The blade is a key component of large wind turbines, responsible for converting wind energy into electrical energy. In recent years, blade health monitoring has gained increasing attention. This paper presents a Mel spectrogram-based method for anomaly detection of wind turbine blades, proposing an aeroacoustic tonal anomaly detection algorithm based on Mel spectrograms. In the algorithm, a period calculation method based on pixel intensity integration is applied to identify the rotation period of the blades. Furthermore, Mel Spectrogram Deviation (MSD) is introduced to effectively extract anomalous components from the Mel spectrograms of wind turbine blades. Through contour feature recognition algorithm based on Mel spectrogram frequency bands, the algorithm is capable of identifying and extracting the frequency bands and shapes of different anomalous components, and constructing a Contour and Band Width (CBW) feature. Based on this feature, the monitoring of aeroacoustic tonal anomalous characteristics occurring in existing wind farms can be achieved. The proposed method has been tested on both experimental and measurement data, and the results demonstrate that it is effective for blade anomaly detection. The proposed method produced no false positives in experimental trials and consistently achieved an overall accuracy of 98.8%–99.3% in measurement-based evaluations.
Li et al. (Mon,) studied this question.