Optimal Filtering of Cyclic Transients and Its Application to Helicopter Gearbox Diagnosis

The optimal filtering of cyclic transients is a major issue in machine diagnostic. As these components are typically associated with incipient faults, their enhancement improves fault detectability and identification. This paper proposes an optimal linear-time-invariant filtering that enhances cyclic transients in the general scenario of a non-steady operating regime. The proposed solution is based on an angle-time cyclostationary modeling of the signal and the noise, contrary to the widely adopted “spectral kurtosis” which assumes the stationarity of the noise. The proposed approach is investigated and compared with the spectral kurtosis on synthetic and real-world vibration signals captured from a helicopter engine operating under nonstationary regime.