Time-varying filtering for nonstationary signal analysis of rotating machinery: Principle and applications

Abstract

Rotating machinery vibration signals are composed of multiple time-varying frequency components under nonstationary conditions, and they usually overlap in the frequency domain. Effective extraction of these time-varying components is vital for rotating machinery fault diagnosis. This paper proposes a time-varying filtering (TVF) method for extraction of time-varying components. Its upper and lower cutoff frequencies are time-varying and independent. The TVF uses locally designed bandpass filters to achieve local filtering in the time domain, and exploits filtering with phase correction (FPC) and local filter lengths to ensure the phase and amplitude consistency of filtered signals. It cannot only extract time-varying mono-components but also separate time-varying frequency bands from highly nonstationary signals, for further time–frequency analysis: (1) fine time–frequency representations based on mono-components extraction, and (2) time-varying demodulation of time-varying frequency band of interests. The proposed method is illustrated and validated through numerical simulation, and signal analyses of typical rotating machines including a hydraulic turbine rotor, an aircraft engine rolling bearing, a multistage planetary gearbox, and a fixed shaft gearbox.