Abstract

Condition monitoring is useful to describe the machine state under current operating regimes, especially, when non-stationary operating conditions appears. Nevertheless, in actual applications the faulty data are not always available. This paper proposes a novel methodology for condition monitoring using dynamic features and one-class classifiers. The dynamic features set comprises the spectral sub-band centroids and linear frequency cepstral coefficients computed from time–frequency representations. A one-class classification stage is carried out to validate the performance of the dynamic features and commonly used statistical features as descriptors of the machine state. Proposed methodology is evaluated by using a test rig, which is composed by outliers (unbalance and misalignment) and target objects (undamaged state). The data set is obtained under variable speed conditions including start-up and coast-down. The attained results outperform other state-of-the-art extracted parameters and the methodology is robust to large speed fluctuations in the machine.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call