MIENgram: an optimal demodulation frequency band selection method for early fault diagnosis of wind turbine bearings

Abstract

Resonant demodulation of vibration signals is an effective method for detecting early faults in wind turbine bearings. Traditional spectral segmentation techniques, such as Fast Kurtogram (FK) and its variants, reduce the optimal demodulation frequency band (ODFB) identification accuracy due to limitations such as fixed bandwidth segmentation and ease of selecting metrics. To address these limitations, this article proposes a novel spectral segmentation method with adaptive bandwidth, named MIENgram, which enables precise filtering of the ODFB from the signal spectrum. This method first uses the Multiple Signal Classification (MUSIC) algorithm to roughly estimate the power spectrum, capturing the basic trend of spectral variations. This spectral trend is then integrated with a VAL tree structure guided by the Fibonacci sequence, forming a filter bank that adaptively aligns with the spectral characteristics. This ensures the preservation of fault characteristic frequencies within the selected frequency bands. Subsequently, the envelope harmonic product spectrum was incorporated into the calculation of the maximum weighted spectral energy frequency factor with envelope negative entropy (ENSEF). This improvement enhances the accuracy of ODFB selection by making the metrics more focused on the main harmonic components in the band and enhancing the adaptive capability of the metrics. Finally, the proposed MIENgram method is validated through simulations, experimental data, and real-world vibration signals from the main shaft bearings of wind turbines. Comparative evaluations against FK, Autogram, and other baseline methods confirm the superior effectiveness and diagnostic accuracy of MIENgram in the early fault detection of wind turbine bearings.