Fault feature enhancement for rotating machinery based on quality factor analysis and manifold learning

Abstract

This paper explores an improved time-frequency signature to enhance the periodic transient shocks of the signal, called impulse-enhanced signature (IES) for identifying rotating machine faults. IES is extracted in the following steps: first, phase space reconstruction is applied to the analyzed signal to present its dynamic signature in high-dimensional space; second, employ quality factor (Q-factor) based decomposition on the phase space to separate the fault transient component from the vibration signal; third, utilize the continuous wavelet transform to present nonstationary information embedded in the signal and finally, IES is obtained by optimizing the low-dimension structure, which is extracted from the phase space using manifold learning. The IES significantly improves the fault information with a highly regular representation, especially for weak fault-induced impulses, and its advantages over other approaches include noise suppression and energy concentration. One simple IES based curve, time marginal amplitude (TMA), is extracted to further detect the fault characteristic frequency and evaluate the performance of IES. Simulation and experiments confirm the effectiveness of the proposed method. Results indicate that IES outperforms traditional empirical mode decomposition envelop analysis for diagnosing rotating machine faults.