Nowadays, many companies use predetermined maintenance for assuring the reliability and safety of their assets. In contrast, the direction of maintenance is changing towards condition based maintenance with the aim of extending assets' useful life. In this context, condition monitoring is necessary to evaluate the state of machines. The objective of this work is to implement an anomaly detection method to assess the condition of rolling element bearings. For that purpose, vibration data is acquired from a test rig, in which healthy and damaged rolling element bearings are mounted. These data are processed by techniques proposed in the literature and a number of indicators are extracted based on time-domain analysis and frequency-domain analysis. A subspace based statistical model is built using the indicators extracted from the healthy data. By applying this one-class classification model to data related to both healthy and damaged states, the results show a good agreement with the real condition of the component. In light of these results, it is shown that this data-driven technique is valid for anomaly detection of a widely used component in industrial applications.
展开▼
