physical processes that lead to wear and tear on machinery.
Acoustic signal warns about damage
– The major advantage of the technique is that we can predict early damage that can cause downtime, says Sergio Martin Del Campo Barraza, PhD-student at LTU-SKF University Technology Centre, Luleå University of Technology.
– The method allows us to simplify the engineering work by processing complex sounds. When the high frequency acoustic signal is detected by a wireless sensor, compared with the different waveforms that are stored in a database. The acoustic signals are classified based on how well and how often they match the waveforms in the database. Different combinations of waveforms can be linked to specific situations that warns about damage to machine components.
The machine learns from database
The database will automatically be updated with machine learning methods. Recurring patterns in the data are identified and linked to physical models or information on other similar machines.
– You might say that the machine is using the experience of situations where the damage occurred, and thus learn to self warn when there is a risk of injury, says Sergio Martin Del Campo Barraza.
The university's research into new technologies for smart machines is, besides SKF-LTU University Technology Centre, are also being made at EISLAB.