Combining the Spectral Coherence with Informative Frequency Band Features for Condition Monitoring Under Time-Varying Operating Conditions

Vibration-based monitoring is very popular for expensive gearboxes found in wind turbines for example. However, the fault information of the damaged components is usually masked by high noise levels and time-varying operating conditions. Informative frequency band identification methods allow the fault information in the signal to be enhanced, which facilitates incipient fault detection. In conventional informative frequency band identification methods, Short-Time Fourier Transform or Wavelet Packet Transform estimators are used to identify the frequency bands of interest. However, in the aforementioned estimators there is a compromise between the time and the frequency resolution of the resulting bandlimited signals. This has been one of the motivations of developing the order-frequency spectral coherence-based IFBI(alpha)gram. This IFBI(alpha)gram is constructed by estimating the Signal-to-Noise Ratio of the predetermined component-of-interest in the frequency bands of a set of order-frequency spectral coherences. Thereafter, the optimal band for detecting the component-of-interest is determined by max-imising the IFBI g gram. The order-frequency spectral coherence simultaneously displays the spectral content and cyclic content of the signal under time-varying operating conditions and enhances weak fault information. Hence, it is very effective for identifying informative frequency bands. However, the suitability of using other features instead of the signal-to-noise ratio measure has not been investigated. Hence, in this work, new informative frequency band feature planes are calculated by combining multiple order-frequency spectral coherences with different features such as the L2/L1-ratio, the negentropy, and the kurtosis. The effectiveness of the diagnostic features is investigated on gearbox data acquired under time-varying operating conditions.