A Probabilistic Signal Representation for Detecting Faults in Highly Fragmented and Down-Sampled Vibration Signals

摘要

Classical vibration signals analysis techniques are developed assuming idealized conditions from which data is collected. In reality, due to physical constraints imposed by the operating environment, the vibration data is usually filtered by thresholding into segments which are then down-sampled, therefore not readily amenable to the application of classical techniques. In this paper, we introduce probabilistically reconstructed signals to represent this particular class of data, and show that the application of classification methods on the reconstructed signal is straightforward with reasonable accuracy.