Comparison between sound perception and self-organizing maps in the monitoring of the bearing degradation

This study aims to monitor and detect bearing defects from measured signals on a wind turbine during 50 operating days using two methods. The first method involves a perceptual approach to classify the selected signals based on 50 measurements. The second method used is an unsupervised classification method called the Self-Organizing Map (SOM). Overall, the perceptive approach proved to be simple and effective compared to conventional methods of treatment and diagnosis of defects, as listeners were able to classify the selected sounds in the order of bearing degradation, allowing the severity of the bearing defect to be tracked. Furthermore, the neural classifier provided relevant information on the evolution of bearing degradation, as it could automatically cluster the vibration signal into four groups corresponding to the bearing life stages. Thus, these results can effectively contribute to well-timed maintenance decisions. In addition, the advantages and deficiencies of one method over the other are briefly discussed in this paper.