Sound-based wind turbine condition monitoring based on a periodicity enhancement spectrogram and time-varying filtering

Condition monitoring is important to ensure safe and reliable operation of wind turbines. Addressing challenges in sound-based condition monitoring, such as non-stationary characteristics, complex structures, and low signal-to-noise ratio. This study proposes an effective nonstationary sound analysis method based on time-frequency periodicity enhancement and time-varying filter. It improves the ability of extracting weak fault features from sound signals, thereby enabling more precise and cost-effective wind turbine health monitoring. Firstly, it sharpens the time-frequency features and highlights the condition information based on the cyclostationary. Then a multi-modal ridge extraction algorithm is proposed to accurately identify ridges of harmonic-like features and transient features. Finally, two diagnostic features are separated by using time-varying filters to obtain an improved timefrequency representation. The accuracy and robustness of proposed method is quantitatively evaluated and validated by numerical simulated signal. Then this method was applied to real-life condition monitoring cases in two wind farms, which demonstrated that this method can effectively enhance and extract the diagnostic features, allowing blade defects of different severity to be identified remotely. This study shows that the sound based diagnostic approach is able to extract wind turbine health status information at low signal-to-noise ratio and provides an effective way for monitoring wind turbine blade conditions.