Removal of non-stationary harmonics for operational modal analysis in time and frequency domain

Operational Modal Analysis allows to assess the modal model of rotating machinery. However, it is needed to pre-process the measured vibration data such that the influence of the harmonic content is suppressed. In this paper, such pre-processing techniques are discussed. The main focus is on evaluating the effects of amplitude modulation, originating for example from speed and load oscillations in rotating machines, on the harmonic removal. Both frequency and time domain editing techniques are considered. To illustrate the need of taking these modulations into account, their effect on the performance of a classical frequency domain editing technique is first discussed. Afterwards, different methods are developed that are able to cope better with frequency and amplitude modulated harmonics. To this end, it is opted to extract the harmonic response in the time domain using complex demodulation. This response can either be used to directly perform the system identification, or it can be used to design filters. Two different filters are proposed with distinct goals. The first one directly sets the harmonic response to the level of the white noise response based on the demodulated signal, whereas the second one is designed to remove the effects of amplitude modulation. For the latter, it is shown that classical editing methods can afterwards be applied in a more efficient manner. •Discussion of methods for removal of non-stationary harmonics (both amplitude and frequency modulated).•Novel time domain filters for suppressing harmonic content.•Sensitivity study of different developed methods based on simulated data of a rotating machine.•Comparison of different harmonic removal methods in time and frequency domain.