Abstract: Based on the analysis of characteristics of fault signals generated by mechanisms and stationary vibration signal, a novel method for extracting impulse components from vibration signals is presented. The original signal is decompose into a linear combination of Chirplet functions that are well concentrated both in time and frequency, aiming to get self-adaptive time-frequency (TF) spectra with high resolution and without noise and cross-terms. Then two methods are used to extract transient impulse components. One is using the TF of the vibration impulse signal model to approximate the TF distribution signal being filtered out to determine the reconstructed impulse signal model. The other method is using Chirplet elementary function to reconstruct directly based on the Chirplet expression characteristics of the transient impulse component. The fault signals of gear and bearing are analyzed by using the two methods respectively. The analysis results verify that the two methods are effective for the impulse signals and have great reference values in mechanical fault diagnosis.

Key words: Chirplet, Self-adaptive time-frequency decomposition, Signal extraction, Time-frequency filtering, Transient impulse